Data-Driven Marketing: The 15 Metrics Everyone in Marketing Should Know

1. 1Define the metric as two measures taken from the same respondents: product or brand awareness (recall and recognition) and recognition of the specific touch point that carried the message. The Nivea for Men Shaver figures track both, and awareness alone cannot tell you which medium worked.
2. 2Fix the tracking design before launch. Build a pre-campaign baseline and a post-campaign read, and put intent to purchase on the same questionnaire. The Nivea pre-post figure pairs product awareness with intent to purchase, so the funnel link is visible from day one.
3. 3Run a weekly tracker during the flight. The Nivea case uses weekly claimed advertising tracking data, which lets you see the curve build or stall while you can still change spend.
4. 4Tag every touch point separately (TV, poster, direct mail, print, online impression). Ask the same recognition question for each, so each medium gets its own awareness-lift reading.
5. 5Plot each medium on a two-axis map: brand awareness against recognition of the touch point. A second map uses the perceived importance of the medium to product awareness against touch-point recognition. Medium by medium, this separates media that are seen and matter from media that are seen and ignored.
6. 6Reallocate along the maps. Move budget toward touch points high on both axes, and rework or cut those low on both. Where a medium is recognized but rated unimportant, change the creative or the message before cutting it.
7. 7Build a response mechanism into impression advertising so awareness leaves a trace in behavior. The Porsche First Mile print ad carries a URL and an SMS number, and the online version runs alongside it.
8. 8Report awareness as the first stage of the funnel, next to the downstream metrics in the next chapters (test-drive, churn, CSAT, take rate). Awareness is 'fuzzy' on its own and does not map directly to sales revenue, so show it as a leading indicator with intent to purchase attached.

The book's real case is the Philips Nivea for Men Shaver campaign. It ran TV and poster/direct-mail advertising (Figure 4.1). The team collected weekly claimed advertising tracking data (Figure 4.2) and compared pre- and post-campaign product awareness and intent to purchase (Figure 4.3). The passages give no numeric results, so none are quoted here. The method works like this. The weekly series shows whether claimed ad recall rises while the flight is live and whether it fades when spend stops. The pre-post comparison shows whether awareness moved and whether purchase intent moved with it. The optimization maps (Figures 4.4 and 4.5) then place each medium on awareness against touch-point recognition. A medium that respondents remember seeing and credit with raising product awareness earns more budget. A medium they do not recall is a candidate for a cut. For a CMO facing a budget cut (the book cites a proposed 36 percent cut), this evidence turns 'fuzzy' branding spend into measured awareness lift per medium.

• If awareness rises but intent to purchase is flat, the message is reaching people without persuading them. Fix the creative or the offer before buying more reach.
• If a medium has low touch-point recognition and low perceived importance, cut or rework it. If recognition is high but perceived importance is low, change the message first.
• If weekly claimed recall decays as soon as spend pauses, treat awareness as a rental. Plan a continuity schedule rather than one burst.
• If an impression ad has no trackable response path (URL, SMS), add one. Without it you can measure only claimed awareness, never behavior.
• If you are defending the budget to non-marketing executives, lead with the pre-post change in awareness and purchase intent per medium. Raw impressions do not make the case.

• Measuring awareness without a pre-campaign baseline. The tell: a post-campaign number that nobody can interpret as lift.
• Reading awareness as a sales result. The tell: executives ask where the revenue is, because awareness is not directly tied to sales revenue. Pair it with intent to purchase and later-funnel metrics.
• Treating all media as one pooled awareness number. The tell: you cannot say which touch point worked, so the budget cannot be reallocated.
• Tracking only at the end of the campaign. The tell: a surprise at the post-read with no weekly data to show when momentum was lost.
• Running impression advertising with no response mechanism. The tell: great recall scores and no web or SMS traffic to confirm they matter.

• ▸Campaign and product: name the specific product being tracked, for example one launch.
• ▸Baseline awareness (pre): record the share who recall or recognize the product before launch, from the same survey wording you will reuse.
• ▸Baseline purchase intent (pre): record intent to purchase on the same respondents or the same sample frame.
• ▸Touch points in flight: list each medium (TV, poster, direct mail, print, online impression) tracked separately.
• ▸Weekly claimed ad recall: log a value each week per medium to see build and decay.
• ▸Touch-point recognition per medium: record the share who recognize each medium's ad.
• ▸Perceived importance per medium: record how much respondents credit each medium with raising product awareness.
• ▸Response mechanism: note the URL, SMS number or other path attached to each impression ad.
• ▸Post awareness and intent (post) and action: record the post-campaign values, the change from baseline, and the budget decision per medium (increase, rework, cut).

1. 1Define the test-drive for your product in one sentence. The source's examples are physical and digital: a Porsche showroom-style engagement, a Ray-Ban Virtual Mirror that simulates sunglasses online (an 'online simulated test-drive'), and an experiential Windows 7 print ad. Pick the one action that lets a customer evaluate the product before buying. Count only that action as the intent signal.
2. 2Place the test-drive inside evaluation marketing. The source defines evaluation marketing as marketing that drives purchase intent by letting customers compare products. Examples are white papers, feature-and-benefit print ads, brochures and product web sites. Put the test-drive offer where comparison happens, not in awareness media.
3. 3Give every test-drive entry point its own trackable address. The source advocates a URL or text-message number on all TV, print and billboard advertising, and says slightly changing the URL or number per medium shows how many people acted because of each one. Use one variant per medium, creative and list.
4. 4Capture the response even when the form is not completed. In the source's custom-URL postcard (www.companyname.com/Mark.Jeffery), anyone who types the URL can be tracked and followed up by phone as a lead without filling in the form. Log every visit to the personalized or test-drive URL as a lead event.
5. 5Compute test-drive propensity as test-drivers divided by people exposed, per medium and segment. This ratio is a working definition, not a formula from the source. Keep exposure counts separate from test-drive counts so the ratio stays interpretable.
6. 6Isolate causality with a small experiment. The source's fix for cause and effect is to run experiments and use the right metric for each activity. Mirror the Microsoft Media Center Edition test: run an experiential (try-it) execution against a traditional execution, same medium, same period, and compare test-drive rates.
7. 7Pair the test-drive rate with pre/post intent-to-purchase survey data where the sale is delayed. The Philips Nivea for Men Shaver campaign tracked product awareness and intent to purchase before and after. Use this because financial ROI does not work for activity separated in time from purchase, and intent is the leading indicator.
8. 8Close the loop to sales. Follow test-drive leads (phone follow-up, order of the custom poster in the Porsche Turbo Cabriolet launch) and track conversion, which is the take rate (Metric #5). Report test-drive propensity as a leading indicator and take rate as the confirmation.

The source's Porsche Turbo Cabriolet launch and the personalized golf postcard show the mechanism. The Porsche mailer delivered a stamped 'raw' metal plate and invited recipients to visit a web site and order a customized poster with 'their' new car color. That ordering step is an evaluative, try-it behavior, not a passive impression. The golf postcard worked the same way. It was addressed to 'Mark', and the back carried a custom URL (www.companyname.com/Mark.Jeffery). Anyone typing that URL could be tracked and phoned as a lead even without completing the web form. Treating each visit as a test-drive event gives you the numerator. Mailers sent gives the denominator. To learn which creative drives intent, split the list. Send half the experiential version and half a traditional version, each with its own URL, as in the experiential versus traditional Microsoft Media Center Edition print experiment. Then compare visit rates and, later, take rates. The source does not publish the experiment's numbers in these passages, so supply your own.

• If the sale happens long after the marketing touch (branding, awareness), then judge it on leading intent metrics such as test-drive or pre/post intent, not financial ROI. The source says a CFO demanding branding ROI is 'asking for the impossible'.
• If a medium has no unique URL or text number, then you cannot attribute test-drives to it. Add one before spending more.
• If the customer's main criterion is price (the Dell model), then make the evaluation quick and put the price and facts up front. If the draw is design and innovation (the Apple model), then lead the test-drive with the experience and de-emphasize price.
• If two executions differ in several ways, then do not credit the winner to one cause. Run a small experiment that changes one element so the impact is isolated.
• If test-drives rise but take rate does not, then fix the post-trial follow-up or the offer, not the top-of-funnel media.

• Reporting impressions or exposure value as intent. The source's example is DuPont's $85 million of TV exposure from Jeff Gordon's sponsorship, which its own CMO said does not map to marketing value. The tell is large exposure figures with no behavioral action count beside them.
• Counting only completed forms as test-drives. This discards the tracked URL visitors who never submitted. The tell is a lead count far below the web-visit count, with no phone follow-up.
• Using one URL or number for all media. Results cannot be attributed. The tell is a single total with no breakdown by medium or creative.
• Comparing campaigns without an experiment. Differences in audience or timing get mistaken for creative effect. The tell is that the test and comparison groups differ in more than the treatment.
• Treating test-drive propensity as the sale. It is a leading indicator. The tell is that no take-rate or conversion follow-through exists to validate it.

• ▸Product and test-drive definition: state the one evaluative action that counts (e.g., simulation used, trial downloaded, order of customized item).
• ▸Medium / creative / list: name the touch point being measured.
• ▸Unique tracking address: the URL variant or text number assigned to this medium only.
• ▸Exposed audience (denominator): number of people reached by this touch point.
• ▸Test-drive events (numerator): tracked visits or trial actions, including those who did not complete the form.
• ▸Test-drive propensity: events divided by exposed audience (working definition).
• ▸Control or alternative execution: the traditional version run in parallel, with its propensity.
• ▸Pre/post intent-to-purchase score: survey intent before and after exposure, if the sale is delayed.
• ▸Downstream take rate: share of test-drive leads that purchase after follow-up.

1. 1Write the single question exactly as the source defines it: 'Would you recommend this product or service to a friend or colleague?' Do not substitute a generic 'how satisfied are you' item; the metric is defined by this question.
2. 2Put the question on a 10-point scale and set the cutoff before you collect data: only a 9 or 10 ('definitely recommend') counts as highly satisfied and loyal. Report the share of respondents at 9–10 as your CSAT. Treat 7–8 and below as not highly satisfied.
3. 3Survey existing customers, not prospects. CSAT is tied to loyalty, and loyalty feeds awareness in the purchasing cycle, so the people who have experienced the product define the signal.
4. 4Run the survey in Excel to start. The source says Excel is well suited to customer satisfaction survey data. Keep one master file per survey wave rather than duplicate copies on each marketer's desktop.
5. 5Pair CSAT with a purchase-intent question in the same survey. The automaker example found a one-to-one correspondence between CSAT and repurchase intention, so capture both and plot them together to validate the link in your own base.
6. 6Segment the results by customer experience, especially customers who had a problem and how it was handled. The automaker finding shows service recovery can raise satisfaction and repurchase intent above the level of customers who never had a problem.
7. 7Track the 9–10 share over time and by segment as a leading indicator. Read a drop as an early warning on future sales and loyalty before revenue or churn shows it.
8. 8If you need Net Promoter Score, treat it as a derivative. Keep the plain 9–10 share as your core CSAT measure so results stay comparable.

A major automaker measured CSAT and brand purchase intent together and found a one-to-one correspondence between CSAT and repurchase intention. The surprise was in the segments: consumers who had problems with their car were more satisfied, and had higher repurchase intent, than consumers who had no problem. The explanation was excellent customer service when a problem occurred, which measurably changed perception of the brand positively. The lesson is operational: the automaker could connect a survey score to future buying behavior, and could see that the service-recovery experience, not just product quality, moved the number. A related DSW figure in the source shows the link between CSAT and future purchase intent for highly satisfied customers, the same relationship in a retail setting. To replicate, ask the recommend question and a repurchase-intent question, split respondents into 'had a problem' and 'no problem,' and compare the 9–10 shares and intent scores across the two groups.

• If a respondent scores 9 or 10, then count them as highly satisfied and loyal; if they score 8 or lower, then do not count them, however pleasant the comment.
• If CSAT and repurchase intent move together in your data, then use CSAT as the leading indicator of future sales and report it ahead of lagging revenue.
• If problem-experience customers score at or above no-problem customers, then credit and protect the service-recovery process instead of treating complaints only as a cost.
• If you need a simple, comparable number across waves, then use the core recommend question and the 9–10 share; use Net Promoter Score only as a derivative.
• If your survey base is large and you need to segment it across many dimensions, then move from Excel to a proper marketing database; Excel is fine for the survey analysis itself.

• Changing the question wording between waves. The tell: scores shift without any change in product or service, and trend lines break.
• Counting 7s and 8s as satisfied. The tell: CSAT looks high but repurchase intent does not follow, because only 9–10 marks loyal customers.
• Treating CSAT as the same thing as awareness. The tell: you survey non-customers or report awareness numbers under the CSAT label, when CSAT is closer to loyalty.
• Excluding customers who had problems, or assuming they are dissatisfied. The tell: no segment cut by problem experience, so you miss that good service recovery can lift repurchase intent.
• Collecting the score but never linking it to purchase intent or behavior. The tell: a dashboard number with no evidence it predicts sales, so it cannot justify spending.

• ▸Survey question: record the exact wording, 'Would you recommend this product or service to a friend or colleague?'
• ▸Scale and cutoff: 10-point scale; highly satisfied = 9 or 10.
• ▸Customer segment: define who is surveyed (existing customers; product, region or tier).
• ▸Wave date and sample size: record when fielded and how many responded.
• ▸Share scoring 9–10: count of 9–10 responses divided by total responses.
• ▸Repurchase-intent score: the paired purchase-intent result for the same respondents.
• ▸Problem-experience split: 9–10 share and intent for 'had a problem' versus 'no problem.'
• ▸Prior-wave CSAT and change: last wave's 9–10 share and the difference.
• ▸Action owner and next step: who responds to the movement and what they will do.

1. 1Establish who your customers are before you measure anything. Churn needs an identified customer base, and firms that cannot identify their customers usually do not know their churn rate. If you lack identity data, collect it with a clear value proposition (the grocery loyalty-card trade: detailed household shopping data in exchange for discounts). State the privacy contract in your policy and keep it easy to find.
2. 2Define churn for each business line: the share of existing customers who stop purchasing in a fixed period. Compute it per product, region and segment, not as one blended number. The source's cautionary case is a company that believed it had no churn problem, then found parts of the business at 45 percent.
3. 3Cut churn by value tier. Retention of high-value customers moves profitability most, so attach CLTV (metric #10) to each segment and rank segments by value at risk, not by raw defection count.
4. 4Add CSAT as the leading indicator. Measure satisfaction and repurchase or purchase intent together and look for the link. One automaker found a one-to-one correspondence between CSAT and repurchase intention. DSW tied CSAT among highly satisfied customers to future purchase intent. Use CSAT to forecast churn before it shows in sales.
5. 5Find where defection concentrates with a segmentation model. EarthLink used a decision tree whose branches show the percentage of customers who do and do not churn within 60 days, plus N per branch. Pick the branches with high churn and enough customers to matter.
6. 6Treat the targeted segment and measure it. EarthLink tracked churn for one dial-up segment 30 to 120 days after treatment. Run it as a small experiment with an untreated comparison group so you can attribute the change to the program rather than assume it.
7. 7Translate the churn change into annual sales using a churn reduction impact template (Figure 4.9). Lower average annual churn over the product life cycle raises annual sales. For long-life products such as cars, computers or washing machines, build in a time delay between the marketing activity and the repurchase.
8. 8Budget it deliberately. Customer-relationship marketing (thank-you notes, loyalty programs, concierge services) averaged about 12 percent of marketing spend in the surveyed portfolios, against 52 percent for near-term demand generation. Check whether your retention spend matches the value at risk.

The source describes a major company that did not think it had a churn problem because it did not know who its customers were. Once customer data was obtained, some parts of the business showed churn as high as 45 percent. A disciplined response follows the EarthLink pattern. First, compute churn per business unit and rank the units by customer value. Second, build a decision tree on customer attributes, with the outcome defined as churning within 60 days (EarthLink's definition). Each branch reports the share of customers who churn and the number of customers (N). Third, choose a high-churn, high-N branch and apply a retention treatment to that segment only, as EarthLink did with one segment of dial-up customers. Fourth, compare churn in the treated group against a comparison group 30 to 120 days later, the EarthLink tracking window. Fifth, enter the churn difference into the churn reduction impact template to estimate annual sales gain, allowing for the lag on long-life products. The treatment cost against that gain gives you the retention ROMI.

• If you cannot name your customers, then fix data capture first (offer a clear what's-in-it-for-me and a stated privacy contract), because any churn figure you compute without it is unreliable.
• If blended churn looks acceptable, then still cut it by segment, because averages hide pockets such as the 45 percent case.
• If a segment has high churn but low CLTV, then deprioritize it; if it has high CLTV, then fund retention first.
• If CSAT or repurchase intent drops in a segment, then act before churn appears, since CSAT is a leading indicator and churn is a lagging one.
• If the product has a long life (car, computer, washing machine), then expect a delay between retention activity and repurchase; judge the program against the product cycle, not the quarter.

• Assuming no churn problem because customer identity data is missing. The tell: nobody can state a churn rate by segment.
• Using a single blended churn rate. The tell: the company-wide number looks fine while specific units lose a large share of customers.
• Treating all defectors as equal. The tell: retention spend is spread evenly and not weighted by CLTV.
• Claiming credit for churn changes without a comparison group. The tell: you report before/after churn but have no untreated segment to compare with.
• Judging retention programs on short-term sales for long-life products. The tell: a program is cut after one quarter even though repurchase cycles are years long.

• ▸Segment definition: name the business unit or customer segment being analyzed.
• ▸Customers in segment (N): count of identified customers at start of period.
• ▸Baseline churn rate: share who stopped purchasing in the period (use your 60-day or annual definition consistently).
• ▸Segment CLTV: average customer lifetime value for the segment, to rank value at risk.
• ▸CSAT / repurchase intent: current score and its link to repurchase for this segment.
• ▸Retention treatment: the specific offer or program, and the control group left untreated.
• ▸Measurement window: when you will read churn after treatment (EarthLink used 30 to 120 days).
• ▸Treated vs. control churn: observed churn in each group; the difference is the program effect.
• ▸Annual sales impact and time lag: projected sales gain from lower churn, with expected delay for long-life products, entered against treatment cost.

1. 1Define the 'take' before the campaign launches. Write down the single action that counts as acceptance: an order, a log-in, a visit, a send-to-a-friend. Take rate is acceptors divided by contacts, and you cannot compute it from a response you never defined.
2. 2Design the campaign for end-to-end measurement, as the Porsche mailer did. Route every contact to a trackable response channel (a web site, a URL, an SMS number) so each acceptance traces back to a contact. The passages show that a measurable design is what captures response data and identifies leads.
3. 3Fix the denominator. Count contacts actually delivered to the target list, not the audience reached in general. Record the list size and the segment before sending, so the rate stays comparable across campaigns.
4. 4Choose the analysis direction that matches your constraint. If the contact count is fixed (a set list or mail drop), vary the acquired customers and read off the outcomes. If the number of acquired customers is fixed (a quota), solve for the contacts you need: required contacts = target customers ÷ take rate. These are the two cases in the Chapter 4 take-rate template (Figure 4.11).
5. 5Convert take rate to acquisition cost. Acquisition cost per customer = total campaign cost ÷ customers acquired. Put cost and take rate side by side, because a higher take rate lowers cost per customer only if cost per contact stays flat.
6. 6Track the funnel behind the headline rate. Porsche logged log-ins, session time, orders and send-to-a-friend invitations, not only the response rate. Report each step so you can see where acceptance stalls.
7. 7Run small experiments on the elements you can change. Microsoft's Security Guidance campaign got a 400 percent improvement in performance just by changing the landing page. Change one element at a time and compare against a control, as in the four-cell Microsoft ad test with 350 people per cell.
8. 8Record each result against the prior campaign and feed it back into the next list, offer and channel choice.

The Porsche 911 Turbo Cabriolet launch mailer sent a letter with a customized stamped 'raw' metal plate. It invited the recipient to choose a car color on a web site and order a personalized poster. Because the mailer was integrated with the web site, Porsche could trace response end to end: 2,700 unique log-ins, an average session of almost 15 minutes, 5,670 posters ordered, and nearly 500 send-to-a-friend invitations. The campaign reported a 30 percent response rate, and 38 percent of Turbo Cab buyers during the period received the mailer. That rate is remarkable for a $130,000 product aimed at busy executives, lawyers and doctors. Two cautions apply. The passages call the 30 percent a 'response rate', so confirm that numerator and denominator match your own take-rate definition before you compare. The passages also give no mail-drop size or campaign cost, so you cannot derive acquisition cost from them. With a quota-style target, you would divide it by the 30 percent rate to size the list. Orders running about 2.1 posters per log-in (5,670 ÷ 2,700, derived) show that downstream behavior differs from the headline rate, which is why you track each step.

• If you cannot trace a response back to a specific contact, then redesign the response path (unique URL, SMS number, integrated web page) before you spend on the campaign.
• If you have a fixed list, then use the fixed-contacts template to project acquired customers. If you have a customer quota, then use the fixed-acquired template to solve for the contacts needed.
• If the take rate is low, then test the landing page or offer first. A single landing-page change produced a 400 percent improvement at Microsoft, and it is cheaper than rebuilding the campaign.
• If a take rate is quoted without its denominator and action definition, then do not compare it to other campaigns until you have normalized both.
• If take rate rises but cost per acquired customer does not fall, then check whether cost per contact rose with it (richer creative, customized mailers) before calling the campaign efficient.

• Counting soft responses as takes (visits, impressions, log-ins) while reporting them as acceptance. The tell: the take rate looks high but acquired customers and revenue do not move.
• Using an undefined or shifting denominator (reached audience one month, mailed list the next). The tell: take rates cannot be reconciled across campaigns.
• Launching without measurement built in, then trying to reconstruct response afterward. The tell: you cannot say which contacts produced which customers.
• Reading take rate alone and ignoring cost. The tell: an expensive customized piece wins on rate but acquisition cost per customer is unchanged or worse.
• Changing several elements at once (offer, creative, landing page) and crediting the result to one. The tell: the next campaign repeats the 'winning' mix and fails to reproduce the lift.

• ▸Campaign name and date: one line identifying the offer, channel and period.
• ▸Defined 'take' (acceptance action): the single measurable action that counts as acceptance.
• ▸Contacts delivered (denominator): number of target contacts actually reached, with list or segment source.
• ▸Takes (numerator): count of acceptances traced to a contact via the tracking path.
• ▸Take rate: takes ÷ contacts delivered, as a percentage.
• ▸Mode (fixed contacts or fixed acquired customers): pick one; if fixed acquired, enter the target and compute required contacts = target ÷ expected take rate.
• ▸Total campaign cost: all creative, production, media and fulfillment spend.
• ▸Acquisition cost per customer: total cost ÷ customers acquired.
• ▸Funnel steps and counts: log-ins, time on site, orders, referrals, or your equivalents, to locate where acceptance drops.
• ▸Test variable and result: the one element changed (e.g., landing page), the control, and the lift observed.

1. 1Fix the unit of analysis first: build CLTV for one customer in a spreadsheet (the book's Figure 6.1 is a single-customer Excel template), then scale to segments. Do not start with a segment average, because averages hide the spread that CLTV exists to expose.
2. 2List the five inputs from the definition as separate rows by period: acquisition cost, margin, cost to serve, retention probability, and discount rate. If one is missing, say so on the sheet rather than folding it into another row.
3. 3Pull the data across functions before modeling. The source says CLTV requires data from multiple functions (Fig. 10.2) and has complex data relationships (Fig. 10.6), unlike simple retail sales data. Name an owner for each input, such as marketing for acquisition cost and operations or service for cost to serve.
4. 4Set retention from churn data, not from a hope. Churn is Metric #3, and the book supplies a churn reduction impact template (Fig. 4.9). Each future period's cash flow is weighted by the probability the customer is still there.
5. 5Discount every future period's net cash flow with the time-value rule from Chapter 5: money received in one year is worth 1/(1 + r) today, and later years take further (1 + r) factors. Check the result against the Excel NPV function (Fig. 5.2). Net the acquisition cost against the discounted stream.
6. 6Stress-test the result before using it. Chapter 5 gives table-function sensitivity analysis and Monte Carlo simulation. Vary retention and the discount rate across worst, expected, and best cases and note whether the segment ranking changes.
7. 7Plot CLTV across your base to get the spectrum (the book shows one for wireless customers, Fig. 6.3). Then cross it with current profitability in a CLTV-versus-current-profitability matrix (Fig. 6.6) to assign a strategy per cell.
8. 8Turn each cell into tactics with a decision tree (Fig. 6.7) and align incentives to it. Test the tactic on a pilot with a control group, because 69 percent of surveyed firms do not run controlled experiments.

RBC (Royal Bank of Canada) is the book's segmentation case. Customers split into three segments (key, growth, prime), and the book shows each segment twice: by CLTV (Fig. 6.5a) and by current profitability (Fig. 6.5b). The book discloses the data as disguised but directionally correct, and the passages give no figures. The point is that the two views rank customers differently. A customer who looks modest on today's profit can carry high lifetime value, so CLTV and current profitability are crossed in a strategy matrix (Fig. 6.6), and a decision tree (Fig. 6.7) turns each cell into tactical execution. Continental Airlines shows the incentive side: representative compensation for 100 high-value and 100 low-value customers was plotted before and after value-based management (Fig. 6.4). The discounting step works like Figure 5.1: at an illustrative 10 percent rate (my assumption, not the book's), $100K received at the end of years 1, 2, and 3 is worth about $90.9K, $82.6K, and $75.1K today, roughly $248.7K in total. Retention then shrinks each of those flows further, and acquisition cost comes off the top.

• If a segment's CLTV is high but its current profitability is low, then treat it as an investment case and fund it from the long-term view (the CLTV-versus-current-profitability matrix exists for this). Do not cut it on this year's margin.
• If a segment's current profitability is high but its CLTV is low, then harvest it cheaply and do not buy retention with heavy spend.
• If the CLTV ranking of segments flips under the worst-to-best sensitivity range, then do not reallocate budget yet. Collect better retention data or run a pilot first.
• If a CLTV input has no owner or source across functions, then label it an assumption on the sheet and put it in the sensitivity range before anyone acts on the number.
• If a value-based tactic is untested, then run it on a pilot with a control group before scaling, since most firms skip this.

• Using current profit as a proxy for lifetime value. The tell: your top customers by this year's margin are the same list as by CLTV, so the analysis is not adding information.
• Leaving out cost to serve or acquisition cost, so CLTV is just revenue or margin summed. The tell: low-value customers still look attractive and the spread across the base is narrow.
• Skipping discounting and retention weighting, so future years count at full value. The tell: CLTV grows steadily with the horizon you choose and never levels off.
• Building CLTV inside marketing alone. The tell: cost-to-serve or retention rows hold placeholder numbers because the data lives in other functions.
• Producing a CLTV spectrum without changing treatment or incentives. The tell: segments are scored but offers, service levels, and compensation are identical for everyone, unlike the Continental before/after shift.

• ▸Segment / customer ID: name the customer or segment this sheet describes.
• ▸Acquisition cost (Period 0): enter the cost to win this customer and name the data owner.
• ▸Margin per period: enter margin for each future period (year or month).
• ▸Cost to serve per period: enter service and servicing costs, sourced from operations or service data.
• ▸Retention probability per period: from churn data; the chance the customer is still active in each period.
• ▸Discount rate (r): enter the rate, then discount each period by 1/(1 + r) raised to that period, checking against the Excel NPV function.
• ▸CLTV = discounted, retention-weighted (margin − cost to serve) summed over periods, less acquisition cost.
• ▸Current-period profitability: enter this year's profit for the cross-check in the CLTV-versus-profitability matrix.
• ▸Sensitivity range and action: record CLTV under worst, expected, and best retention and discount rate, then record the matrix cell and chosen tactic.

1. 1Pick the template that matches the initiative. A demand-generation program uses the campaign ROMI template: n campaigns in the base case and m new campaigns in the upside, over N periods. A new product launch uses the product template: n segments over N years, usually a few years of useful life. Either can run in months or years, so match the period to the program (the source shows a three-year program and a nine-month campaign).
2. 2Build the base case first: the revenue and costs you expect in Year 0 if you do nothing new. ROMI is the incremental return over this baseline, so a weak baseline inflates every later number.
3. 3List every incremental cost on its own line: new product development and marketing spend, by period. Keep all dollars in one unit (the source templates use thousands) so the sheet stays readable.
4. 4Convert each marketing impact into a revenue driver you can defend with research, not a headline revenue guess. The web-portal launch uses two: an order-size revenue lift (% improvement) from market research, and a market-share increase with worst, expected and best cases from market research. Multiply each driver through the base-case revenue and subtract the incremental costs to get incremental profit per period (Metric #6, Profit).
5. 5Discount the incremental cash flows to Time 0. $1 received in one year is worth $1/(1 + r) today, so each year's flow is divided by (1 + r) raised to that year. Use the Excel NPV function for NPV (#7), then compute IRR (#8) and Payback (#9) from the same flow. Report all four together, because profit ignores timing, payback ignores everything after break-even, and NPV and IRR cover the middle.
6. 6Stress-test the numbers. Re-run the model on the worst, expected and best market-share cases and on a lower order-size lift. Record which assumption moves NPV and payback most, and put that assumption first in line for validation.
7. 7Validate the biggest driver with a small experiment before committing the full budget. The source's answer to causality is small experiments, and its agile guidance is to design for measurement and fail fast. Replace the research-based assumption with measured take-rate or lift as results arrive, and re-run the model.
8. 8Present the result as a decision, not a metric dump. State the base case, the expected NPV, IRR and payback, the worst-case outcome, and the experiment that will confirm or kill the key assumption. This is the case the source says a CMO needs when a chairman proposes cutting the marketing budget by 36 percent.

The source's ROMI example is a web-portal new product launch (Figures 5.8–5.10), but the passages give the structure rather than the numbers. The structure: base-case revenue and costs in Year 0; new product development and marketing costs; an order-size revenue lift from market research; and market-share increase assumptions for worst, expected and best cases. The incremental profit that results is discounted into NPV, IRR and payback. The arithmetic below is illustrative only, using the source's $100K-per-year discounting pattern. Assume the launch yields $100K of incremental profit at the end of each of three years and r = 10% (an assumption, not a source figure). The present values are 100/1.10 = 90.9, 100/1.21 = 82.6 and 100/1.331 = 75.1, so NPV of the inflows is about $248.7K. If the up-front development and marketing outlay is $200K at Time 0, NPV is about +$48.7K. Payback is just under two years on undiscounted flows, and IRR sits above 10% because NPV is positive at 10%. Re-running with the worst-case market-share assumption shrinks the annual inflow, and the point where NPV turns negative shows how much margin the case has. That break-even is the number to bring to the budget meeting.

• If the initiative is a set of campaigns, use the campaign template (base campaigns vs new upside campaigns). If it is a new product, use the product template by segment. Don't mix them, because the impact drivers differ.
• If NPV is positive in the expected case but negative in the worst case, then fund a small experiment on the driver that flips it before releasing the full budget.
• If NPV and payback disagree (a positive NPV but a payback beyond what the business will tolerate), then show both and let the decision-maker choose on the cash constraint. Don't pick the flattering metric.
• If the program is short (the source's nine-month campaign), then model in months. Annual rows hide when the cash actually comes back.
• If an input comes from market research rather than measured results, then label it as an assumption and give it worst, expected and best values. Don't carry a single point estimate.

• Reporting profit or revenue without discounting. The tell: multi-year flows are summed with no NPV, and a late-arriving payoff looks the same as an early one.
• Skipping the base case, or putting incremental and base revenue in the same line. The tell: the ROMI claims credit for sales that would have happened anyway, and nobody can state Year 0 revenue and costs.
• Using a single optimistic point estimate for the key driver (market-share gain, order-size lift). The tell: there is no worst or best case, and no one can say which assumption the result depends on.
• Running the analysis once and never replacing assumptions with measured results. The tell: the model's inputs are still the pre-launch research figures after the campaign has been live for months.
• Showing one metric in isolation. The tell: a high IRR is presented with no scale (NPV in dollars) or no payback, so a small program looks better than a large one.

• ▸Initiative type and period unit: campaign or product launch; months or years; N periods; all dollars in thousands.
• ▸Base case (Year 0): revenue and costs expected without the initiative.
• ▸Incremental costs by period: development and marketing spend, one line per item.
• ▸Impact drivers: the order-size lift % and market-share gain (worst / expected / best), with the source of each (research, experiment, or judgment).
• ▸Incremental profit by period: incremental revenue from drivers minus incremental costs.
• ▸Discount rate r: the rate used to turn each period's flow into Time 0 dollars.
• ▸Results: NPV, IRR, payback period, and total profit, expected case.
• ▸Sensitivity: NPV and payback in the worst and best cases, plus the single assumption that moves them most.
• ▸Validation experiment: a small test of the key driver, its owner, the date, and the result that would change the decision.

1. 1Define the outcome in three lines before you measure anything: sales growth, profitability, and market position relative to named competitors. The source's claim is relative: firms that master data-driven marketing have significantly better financial performance than competitors. A figure with no competitor benchmark is not evidence of that claim.
2. 2Set a leader-versus-laggard baseline. The source's research covers 252 firms and $53 billion of annual marketing spend, and it compares leaders with laggards on marketing processes, technology, and investment portfolio mix. Position your firm on the same dimensions: marketing process maturity, technology, and how the marketing portfolio is split. Record whether your staff understands the essential metrics. In the research, 55% of executives said theirs does not, and more than 80% of organizations do not use data-driven marketing.
3. 3Build the chain from activity to financial outcome with the balanced scorecard and the marketing behavioral impact model (Chapter 3). Put the five nonfinancial metrics (brand awareness, test-drive, churn, CSAT, take rate) next to the financial ones (profit, NPV, IRR, payback, ROMI, customer value). Every nonfinancial metric must point to a line in the P&L. If it does not, drop it.
4. 4Convert each major program into the four financial metrics: profit, NPV, IRR, and payback. Discount future cash to today's dollars, because $1 received in a year is worth $1/(1 + r) today. Do this for multi-year programs and for short campaigns, which the source treats on a nine-month basis.
5. 5Run ROMI as a base case versus an upside case. The base case is what happens without the new initiative. The upside adds the incremental revenue and costs of the new campaigns or the new product, by segment, over N periods. ROMI is the incremental profit that the marketing investment creates, not total sales.
6. 6Stress-test the numbers with sensitivity analysis. Use worst, expected, and best cases for the drivers, such as the market-share increase and the revenue lift. Base those assumptions on market research or on small experiments, not on opinion. Report the range, and report whether the investment still pays back in the worst case.
7. 7Establish causality with small experiments before you claim firm-level credit. The source uses an experiential versus traditional print test for Microsoft's Media Center Edition. Roll the winning treatment into the ROMI case, and keep the experiment data as the evidence behind the assumption.
8. 8Report to executives on a fixed cadence. Show your performance against competitors, ROMI by initiative, and the metrics that drive those results. Use this when budgets are challenged. It lets you defend or reallocate spend with numbers, not defend awareness as a 'fuzzy' benefit.

The source's opening case is a Fortune 500 CMO in October 2008. The chairman wants to cut the marketing budget by 36 percent, and the CMO has a 2:00 P.M. meeting the same day. The source does not report the outcome, so what follows is how to use the framework in that situation. First, list each major initiative and place its nonfinancial metrics (awareness, churn, CSAT, take rate) in the scorecard chain that ends in profit. Second, for each initiative, build a base case and an upside case, and compute incremental profit, NPV, and payback. Discount at a stated rate. As an illustration only, $100K per year at an assumed 10% is worth about $90.9K for year one, not $100K. Third, rank initiatives by ROMI and payback so that a 36% cut is made from the weakest returns, not across the board. Fourth, show worst, expected, and best cases so the chairman sees how fragile each return is. Fifth, add the research point that firms mastering data-driven marketing outperform competitors financially, and show where your firm sits on the same dimensions. The chairman then has a number-based trade-off to evaluate, not a defense of branding.

• If a metric has no traceable path to profit, sales, or market share, then remove it from the executive scorecard and keep it only as a diagnostic.
• If an initiative's ROMI is positive only in the best case, then do not scale it; run a small experiment to firm up the key assumption first.
• If a program's cash returns arrive over several years, then judge it on NPV and IRR, not on first-year profit, and check payback against how long you can wait.
• If your firm outperforms on sales but not on profit, then re-rank spend by customer value and profitability, not by volume.
• If leadership proposes an across-the-board budget cut, then present initiative-level ROMI and payback so the cuts fall on the lowest returns.

• Claiming credit for total sales growth. The tell: the ROMI has no base case, so baseline revenue is counted as marketing's return.
• Reporting performance with no competitor comparison. The tell: slides show year-over-year gains but no market share or peer benchmark, so you cannot say whether you are a leader or a laggard.
• Using a single-point forecast. The tell: one ROMI number with no worst or best case, and no sensitivity analysis when finance challenges an assumption.
• Ignoring time value of money. The tell: multi-year returns are summed undiscounted, and NPV, IRR, and payback do not appear.
• Measuring too many metrics, or ones nobody understands. The tell: scorecards with dozens of items while staff cannot explain the essential metrics, which the source reports for 55% of firms.

• ▸Initiative and period: Name the campaign or launch and the number of years or months analysed.
• ▸Competitor benchmark: Name two or three competitors and the sales growth, profit, and market share comparison you can source.
• ▸Base case: Revenue and cost without the initiative, by period.
• ▸Upside case: Incremental revenue and costs by campaign or segment, with the research or experiment that supports each assumption.
• ▸Discount rate: State r and the PV of each period's cash flow at Time 0.
• ▸Financial metrics: Profit, NPV, IRR, and payback for the initiative.
• ▸ROMI: Incremental profit divided by marketing investment, in base, worst, expected, and best cases.
• ▸Leading metrics: The nonfinancial metric (awareness, churn, CSAT, or take rate) that drives the revenue line, with its current value.
• ▸Decision and owner: Fund, scale, test, or cut, plus the person accountable and the review date.

1. 1Write the data-driven marketing strategy first, then pick data and tools. The book's order is strategy, then infrastructure (Figure 2.2 translates strategy into infrastructure for a large firm). Name the marketing decisions you want data to settle before you buy anything.
2. 2Start small and collect only the data those decisions need. Continental began with focus groups and experiments, not a full build. Early results created momentum and helped line managers win executive support.
3. 3Test before you scale. Run a small experiment that isolates as many variables as possible and compares a treated group against a control group. The source reports almost 70 percent of marketing organizations do not pilot campaigns against a control, so this step alone separates you from most peers.
4. 4Define the metrics at each phase of the marketing cycle. Link each activity to a metric, using the marketing behavioral impact model (Figure 3.1) and the cycle-phase metrics (Figure 3.2). Then roll them into a balanced scorecard for marketing (Figure 3.3). Cover nonfinancial metrics (awareness, churn, CSAT, take rate) and financial ones (profit, NPV, IRR, payback, ROMI).
5. 5Close the gap between channel metrics and revenue. Recognize that last-click attribution credits one touch with 100 percent of the sale. Where you have click-stream data, weight assist clicks by their contribution to the final sale.
6. 6Obtain the customer data you lack. Integrate fragmented databases so you can calculate true profitability and customer lifetime value. Continental needed 45 integrated databases before it could measure the true revenue effect of a campaign.
7. 7Size the infrastructure to requirement complexity, not just data volume. Decide whether to forklift data into the enterprise data warehouse or rearchitect it. Low-complexity retail sales relationships need far less than customer lifetime value relationships (Figures 10.5 and 10.6).
8. 8Change the reward system. Experiments go unused because rewards are based on activity, not results. Pair measurement with governance and a closed-loop campaign management process (Figure 11.1), upgraded in three phases.

Continental Airlines began its data-driven journey with a small initiative: a letter to targeted customers encouraging them to sign up for the Presidents Club. It used focus groups and experiments rather than waiting for full infrastructure. Early results showed a clear uptake in club sign-ups. Because the club is a high-margin, high-profitability offering, the initiative showed a very positive ROMI. That momentum helped line managers secure executive support. Later, once systems integrated 45 databases and could calculate true profitability and CLTV, Continental found that customers who received the letter had an 8 percent improvement in revenues. The early proxy result was confirmed by a fuller measure. Mike Gorman, senior director of customer management, said that once the database capability existed, tools that created enormous value could be spun off quickly.

• If someone says causes are too tangled to measure, then run a small controlled experiment with a control group instead of debating attribution across overlapping campaigns.
• If the data to compute true profit or CLTV does not exist yet, then start with a proxy such as sign-ups, and re-measure once the databases are integrated.
• If the credit for a sale goes entirely to the final click, then look at assist clicks and reweight them. The last-click keywords get roughly 50 percent in the Artemis example, not 100 percent.
• If requirement complexity is low (simple retail sales relationships), then build the lighter infrastructure. If it is high (customer lifetime value), then plan a rearchitecture.
• If experiments are not happening despite awareness of their value, then fix the reward system, because it pays for activity rather than results.

• Building infrastructure before defining the strategy. Tell: a large data investment with no list of decisions it is meant to improve.
• Running overlapping campaigns with no control group. Tell: nobody can say what worked, and the team falls back on 'too many causes to tell'.
• Crediting the last click with the full sale. Tell: branded-search keywords look like heroes while the earlier searches that started the journey get cut. In the source, one customer's six searches over three weeks were credited entirely to the last, a $1,205 holiday package.
• Waiting for perfect data before acting. Tell: no initiative launches until all databases are integrated, so no early win builds executive support.
• Rewarding activity instead of results. Tell: teams are evaluated on campaigns launched, and experiments are skipped as extra work.

• ▸Marketing decision to improve: state the one decision this pilot informs.
• ▸Hypothesis: what you expect to change and why.
• ▸Test and control groups: define who is exposed and who is held out, and how many in each.
• ▸Variables isolated: list what stays constant so the effect can be attributed.
• ▸Primary metric and phase: name the metric, the marketing-cycle phase, and whether it is financial or nonfinancial.
• ▸Proxy versus true measure: note the quick proxy now and the profit or CLTV measure to check later.
• ▸Data and systems required: list the sources and databases needed, and which are not yet integrated.
• ▸Success threshold and ROMI estimate: set the result that triggers scale-up and the expected return.
• ▸Executive sponsor and reward alignment: name who backs it and how the team is rewarded for results.

1. 1Write the business objective before the creative brief. Name the one or two key business objectives the campaign serves. The source research finds that 73 percent of organizations do not rate campaigns against key business objectives before a funding decision, so this step is where most of them fail.
2. 2Pick the few metrics that prove the objective, not every available one. The source's position is to focus on the 15 that matter rather than 50 or 100. Tie each metric to a phase of the marketing cycle (awareness, response, purchase, loyalty) so every activity has a number attached.
3. 3Build the scorecard before launch. Rate the campaign against the objectives, and put a forecast on it (ROMI, NPV, or CLTV, as the objective requires). Add a business case for funding. The source reports that 53 percent do not forecast and 57 percent do not use business cases.
4. 4Build the tracking mechanism into the campaign itself. Use personalized log-ins, custom web URLs, or a response path that ties each touch to a customer record, so response data is captured as a by-product of participation instead of reconstructed afterward.
5. 5Decide the comparison before you spend. Choose a before-and-after measure (the Best Buy approach: sales lift as a percentage change, measured before and after) or a pilot against a control group. The source finds that 69 percent run no control-group experiments.
6. 6Make hard-to-measure activities measurable by adding an activation layer. Brand-style advertising such as the Tyvek print ads is difficult to score on its own. Pair it with an activity that carries a trackable response, and keep score on all major activities.
7. 7Specify the secondary signals up front. Decide whether you will capture referral behavior (send-to-a-friend as a word-of-mouth signal), time on site, and lead capture, so you do not discover them after launch.
8. 8Document the screening and prioritization process so the same scorecard and business-case test applies to every campaign. The source finds that 61 percent have no defined, documented process.

The 2008 Porsche Turbo Cabriolet launch shows a campaign designed for measurement. Existing Turbo Cab owners received a stamped raw metal plate timed to the press announcement. The mailing included personalized log-in credentials and invited recipients to the web site to choose their car color and order a personalized poster. Because the direct mail was integrated with the web site, tracking ran end to end. The campaign recorded 2,700 unique log-ins, an average session time of almost 15 minutes, 5,670 posters ordered, and nearly 500 send-to-a-friend invitations (a WOM signal). Overall response rate was 30 percent, and 38 percent of Turbo Cab buyers in the period received the mailer. Those figures exist only because the log-in credentials and web path were part of the design. The result is notable for a $130,000 product aimed at busy executives, lawyers, and doctors, and the site also captured response data and identified potential leads.

• If the campaign cannot name the business objective it serves, then do not fund it. The scorecard needs an objective to rate against.
• If the activity is hard to measure directly (branding, awareness), then pair it with an activation or response mechanism and track the response, rather than dropping measurement.
• If you can run a pilot, then run it against a control group before scaling. If you cannot, then at minimum measure before and after, as Best Buy did with sales lift in the Jills-targeted stores.
• If the tracking path does not identify the customer (log-in, custom URL, coded response), then fix the path before launch. You cannot add it afterward.
• If a metric will not change a funding or optimization decision, then cut it from the scorecard.

• Measurement added after launch. The tell: reporting starts with 'what data do we have?' rather than 'what did we plan to track?'
• Funding decisions made without scoring against objectives. The tell: no pre-funding scorecard exists, and campaigns are justified by anecdote or creative appeal.
• No control or baseline. The tell: results are stated as raw totals with no before-and-after change or comparison group, so causality cannot be argued.
• Metric overload. The tell: dashboards list dozens of numbers and no one can say which one decided the budget.
• Ignoring unmeasurable-looking activities. The tell: brand or sponsorship spend (like the Tyvek print advertising and NASCAR sponsorship) has no associated score, so it is first to be cut when the budget is challenged.

• ▸Campaign name and vision: one sentence on what the campaign is meant to change.
• ▸Key business objective(s): the one or two objectives the scorecard will rate against.
• ▸Primary metrics (max 3-5): each metric tied to a marketing cycle phase and the objective it proves.
• ▸Forecast and business case: expected ROMI, NPV, or CLTV, with the assumptions written out.
• ▸Tracking mechanism: the specific device that captures customer response (personalized log-in, custom URL, coded reply) and where the data lands.
• ▸Comparison design: control-group pilot or before/after measure, with baseline value and measurement window.
• ▸Secondary signals to capture: for example referrals or send-to-a-friend, time on site, leads identified.
• ▸Scorecard rating and funding decision: how the campaign scores against each objective, who decided, and the date.
• ▸Post-campaign review: actual versus forecast, plus what to change in the next campaign's design.

1. 1Start with the profit question, not the response rate. Put the marketing cost (volume × unit cost) next to the incremental revenue, then apply the gross margin. Sears: 250 million catalogs at about $1 each is roughly $250M of cost against $900M of incremental revenue, which is about 25% of revenue. At a retail margin of 10% or less, that is a loss of more than $100M a year. Run this check before you design any test.
2. 2Establish what the baseline is before you attribute a lift. The Romanian sponsorship was credible because no other marketing was running there, so the sales lift could only come from the sponsorship. If you cannot say what else was running in the market, you cannot claim causality.
3. 3Where you cannot get a clean market, build a controlled test. The U.K. rollout isolated confounding effects with experimental design: test stores received the program and were compared against stores that did not. Change one program at a time, and record what else is running in each group.
4. 4Replace coarse cutoffs with finer segments. Sears had targeted the top 40% of households by recency, frequency, and monetary value, with only regional versioning. It moved to 25 segments built on variables, attributes, and purchase characteristics, using an enterprise data warehouse and analytics.
5. 5Version the creative to the segment. Sears chose products and categories relevant to each group. Use the same logic when buying a targeted list: segment first, then match the offer to the segment.
6. 6Drop the hard cutoff and test the next tier. Sears reached lower-value customers with upside potential ('upsell' opportunities) instead of excluding everyone outside the top 40%. Treat these customers as a test cell and track them separately.
7. 7Build tracking into the piece itself. A custom URL per person (www.companyname.com/Mark.Jeffery) lets you follow up a visitor as a lead even if the web form is never completed. The Porsche mailer used personalized log-ins for the same purpose. Both give end-to-end tracking.
8. 8Measure the mechanisms behind the lift, not only revenue. Sears tracked trips per mailed customer (+1%), average purchase dollars per trip (+5%), and gross margin (+2%). Then convert the incremental profit into NPV, IRR, and payback, subtracting COGS and all program costs, and keep the baseline out of the figure.
9. 9Pre-write your answer to 'How do you know the campaign caused it?' The question will be asked at the ROMI presentation. Your answer should name the control, the confounders you excluded, and the experimental design.

Sears' 2001 direct-mail program mailed more than 250 million catalogs a year to 14–18 million customers in 18 mailings, producing $900M of incremental sales. At about $1 per piece, cost was roughly $250M, about 25% of revenue. With retail margin near 10%, the program was losing over $100M a year despite the large sales number. Sears used an enterprise data warehouse to split customers into 25 segments, versioned the mailers by product and category, and dropped the strict top-40% cutoff to reach lower-value customers with upside. Revenue rose $215M a year, taking the program from $900M to about $1.1B. The component metrics were trips per mailed customer +1%, average purchase dollars per trip +5%, and gross margin +2%. The margin gain came from featuring the 'right' product, so the sale did not depend on off-price promotions. The catalog targeting project alone had an NPV above $40M.

• If a campaign shows large incremental revenue, then compute cost as a share of revenue and compare it to gross margin before calling it a success. A 25% cost against a 10% margin is a loss.
• If no other marketing ran in the market, then the lift can reasonably be attributed to the program, as with Romania. If other activity overlapped, then run a controlled test, as in the U.K. stores, before claiming causality.
• If your segmentation is high/medium/low or a single top-percentile cutoff, then refine it into more segments and version the content. Test the excluded lower-value customers instead of dropping them.
• If your customer base is small, then buy targeted lists and personalize the piece. A personalized URL or log-in gives trackable response.
• If NPV is positive, IRR is high, and payback is under two years (the sponsorship case: $917K NPV, 132% IRR, 1.4 years), then the program warrants a closer look at more funding. Rerun it after the controlled test results come in.

• Reporting gross incremental revenue as success. The tell: the report has no marketing cost as a share of revenue and no margin applied.
• Claiming causality with no control. The tell: someone asks 'how do you know the campaign caused it?' and the only answer is that sales went up after launch.
• Using only a coarse cutoff such as top 40% by recency, frequency, and monetary value. The tell: everyone in a region gets the same mailer, and lower-value customers with potential are never tested.
• Assuming a large organization's data and budget are prerequisites. The tell: you postpone targeting because you lack a data warehouse, when a purchased targeted list with personalized URLs would give you a test.
• Dropping the baseline discipline in ROMI. The tell: reported lifts mix organic sales with program sales, or two quoted lifts for the same market disagree (the passages cite 108% and 150% for Romania). Reconcile them before presenting.

• ▸Program and objective: state what the campaign must change (trips, purchase dollars, margin, leads).
• ▸Cost check: volume × unit cost = total marketing cost; divide by incremental revenue to get cost as a % of revenue.
• ▸Margin check: incremental revenue × gross margin % − marketing cost = incremental profit. Flag the program if it is negative.
• ▸Test and control design: define the test group and the matched control group (for example, test stores vs. comparison stores), and list the confounding activities you are isolating.
• ▸Segments and versions: list each segment, the variables defining it, and the product or category version it receives.
• ▸Extension cell: name the lower-value group with upside that you will mail outside the old cutoff.
• ▸Tracking mechanism: custom URL, personalized log-in, or other identifier tied to each recipient.
• ▸Lift metrics: baseline and observed values for trips, average purchase dollars per trip, gross margin, and any response counts.
• ▸ROMI summary: incremental revenue less COGS and all program costs, then NPV, IRR, and payback, with the baseline removed.
• ▸Causality answer: one or two sentences you would give to 'How do you know the campaign caused it?'

1. 1Start from the marketing strategy, not the technology. Write down the data-driven decisions you need to make (campaign selection, segmentation, event-triggered offers, asset tracking). Then work backward to the data and systems each decision requires. The book frames this as translating data-driven marketing strategy into infrastructure (Figure 2.2).
2. 2Baseline yourself against the divide. Answer the survey's five questions: Do you use a centralized campaign database (57% of firms don't)? An EDW to track customer interactions (70% don't)? An EDW plus analytics to pick campaigns (71% don't)? An integrated data source for automated event-driven marketing (80% don't)? Software such as MRM to track campaigns and assets (82% never do)? Each 'no' is a named infrastructure gap.
3. 3Name the four components and tie each to a use. EDW holds customer interactions with the firm and with campaigns. Operational CRM acts on that data. Analytics selects campaigns and segments. MRM tracks campaigns and assets. A component with no decision attached to it is spend without a purpose.
4. 4Ask 'which data do you really need?' before buying anything. Customer lifetime value (CLTV) pulls data from multiple functions (Figure 10.2), so list the source functions for each metric you plan to report. Collect the right data, not all the data. The source's complaint is that marketers have lots of data and none of it is useful.
5. 5Size the build on two axes: data size and requirement complexity (Figure 10.4). Low-complexity retail sales data relationships (Figure 10.5) call for a ranch house. Complex relationships such as those behind CLTV (Figure 10.6) push toward an Empire State build (Figure 10.3). Do not let data volume alone set the scope.
6. 6Decide whether to forklift existing data into the EDW as-is or rearchitect it. Make this choice explicitly per data domain, not once for the whole warehouse.
7. 7Close the loop. Connect the infrastructure to closed-loop marketing campaign management (Figure 11.1): data informs campaign selection, the campaign runs, results flow back into the EDW. Place your organization on the three stages of MCM capability (Figure 11.5) and build one stage ahead, not three.
8. 8Budget it as its own line. Surveyed firms (254 respondents) put about 14% of marketing spend into infrastructure and capabilities (technology and training), against 52% for demand generation, 10% for branding and awareness, and 12% for customer equity. Compare your split to that baseline before you commit.

Sears ran a direct-mail initiative that drove about $900 million in incremental revenue at a marketing cost of roughly 25% of those revenues (about $225 million). U.S. retail margins are about 10% or less (about $90 million on $900 million), so the initiative was losing more than $100 million a year, on an order-of-magnitude basis. The infrastructure answer was an EDW plus analytics. Sears split the targeted customers into 25 segments using a robust series of variables, attributes, and purchase characteristics. It versioned the mailers by the products and categories relevant to each segment. It also dropped the strict cutoff that mailed only the top 40% of customers, so it captured upsell among lower-value customers with upside. Mailer revenue rose $215 million a year, turning a $900 million initiative into a $1.1 billion one. The detailed metrics matter as much as the total: trips among mail recipients rose 1%, average purchase dollars per trip rose 5%, and gross margin rose 2% because the 'right' product sold without off-price promotions. Segmentation is an old idea; what changed was that data-warehousing technology made fine-grained segmentation practical (25 segments versus the old high/medium/low).

• If a campaign's cost can't be set against the revenue it drives, fix the data first: the infrastructure must at minimum link campaign cost to recorded customer response before you add more tools.
• If your data relationships are simple (low-complexity retail sales data), scope a ranch house. If a metric like CLTV needs data from multiple functions with complex relationships, plan for the Empire State end of the range.
• If you can already run mass campaigns but can't choose or target them with data (71% of firms), put EDW plus analytics ahead of MRM or new channels; the Sears case shows the return sitting in targeting.
• If the existing data is usable, forklift it; if its structure can't support the relationships your decisions need, rearchitect it. Decide per data domain.
• If a proposed data source or tool maps to no specific marketing decision or metric, drop it from scope.

• Building to data volume instead of requirement complexity. The tell: a big warehouse that still can't produce a customer-level metric like CLTV because the cross-functional relationships were never modeled.
• Collecting everything. The tell: lots of data, none of it useful, and nobody can say where to start (the Obstacle 1 complaint).
• Buying technology without the process around it. The tell: an EDW exists, but campaigns are still chosen by habit and there is no business case or ROMI defined before funding.
• Using coarse segmentation despite the new capability. The tell: marketers still targeting 'high, medium, low' or a rigid top-40% cutoff, as Sears did before it moved to 25 segments.
• Treating infrastructure as a one-time purchase. The tell: no closed loop, so campaign results never return to the data store and capability stays at an early MCM stage.

• ▸Decision to support: the specific marketing decision (e.g., which customers receive which mailer) this infrastructure must enable.
• ▸Divide check: yes/no on each of the five survey capabilities (central campaign database, EDW for interactions, EDW plus analytics for campaign selection, integrated data for event-driven marketing, MRM).
• ▸Metrics required: list each metric the decision needs (e.g., CLTV, take rate, ROMI).
• ▸Data sources by function: for each metric, which functions supply the data and where it lives today.
• ▸Requirement complexity: low (simple sales relationships) or high (multi-function relationships like CLTV).
• ▸Data size: rough scale of the data, recorded separately from complexity.
• ▸Build scope: ranch house or Empire State, with a one-line rationale.
• ▸Data approach: forklift or rearchitect, per data domain.
• ▸Budget and MCM stage: infrastructure % of marketing spend versus the 14% average, plus current and target stage of campaign management capability.

1. 1Diagnose which of the five Obstacle 5 symptoms you have: avoiding accountability, incentives paid for activity, no culture of measurement, missing skills, or resistance to new ideas (including 'metrics kill creativity'). Write down the evidence for each. The fix differs by symptom, and the first two are structural, not attitudinal.
2. 2Classify the organization as rational, bureaucratic, or political. Don't assume 'the best idea will prevail.' Bureaucratic firms need the order to come down the chain. Political firms need you to map the centers of power (budget authority, staff, kingdoms), because the most powerful people are often not the senior executives.
3. 3Recruit senior executive sponsors who hold power, appreciate results, and already like data-driven decisions. In a large firm, form a guiding coalition, an executive council that oversees strategy development and execution and monitors progress. In a small firm, the boss's buy-in plus demonstrated results is enough.
4. 4Keep marketers in the lead. Don't let the infrastructure build (likely a multimillion-dollar investment) become an IT-led project. The source calls that the 'kiss of death.' IT is a partner, not the owner.
5. 5Realign incentives from marketing activity to results. Replace 'campaigns launched' with outcome metrics. The Circuit City 'constant sales' strategy drove revenue while losing money, so judge by profit, not by volume of activity. Do this early. Training can't fix a pay and recognition system that rewards activity.
6. 6Fund training properly. Teach the new approaches, tools, techniques, and skills. Use group action-learning exercises, not rote sessions. Schedule follow-up coaching to reinforce, and bring in dynamic external speakers who can credibly voice your ideas. Start with the metrics people say they don't understand (55% of managers report staff don't understand NPV and CLTV).
7. 7Run top down and bottom up together. Equip frontline marketers with tools and processes, and have executives visibly lead. Create momentum with an easy win so sponsors and skeptics see results. Make results and methods transparent so others can see what worked.
8. 8Treat culture as one of four components alongside strategy, work processes, and technology. 'You build it, they will come' doesn't work with employees. They must want to deliver exceptional performance.

Royal Bank of Canada began its data-driven journey by looking at what it already did. Each year at IRA-equivalent contribution time, marketing gave the sales force an alphabetically ordered list. Each salesperson called the top 10 candidates and averaged one to two acceptances. Cathy Burrows, who led the initiative, framed the question as 'How can I do this better, cheaper, faster, and smarter?' The marketing team built a model that ranked and scored the list by propensity to contribute more than $5,000. For culture purposes, notice what the easy win does. It starts from a baseline everyone recognizes (1–2 of 10) and replaces the alphabetical list with a scored one. It changes no one's job except to hand them a better list, and it produces a result that a sponsor with power can point to. To copy this, pick one recurring activity, record its baseline yield before changing anything, change one thing, and report the delta to your executive council. The passages don't give RBC's post-model figures, so use your own measured results.

• If your organization is bureaucratic, then secure the top executive's directive and route requests through formal protocol. If it is political, then map who holds budget and staff authority and win those power centers first.
• If sponsors are enthusiastic but staff don't use metrics, then check incentives before adding training. Pay for activity produces activity regardless of skills.
• If IT proposes to own the infrastructure project, then insist marketing leads and IT supports. A technology-led initiative is the failure pattern named in the source.
• If someone says metrics will kill creativity, then show a small win that measures a creative test without constraining it, rather than arguing in principle.
• If you are in a small firm with the boss's buy-in, then skip the council. Demonstrate results quickly and let them drive adoption.

• Sponsorship in name only: an executive endorses the program but never reviews progress. The tell is that the guiding coalition doesn't meet or monitor and the initiative stalls at the first budget fight.
• Incentives left on activity: staff are praised for campaigns shipped and spend sent out. The tell is that the 'we don't measure because we don't want accountability' behavior persists despite new dashboards.
• IT-led build: the data infrastructure becomes a systems project. The tell is that marketers say 'IT builds systems, but they are not the tools we need.'
• Training treated as an event: one rote session, no follow-up. The tell is that people nod in the session and revert to old methods within weeks, and metrics like NPV and CLTV remain unfamiliar.
• Assuming a rational culture: you rely on the merits of the idea. The tell is that a clearly better approach gets ignored because it threatens a power center or bypasses protocol.

• ▸Culture type (rational / bureaucratic / political): state your call and the evidence behind it.
• ▸Obstacle 5 symptoms present: tick any of no accountability, activity incentives, no measurement culture, skills gaps, resistance, 'kills creativity' fear.
• ▸Power map: list the three people or groups with real budget or staff authority, and note which are not senior executives.
• ▸Executive sponsor(s): name who has power, values results, and likes data-driven decisions.
• ▸Guiding coalition: list members, meeting cadence, and what progress they monitor.
• ▸Current incentive basis vs. target: write what people are rewarded for today (activity) and the result metric you will replace it with.
• ▸Training plan: name the metrics or skills to teach, the action-learning format, follow-up coaching dates, and any external speaker.
• ▸Easy win: pick one recurring activity, record its baseline result, and describe the single change you will test.
• ▸Leadership check: confirm marketing, not IT, leads, and name the marketing lead.