Principles of Marketing Engineering and Analytics

1. 1Split the construct into six columns before you touch any data: search, trial, initial purchase, repeat purchase frequency, quantity purchased, and brand switching or loyalty. Each column needs its own definition and time window. A single 'engagement' number hides which behavior moved.
2. 2Source each column from data traces, not opinion. The text says transactions and online interactions, including social networks, now leave electronic records of customer behavior. Map each of the six behaviors to a specific system or log. Mark any column with no trace as a gap.
3. 3Label every measure by type, using the book's three families: should-do (objective), plan-to-do (perceptual), have-done (behavioral). Customer behavior is the have-done family. Keep stated intentions and survey answers in a separate column so they are never mistaken for observed action.
4. 4Build one row per customer with the six behavior columns plus the observable characteristics you already hold, such as demographics or, for physicians, specialty. Individual-level rows let you aggregate to a segment later. Segment-level averages cannot be disaggregated back to individuals.
5. 5Test whether a candidate segmentation separates behavior. The book advises segmenting on characteristics related to underlying needs in the category. For each segment, compare repeat frequency, quantity, and switching. If they do not differ, the segmentation does not explain behavior, however neat the profile.
6. 6When observable traits leave behavioral differences unexplained, hypothesize an unobserved driver. The book's example is a physician's orientation toward patients versus scientific research, which shapes prescribing. Then search for observable correlates of that driver, such as physician age or type of medical school. This is the latent class logic.
7. 7End each analysis with a decision: which customers to target individually, and which marketing action should change which behavior. The book warns that more data does not improve decisions unless managers extract actionable insights, so a behavior table with no action attached is unfinished.

The book's Hunch.com study shows both the value and the limit of a priori segmentation. In April 2011, 388,000 Hunch users answered a question about computer loyalty: 52 percent called themselves PC people and 25 percent Mac devotees, which leaves 23 percent in neither camp. Hunch cross-referenced these answers with other questions and found differences in taste. Mac users leaned toward designer or retro styles, red wine, and The New York Times. PC users leaned toward casual clothes, white wine, and USA Today. Apply the six-column test. The 'loyalty' here is self-identification, a stated label, not observed repeat purchase, switching, or quantity. The authors themselves note that it is not clear how the results could be used for strategic decision making. To make this construct usable, you would pull actual purchase histories for each camp. You would then check whether the 'Mac' label predicts repeat frequency and brand switching, and only target on the label if it does.

• If a measure comes from a survey of intentions or self-description, then record it as plan-to-do and never as customer behavior. Use have-done data for the behavior columns.
• If a segmentation produces vivid profiles, as the Mac/PC study does, but you cannot name the behavior or decision it changes, then treat it as descriptive only and do not act on it.
• If segments share observable traits but differ in behavior, then look for an unobserved orientation (latent class) and search for observable correlates you can target on.
• If you have a choice of segmentation basis, then choose characteristics related to underlying customer needs in the category, because they are the ones most likely to separate behavior.
• If data volume is large but no behavior column links to an action, then stop collecting and define the decision first.

• Treating self-reported brand loyalty as observed loyalty. The tell: the 'loyalty' field comes from a survey answer and no purchase record backs it.
• Collapsing the six behaviors into one score. The tell: a segment 'rises' but nobody can say whether trial, frequency, quantity, or switching drove it.
• Segmenting on convenient demographics that do not relate to needs. The tell: segment profiles are interesting, as with Mac vs. PC users, but behavior metrics are nearly identical across segments.
• Accumulating data without extracting insight. The tell: large datasets exist but no report states which customers to treat differently.
• Relying on segment averages when decisions are made at the individual level. The tell: you cannot say what any one customer's frequency or quantity is.

• ▸Segment or customer group: name the group and the characteristic it is defined on.
• ▸Need link: state the underlying category need that justifies this characteristic.
• ▸Search behavior (have-done): the observed trace and its source system, or 'no trace'.
• ▸Trial and initial purchase (have-done): count or rate within a stated time window.
• ▸Repeat purchase frequency (have-done): purchases per period per customer.
• ▸Quantity purchased (have-done): units or amount per purchase.
• ▸Switching or loyalty (have-done): observed share retained vs. moved to another brand.
• ▸Stated intention (plan-to-do, kept separate): survey or self-label, never mixed with the columns above.
• ▸Unobserved driver hypothesis and observable correlate: the latent orientation you suspect and the trait that proxies it.
• ▸Decision triggered: the one targeting or marketing action this ledger changes, and the behavior it should move.

1. 1Pick the outcome metric and write it down as one of three: unit sales volume, revenue, or market share. Add the time frame and the unit of analysis (brand or product line). Do not mix them. Brand sales equals brand share times total market demand, so the three are linked but not interchangeable.
2. 2Decide the level of demand you model. The passages call analyzing brand sales directly 'the most common approach'. Model brand share and total market demand separately when you need to know whether a change came from your brand gaining share or from the category growing.
3. 3Build the outcome as a sum of individual responses where you can. Markets are composed of individuals, so you can model individual or segment responses and aggregate them into total market response. Use the aggregate-only model when you lack individual-level data.
4. 4Choose the response shape explicitly. Ask whether every dollar of marketing effort has the same effect on sales (a linear response) or whether some spending ranges produce larger or smaller returns (an S-shaped response). Write the form, such as SALES = a + b × ADVERTISING, and name the parameters (a, b) with their marketing meaning, for example market potential or price elasticity.
5. 5Calibrate the parameters. Use statistical estimation, a judgmental process, or both, and record which one you used for each parameter. Parameters 'infuse life into an abstract model', so an uncalibrated form tells you nothing about your market.
6. 6For share outcomes, enforce the logical constraints. Each firm's share must fall between 0 and 100 percent (range restriction), and shares across brands must sum to 100 percent (sum restriction). Use an attraction model: Share = Attractiveness of Offering ÷ (Attractiveness of Offering + Attractiveness of Competitive Offerings), with attractiveness judged from the customer's perspective.
7. 7State whether the model is static or dynamic and whether it includes competition. Decide whether you need the flow of actions and responses over time or a single snapshot. Decide whether competitor actions are modeled explicitly or treated as environment.
8. 8Link the outcome to the objective. Under a short-run profit objective, Profit = Unit margin × Quantity − Costs, so a volume or share gain only counts if it survives the margin change that produced it. Then test the forecast with scenarios: list the uncontrollable factors (customer behavior, competitors, economy, regulation) and see which ones the outcome depends on.

The passages give a market-outcome forecast for hybrid (electric-combustion) vehicles. A consensus estimate, had it been realized, would have put 150,000 hybrid-drive vehicles on the road by 2004. Actual 2004 sales were approximately 83,000 units. That is a gap of 67,000 units, with the outcome at about 55% of the 150,000 figure. (The passage itself calls the 2004 result 'remarkably accurate' for a ten-year-ahead forecast, so treat the gap as a diagnostic, not a verdict.) The optimistic scenario for hybrids lists the assumptions a volume outcome would need to hold: baby boomers with more disposable income opt for hybrids, gas prices rise above $3.50 and stay there, the government offers bigger tax breaks and raises mileage standards to 35 MPG, green living becomes fashionable, and competing technologies do not catch on. Almost all of these are uncontrollable or weakly controllable. So when the outcome falls short, you check which assumption failed before you blame the response model. The scenario method then sorts your responses into core strategies (apply for patents), contingent strategies (proceed if a test market succeeds) and hedging strategies (buy land where a factory could go), so you do not bet the whole outcome on one future.

• If you need to know why sales moved, split brand sales into brand share and total market demand. If the category grew and share fell, you lost ground even though volume rose.
• If you forecast share, then use a model that enforces the range (0–100%) and sum (100%) restrictions, such as an attraction model. A model that lets shares exceed 100% across brands fails.
• If extra marketing spend may add more or less than the last dollar did, then fit an S-shaped response, not a linear one. Do not assume constant returns.
• If the outcome depends mostly on customer behavior, competitor moves, the economy or regulation, then build scenarios and assign core, contingent and hedging strategies. A single-point forecast hides that dependence.
• If you have no individual-level data, model overall response (the sum of individual responses). If you target specific customers or run direct marketing, model individual response and aggregate it.

• Reporting volume without share. The tell: sales are up and nobody can say whether the category grew or you took share.
• Using a share model that ignores competition. The tell: modeled shares across brands don't sum to 100%, or one brand's share exceeds 100% when a competitor cuts price.
• Fitting a linear model to an S-shaped response. The tell: the model recommends ever more spend, and actual returns flatten or jump at certain spending levels.
• Treating the forecast as the outcome. The tell: the plan is built on one number (such as 150,000 units) with no record of the assumptions behind it, so a miss cannot be diagnosed.
• Maximizing volume while ignoring the profit equation. The tell: quantity rises but unit margin falls, and Unit margin × Quantity − Costs does not improve.

• ▸Outcome metric: write unit sales, revenue or market share, and the brand or product line it applies to.
• ▸Time frame and level: state the period and whether you model brand sales directly or brand share and total market demand separately.
• ▸Response form: write the equation (for example SALES = a + b × ADVERTISING) and state whether it is linear or S-shaped.
• ▸Parameters and calibration: list each parameter, its marketing meaning (market potential, price elasticity), and whether you estimated it statistically, by judgment, or both.
• ▸Competition treatment: state whether competitor actions are explicit in the model or treated as environment, and give the attraction formula if you model share.
• ▸Constraint check: confirm shares stay between 0–100% and sum to 100% across brands.
• ▸Aggregation basis: state whether the outcome is the sum of individual or segment responses or an overall response model.
• ▸Profit link: write Unit margin × Quantity − Costs for the forecast outcome.
• ▸Scenario assumptions: list the uncontrollable factors the outcome depends on, and the core, contingent and hedging strategy for each.

1. 1Write the objective before touching any model. Name the components (profit, market share, sales goal), the time horizon, and how you handle uncertainty. The source says objectives must specify all three, and must settle which objective to pursue when they conflict.
2. 2Start with short-run profit, the simplest and most common tactical objective: Profit = (Unit price − Unit variable cost) × Sales volume − Relevant costs, which is Unit margin × Quantity − Costs. Compute it for the single marketing element under study, holding the environment static.
3. 3Build a response model that shows how sales volume changes with the marketing action. For price, margin rises as price rises and volume generally falls. For advertising, margin is fixed, quantity rises, and costs rise too. Profit moves only through the net of these effects.
4. 4Split relevant costs into discretionary and fixed. Always include the discretionary costs of the activity under study. Allocate fixed costs (plant, overhead) to the marketing activity appropriately rather than ignoring them.
5. 5Hold marketing spend to the same standard as any other firm investment. Express each program as an ROI or profit contribution, because senior management now demands this justification.
6. 6Run what-if comparisons across alternatives, including inaction. The source says Marketing Engineering frames the opportunity costs of alternative actions or inaction, so state what each option forgoes.
7. 7Check the analytics investment against the performance you expect from it. CMOs plan to raise analytics spend from 4.6% to 21.9% of budget, yet rate its contribution to company performance at only 3.7 of 7. Attach a performance target to each analytics dollar.
8. 8Convert outputs into action guidelines or reports and integrate them into company processes. Insights that do not reach a decision do not move firm performance.

The source offers a disconnect that shows the problem. In the CMO survey (Moorman 2017), CMOs project the share of budget spent on analytics rising from 4.6% in 2017 to 21.9% in 2020. Asked to what degree analytics contributes to company performance (1 = not at all, 7 = very highly), they average 3.7, down slightly from 3.8 in February 2016. Spending is planned to rise steeply while expected performance is lukewarm. Apply the profit equation to close the gap. Take one program, such as advertising: margin is fixed, so you must show that the added quantity times unit margin exceeds the added advertising cost plus allocated fixed cost. If it does not, the spend does not raise profit, whatever the analytics said. Running this test on each program turns a budget line into a measurable contribution to the top and bottom lines. The source gives no figures for a specific program, so supply your own inputs.

• If the objective is a tactical one-period decision, then maximize short-run profit using unit margin × quantity − costs.
• If the lever is price, then expect margin up and volume down, and test the net. If the lever is advertising, then margin is fixed and you must verify that the volume gain outweighs the added cost.
• If a cost is tied to the marketing activity under study, then always count it as discretionary. If it is plant or overhead, then allocate it appropriately.
• If you cannot state an ROI or profit effect for a marketing expenditure, then treat it as unjustified by the standard senior management now applies.
• If objectives (profit, share, sales) conflict or the future is uncertain, then fix the time horizon and choose which objective to pursue before modeling.

• Counting revenue or volume instead of margin. The tell: sales rise after a price cut or ad push, but profit is never calculated.
• Leaving out discretionary or allocated fixed costs. The tell: the program shows a profit that disappears when finance adds overhead.
• Spending more on analytics without a performance link. The tell: budget share climbs but managers rate the contribution near the 3.7 midpoint.
• Leaving objectives vague on horizon and uncertainty. The tell: teams argue over results because they never agreed whether they were chasing profit, share, or sales.
• Assuming more data yields better decisions. The tell: large databases and reports but no actionable insight changing a decision.

• ▸Objective and time horizon: state the objective (profit, share, sales) and the period.
• ▸Marketing lever: price, advertising, or other single instrument under study.
• ▸Unit price and unit variable cost: enter both, then compute unit margin.
• ▸Expected sales volume (response model): state volume at the proposed level and the assumption behind it.
• ▸Discretionary costs: list costs tied directly to this activity.
• ▸Allocated fixed costs: list plant and overhead share assigned to the activity.
• ▸Profit = unit margin × quantity − costs: compute the result.
• ▸Alternative and inaction comparison: compute profit for at least one alternative and for doing nothing.
• ▸ROI and decision: state ROI versus the firm's investment standard, then approve, revise, or reject.

1. 1List every marketing action you actually control and sort each into product design, price, advertising or selling effort (communications), or distribution. The passages name 'product design, price, advertising, selling effort, etc.' as the inputs (arrow 1 in Exhibit 1.8). If a lever is not controllable, it is not part of the mix.
2. 2Separate the mix from the noncontrollable variables: market size, the competitive environment, and environmental conditions such as regulation or the economy. Write both lists side by side. Competitor actions (arrow 2) and environmental conditions (arrow 3) hit the same market response as your inputs, so a result cannot be credited to your mix until they are accounted for.
3. 3Name the measurable outputs each lever is supposed to move: awareness level, preference level, sales level, profit. Match each lever to one output before you spend on it, for example advertising to awareness and a promotion to sales.
4. 4Set the objective measure for each lever before acting, for example sales in response to a promotion, or percentage of the target audience that recalls an ad. This is arrow 5, the evaluation step. Without it you cannot tell whether the mix is working.
5. 5Build a response model that links inputs to outputs, even a simple spreadsheet equation. A 'dumb' spreadsheet only records numbers. A 'smart' one embeds an equation showing how advertising affects sales and revenues, so you can judge whether more or less advertising is justified.
6. 6Run what-if comparisons across alternative mix settings, including the options you will reject. Estimate the forgone profit of each rejected alternative. You only observe the consequences of actions you took, so the model is your only view of the opportunity cost.
7. 7Choose the method for each lever from the structure the source uses. For price, use cost-, demand-, or competition-oriented approaches. For product design, use conjoint analysis. For communications, use advertising, sales force, and promotion decisions.
8. 8Close the loop. Compare observed outputs with objectives, then adapt the actions (arrow 6). Add human judgment to the model's recommendation before committing, because the source states that models require judgment.

The source gives a pricing case. A manager must choose between two pricing policies and picks the lower-priced one. Price is the controllable input. Sales and profit are the outputs. Market size and competitor response are noncontrollable and sit alongside the input in the response model. Because the manager observes only the lower-price outcome, the higher-priced option stays invisible. A response model lets the manager estimate the forgone profitability of the higher-priced option, so the choice is made against a quantified opportunity cost. After launch, observed sales and profit are compared with the objective (arrow 5), and price is adjusted if the result falls short (arrow 6). The source's hybrid-car scenario shows the other side of the ledger. Gas prices, tax breaks, and media attention are noncontrollable factors. They can swing demand regardless of how well the mix is set, so they belong in the scenarios you test the mix against.

• If a variable is not under your control (market size, competitor moves, regulation, gas prices), then keep it out of the mix list and model it as an environmental or competitive condition or as a scenario.
• If you are choosing between two mix settings (for example, two price policies), then estimate the forgone result of the rejected option before you decide.
• If a lever has no stated objective measure (such as sales response or ad recall), then do not fund it until you define one.
• If a model's recommendation conflicts with informed judgment, then investigate the assumptions rather than defer to either. Models need judgment, and poor response models can lead the decision astray.
• If results miss the objective, then adjust the actions (arrow 6) after checking whether competitor or environmental changes explain the miss.

• Treating uncontrollable factors as part of the mix. The tell is a 'mix' that lists competitor actions or market conditions as things you will set.
• Judging the mix only by what you executed. The tell is the absence of any estimate of forgone profit for rejected alternatives.
• Using a dumb spreadsheet that records spend and sales without an input-output equation. The tell is that nobody can answer whether more or less advertising is justified.
• Setting levers with no objective measure. The tell is that evaluation after the fact relies on impressions instead of sales response or ad recall.
• Using a poor response model without questioning it, like a golfer who does not stand square to the ball. The tell is that results slice far from the target while the model's inputs go unexamined.

• ▸Product action: the design decision you control and the output it should move.
• ▸Price action: the policy chosen, the alternative rejected, and the method used (cost-, demand-, or competition-oriented).
• ▸Place/distribution action: the channel decision and the output it should move.
• ▸Promotion action: advertising, selling effort, or sales promotion, and the output it should move.
• ▸Noncontrollable variables: market size, competitor actions, and environmental conditions to track.
• ▸Expected outputs: awareness level, preference level, sales level, profit.
• ▸Objective measure per lever: for example, sales in response to a promotion or percentage of the target audience that recalls an ad.
• ▸Forgone-option estimate: the modeled profit of the best rejected alternative.
• ▸Adaptation trigger: the gap between observed output and objective that causes you to change the action.

1. 1Define the competitive set by the customer's alternatives, not by your category. The German Railroad's price structure lost to automobile usage, so the car was the competitor that mattered. List every alternative a customer weighs, including substitutes outside your industry.
2. 2Log each competitor's actions under the four headings in the definition: pricing, promotion, new product introductions, and distribution. Record the date, the action, and the segment it targets. Keep the log tied to the decision you face, because the source warns that an abundance of data can be a bigger problem than a lack of data.
3. 3Compare your offer with the competitor's on the attributes customers actually trade off. The railroad did this with a large-scale conjoint analysis, and the result showed that a flat per-kilometer price was the weak point against driving.
4. 4Build or adapt a market response model that includes competition, meaning how your share or sales respond to your actions and to theirs. The book treats market share and competition as a distinct part of response modeling. Use it to translate a rival's move into a predicted effect on you.
5. 5Write down at least three candidate responses, including holding course. Use the what-if capability to estimate profit for each. Do not stop at the response that looks most natural.
6. 6Quantify the opportunity cost of every response you reject, such as the forgone profitability of the higher-priced option if you choose the lower price. Managers observe only the consequences of actions they actually take, so the counterfactual exists only if you calculate it now.
7. 7Apply human judgment to the model output before you commit. The source states that marketing models require judgment, and that managers use models to test and sharpen their own mental models, not to replace them.
8. 8Implement, then compare actual results with your recorded what-if forecast. Embed the model in your information and decision systems so the analysis recurs. Earlier engineering attempts produced short-lived successes because the success was never embedded in the organization.

The German Railroad Corp., a $15 billion company, priced travel between any two points as a simple multiple of the distance. That structure was not competitive with automobile usage. The competitor was a substitute with its own cost logic, not another railway. Instead of cutting every fare, the company ran a large-scale conjoint analysis to learn what customers valued. It then launched the BahnCard, which gives cardholders large discounts off the standard per-kilometer fares. Many more passengers began to see the train as an attractive alternative to driving. With 3.5 million cardholders, the BahnCard raised profits by more than $200 million per year (Dolan and Simon 1996). The lesson for competitive actions is that the response was a restructured price offer, chosen by measuring customer trade-offs against the alternative. It was not a reflex discount.

• If customers compare you with a substitute from another industry, then add that substitute to the competitor set and test the response with conjoint analysis before you change price, as the railroad did against the automobile.
• If you must choose between a lower-priced and a higher-priced response to a rival, then use what-if analysis to calculate the forgone profit of the option you reject before you commit.
• If a rival's move produces a flood of new data, then use only the data tied to the decision at hand, because abundance is a bigger problem than scarcity.
• If the model's recommendation conflicts with your read of the market, then reconcile the two and decide with judgment, because the source holds that models require judgment.
• If you plan to answer competitors repeatedly, then embed the model in your decision systems, because one-off analyses produced only short-lived successes.

• Defining competitors only as direct category rivals. The tell is that your share erodes to a substitute, as rail did to the automobile, and your competitor list never named it.
• Reacting to a competitor's action without a counterfactual. The tell is that nobody can say what profit the rejected option would have earned.
• Treating model output as the answer. The tell is that no one has questioned the assumptions, and the decision cannot be explained apart from the model.
• Collecting every available competitor data point. The tell is large dashboards and slow decisions, with no data tied to a specific choice.
• Running the analysis once and shelving it. The tell is that the next competitive move triggers a fresh scramble instead of an update to an existing model.

• ▸Competitor or substitute: name the firm or alternative the customer is weighing against you, including non-industry substitutes.
• ▸Observed action and type: record the move and classify it as pricing, promotion, new product, or distribution, with the date.
• ▸Segment affected: state which customer segment the action targets.
• ▸Customer-valued attributes: list the attributes where the offers differ, and note whether conjoint or other evidence supports them.
• ▸Predicted effect on us: give the response model's estimate of the impact on your sales or share, with the key assumption.
• ▸Candidate responses (3+): list each option, including holding course.
• ▸Profit by option and forgone profit: record what-if profit for each option and the forgone profit of every rejected option.
• ▸Judgment override and rationale: note any departure from the model and the reason.
• ▸Actual result vs. forecast: after implementation, record what happened against the forecast and what to adjust in the model.

1. 1Name the environmental variables before you model anything. From the passages, the candidates are market size, population size and growth, prior sales levels, economic and financial conditions, political and regulatory change, technological change, and the competitive environment. Treat each as noncontrollable: it enters the model as an input you track, not a lever you pull.
2. 2Place them in the decision loop. Marketing actions (arrow 1), competitor actions (arrow 2), and environmental conditions (arrow 3) combine to drive market response, which produces outputs (arrow 4). You evaluate those outputs against objectives (arrow 5) and adapt your actions (arrow 6). A model that omits arrow 3 attributes environment-driven movement to your own marketing.
3. 3Split environmental effects using Brand Sales = Market Sales x Brand Market Share. Environmental variables (population size, growth, prior sales levels) mainly move the Market Sales term, through forecasting or demand build-up models. Your marketing mix and competitors' actions mainly move the share term. Estimate the two terms separately.
4. 4Check the share term for logical consistency. Each firm's share must stay between 0 and 100 percent (range restriction), and shares summed over brands must equal 100 percent (sum restriction). An attraction model, where share = your attractiveness / (your attractiveness + competitors' attractiveness), satisfies both.
5. 5Rate environmental risk for each candidate segment. Use the segment-attractiveness criteria: size and growth (e.g., growth forecasts of adopting new technologies), and environmental risk (economic, political, and technological change). Weigh size against your own resources, because large markets attract aggressive competitors, tougher competition, and lower prices and profitability than niche markets.
6. 6Build scenarios for the uncontrollable factors you cannot forecast. A scenario is a plausible, coherent story, not a forecast. Start by identifying the stakeholders who shape the future of the industry (customers, competitors, regulators, distributors, strategic partners), then write each scenario around the economic, regulatory, and customer-behavior factors you cannot control.
7. 7Recalibrate when the environment moves. Add the new data to your database, refit the response model for that market, and re-derive the optimal action from the refit, instead of applying a rule of thumb.
8. 8In fast-changing markets, prefer current-looking data. Behavioral measures depend on historical data that may not extend to the future, so supplement them with objective value-in-use measures or perceived-value measures anchored to a described future scenario.

The source's hybrid-car scenario shows how to turn environmental conditions into an explicit story. The optimistic scenario strings together several noncontrollable factors. The population ages, and baby boomers with more disposable income buy hybrids. Gas prices rise above $3.50 and stay there. The government offers bigger tax breaks, higher CO2 emissions fines, and a 35 MPG mileage standard. Media attention makes green living fashionable, and competing technologies do not catch on. Each element is a demographic, economic, regulatory, social, or technological variable that the carmaker cannot control. The scenario is not a forecast. It gives you a coherent future against which to test marketing actions. The passages also report a forecast made about 10 years earlier: 150,000 hybrid-drive vehicles by 2004, against actual 2004 sales of approximately 83,000 units. That gap shows why you keep the environmental assumptions visible and revisit them.

• If sales miss forecast, then first ask whether market sales or market share moved (Brand Sales = Market Sales x Share) before you change promotion spending.
• If a marketing action produces zero incremental sales, then test two explanations: no effect at all, or a competitive response that grew the product class but cut your share.
• If a market is dynamic and its environment is shifting, then rely less on behavioral (historical) value measures and more on objective or scenario-anchored perceived-value measures.
• If a segment is large and fast-growing, then check whether it fits your resources and whether the size attracts aggressive competitors that erode price and profit, before you commit.
• If a decision depends on uncontrollable factors you cannot forecast with confidence, then write several scenarios and use them to stress the decision.

• Treating a sales decline as a promotion problem. The tell: the proposed fix (e.g., 'increase promotional spending 10 percent') assumes a short-term response of at least the shortfall, with no check on market size, competitors, or environment.
• Leaving environmental variables out of the response model. The tell: the model fits history but misses when population, economy, or technology shifts, and the error gets credited to marketing.
• Using only historical purchase data in a dynamic market. The tell: estimates embed constraints from past conditions (what was available, how urgent the need was) that may not exist in the future.
• Mistaking a scenario for a forecast. The tell: one favorable story, such as the hybrid scenario, is used as the plan, with no alternative futures considered.
• Assuming bigger and faster-growing segments are better. The tell: you enter a large market and meet aggressive competitors, lower prices, and lower profitability than a niche would offer.

• ▸Decision at hand: state the marketing decision (allocation, targeting, positioning) this register supports.
• ▸Environmental variables: list the noncontrollable factors in play (market size, population growth, economic conditions, regulation, technological change, competitive environment).
• ▸Effect channel: for each variable, mark whether it moves Market Sales, Brand Market Share, or both.
• ▸Current value and trend: record the latest data and direction for each variable, with the source.
• ▸Environmental risk rating: rate each variable's economic, political, or technological risk to the segment (low/medium/high).
• ▸Scenario story: write one plausible, coherent scenario in a few sentences, naming the stakeholders involved.
• ▸Alternative scenario: write at least one contrasting scenario in the same format.
• ▸Model recalibration trigger: define the data change that forces you to update the response model.
• ▸Action and evaluation: record the action chosen, the objective it is measured against, and the review date.

1. 1Define the need context first. Name one specific need and the product setting (thirst at an outdoor event, not 'beverages'). The source shows value shifts by setting: water is valued outdoors, beer at a bar.
2. 2Place the need on a scale from general to specific. Thirst is general; thirst to reduce body temperature or replenish lost fluids is specific. Write the specific version, because it determines which offerings satisfy it.
3. 3Describe the need on its physical and social dimensions. Physical: sweetness, temperature, alcoholic content, smoothness, carbonation, flavor. Social: family setting, restaurant, bar. Record which dimensions differ between groups.
4. 4Score the temporal profile: urgency (time within which the customer believes the need must be resolved), frequency (how often it recurs), and duration (how long the tension lasts). Record each for the group, e.g. several times a day versus every five years.
5. 5Score the information profile: newness, complexity, and clarity of the need. New or complex needs mean more learning, longer buying processes, consultation with others, and more alternatives considered. Unclear needs mean heavy information search and trial.
6. 6Separate the functional need (what is required to do the job) from the perceived need (what the customer believes is required). Write both for each group. The gap tells you which measurement approach to use and what to communicate.
7. 7Note the availability or scarcity of options to satisfy the need. The source names it, together with the characteristics of the need, as a driver of where value falls.
8. 8Translate each group's profile into a response hypothesis for the 4Ps. High urgency means value in immediate satisfaction. High information needs mean value in offerings that supply communication, even if they are not objectively superior. Group customers whose profiles match and treat them as one segment.

The source contrasts two buyers of the same kind of need. A consumer's thirst recurs several times a day, lasts minutes, and is satisfied from a familiar set of drinks. Physical dimensions (temperature, carbonation, sweetness) and setting (outdoors versus bar) determine value: water is valued outdoors, beer at a bar. An organization's need for a new telephone switching system occurs about every five years and lasts several months until the upgrade is ready. It becomes urgent once the organization reaches calling capacity, because it depends on reliable responses to customer calls. Its complexity means longer buying processes, consultation with others, and more alternatives considered. The two profiles call for different marketing stimuli. The thirst buyer responds to availability and fit to the setting. The switching-system buyer responds to communication that reduces information burden and to speed once capacity is hit. Segment by these characteristics, not by product category alone.

• If a need is urgent and important, then weight your offer toward immediate satisfaction, because such customers value offerings that satisfy the need immediately.
• If the need is new, complex, or unclear, then plan for a longer buying process, consultation with others, and heavy information search, and supply communication or trial.
• If members of a group have unclear needs, then expect them to be more susceptible to marketing information such as salesperson advice, so make your information accurate and structured.
• If the functional need and the perceived need differ within a group, then build messaging and measurement around the perceived need, and decide separately whether to correct the gap.
• If the same need shows different physical or social dimensions across settings, then treat each setting as a distinct segment context instead of averaging.

• Defining a segment by product category or demographics alone. Tell: you cannot say what need, urgency, or setting separates the group from its neighbor.
• Treating functional and perceived needs as identical. Tell: the offer is objectively strong but customers do not respond, because it addresses what is required rather than what they believe is required.
• Ignoring temporal dimensions. Tell: one message and timing for buyers whose need recurs daily and buyers whose need recurs every five years.
• Under-serving information needs. Tell: a technically superior offer loses to a competitor that communicates better, since customers with high information needs favor offerings that supply it.
• Describing the need too generally (thirst, not a specific thirst in a specific setting). Tell: the offer set is undifferentiated and value judgments cannot be predicted.

• ▸Segment name and need context: one line naming the group and the specific product setting.
• ▸Need specificity: write the need from general to specific (e.g., thirst, then thirst to replenish lost fluids).
• ▸Physical and social dimensions: list the physical attributes and social settings that characterize the need for this group.
• ▸Temporal profile: record urgency, frequency, and duration in concrete terms.
• ▸Information profile: rate newness, complexity, and clarity of the need (high/medium/low) and note expected search and consultation.
• ▸Functional vs. perceived need: write what the job requires and what the customer believes it requires.
• ▸Option availability: list the options available to satisfy the need and how scarce they are.
• ▸Expected response to 4Ps: state the stimulus (speed, information, setting fit) this group should respond to and why.
• ▸Comparison with adjacent segment: name the single characteristic that most separates this group from the nearest other.

1. 1Split the offering into its three components: physical, service, and perceptual. Write the attributes of each. The passages state that every product carries a perceptual component (the BMW buyer gets the car, the warranty, and the 'Ultimate Driving Machine' perception). Customers form beliefs about all three.
2. 2Name the need the offering serves, at the right level of specificity. Move from general (thirst) to specific (thirst to reduce body temperature or replenish lost fluids). Then list the physical dimensions (sweetness, temperature, carbonation) and social dimensions (family, restaurant, bar) of that need. Preferences are judged against these.
3. 3Rank the need by importance, using Maslow's hierarchy (physiological, security, social, status, self-actualization). Ask whether communication can tie the offering to a higher-order need. NEC positioned its mobile phone as a safety device, especially for women. Soap makers stress beauty over cleanliness.
4. 4Define the perception and preference outputs you will track. The response-model framing names customer awareness level, product perceptions, and preference level as measurable outputs. Set the firm's objective for each, such as the percentage of a target audience that recalls an ad.
5. 5Collect the data in three layers: objective (should-do), perceptual (plan-to-do), and behavioral (have-done). Keep them labeled separately. Stated perceptions and actual behavior can diverge, because actual purchases are constrained by availability, budget, and other factors that redirect customers from a preferred option to a secondary one.
6. 6Record the context for each preference. The same customer may value water outdoors and beer at a bar. A preference without its usage setting is incomplete.
7. 7Map each marketing input (product design, price, advertising, selling effort), plus competitor actions and environmental conditions, to the perception and preference outputs. Then evaluate against objectives and adapt the actions. This is the decision loop in the response-model framework.
8. 8Calibrate the links. Where you have experimental or market data, use objective calibration. Where the firm has never varied an input, or the history is stale (new entrants, price changes, shifting preferences), use judgment-based subjective calibration. Flag every estimate as one or the other.

NEC wants its mobile phone to be preferred. Following the passages, the team starts with the need, not the handset. A phone's physical attributes satisfy a low-order need, but safety sits higher in Maslow's hierarchy and is more important to the customer. NEC positions the phone as a safety device, especially for women. The marketing action is the positioning message. The outputs to track are awareness of that message, the perception that the phone is a safety device, and preference versus competing phones. Soap makers use the same move, stressing appearance and beauty (higher-order) over cleanliness (lower-order). The check on the result is behavioral. A stated preference is not a purchase, because availability and budget can redirect a customer to a secondary option. So the team compares plan-to-do perception measures with have-done sales, and re-calibrates if they diverge.

• If a preference was measured only by stated perception, then label it plan-to-do and test it against have-done behavior before forecasting sales from it.
• If the offering's functional need is low in the hierarchy, then test whether communication can link it to a higher-order need, as the NEC and soap examples do.
• If a preference differs by usage setting (water outdoors, beer at a bar), then segment the measurement by setting rather than averaging.
• If you have never varied an input, or your history predates competitive entries, price changes, or shifting preferences, then use subjective calibration and label it as judgment.
• If awareness is low, then fix communication before reading preference data, since a customer cannot prefer an offering they do not know.

• Treating perception as purchase behavior. The tell: forecasts built on survey preference overshoot actual sales, with no adjustment for budget or availability.
• Measuring only the physical attributes and ignoring the perceptual component. The tell: the product spec matches a rival's, yet customers still choose on brand image.
• Defining the need too broadly ('thirst'). The tell: preference data conflicts across settings because the usage context was never recorded.
• Calibrating on stale data. The tell: the model rests on year-old data from a fast-moving market such as smartphones, or on historical advertising spend that never varied.
• Collecting large volumes of data without extracting insight. The tell: dashboards full of metrics but no change in marketing actions.

• ▸Offering and target customer: state the product or service and the specific customer group.
• ▸Attribute components: list physical, service, and perceptual attributes separately.
• ▸Need statement: write the need from general to specific (e.g., thirst, then replenish lost fluids), with its physical and social dimensions.
• ▸Need level in the hierarchy: mark physiological, security, social, status, or self-actualization, and any higher-order link you can credibly claim.
• ▸Usage setting: record where and when the offering is consumed and how preference changes by setting.
• ▸Awareness measure and objective: name the metric (e.g., percent recalling the ad) and the target.
• ▸Perception and preference measures: record the current perception of key attributes and preference versus competitors.
• ▸Evidence tier: label each figure plan-to-do (stated) or have-done (behavioral).
• ▸Constraints and calibration: list availability and budget constraints that could redirect choice, and mark each estimate as objective or subjective.