Social Media Metrics - How to Measure and Optimize Your Marketing Investment (New Rules Social Media Series)

1. 1Start from the business objective, not the metric list. The source's advice, attributed to David, is to figure out your business objectives first and then apply metrics accordingly. Write down which of the Big Three the program serves: increased revenue, lowered costs, or improved customer satisfaction.
2. 2Split the construct into two layers. Layer 1 is interaction: number of interactions (55), interaction/engagement rate (56), frequency of social interactions per consumer (57), percentage of videos viewed (58), polls taken/votes received (59), tags added (66), and user-generated submissions received (62). Layer 2 is action: clicks (52), time spent on site through social media referrals (50), registrations from third-party social logins (68), contest entries (70), discount redemption rate (86), leads generated (88), products sampled (89), visits to store locator pages (90), and responses to socially posted events (97). Layer 1 shows people care. Layer 2 shows they did something.
3. 3Pick a small set of metrics, one or two per layer, that fits the objective. Revenue goals pull toward discount redemption, store locator visits, leads, and conversion change due to user ratings and reviews (91). Cost goals pull toward customers assisted (75) and savings per customer assisted (76). Satisfaction goals pull toward volume of customer feedback generated (80) and customer satisfaction (79). Drop the rest of the list for now.
4. 4Answer the three Chapter 5 questions for each program. Do people click through to your site from their social interactions? Do they engage with your organization in new and different ways? Are more people drawn into a profitable, sustainable relationship with you? If a metric answers none of the three, retire it.
5. 5Segment by source so you can tell which social activity caused the action. Use registrations by channel (69), method of content discovery (51), and percentage of traffic generated from earned media (53). Tag each blog post, tweet, and video with where it sends people, so a click or registration traces to its origin.
6. 6Follow each action to a cash or cost outcome. The source says that once you connect the dots to income, everybody knows what you are talking about. Link clicks and registrations to impact on online sales (84), impact on offline sales (85), customer retention (92), or customer lifetime value (93). Where the chain breaks, state why the metric is still worth tracking.
7. 7Measure the ratio, not just the count. Interaction/engagement rate (56) divides interactions by the audience reached. Pair it with a downstream rate such as clicks per interaction or redemptions per click, so a drop shows you which step failed.
8. 8Review against the objective each cycle. Keep, change, or kill each program by its action-to-outcome result. Reevaluate your Key Performance Indicators as you learn, while the Big Three goals stay fixed.

The source supplies the metric list and the contrast, not a numbered case, so this walk-through is an illustrative construction from its own metrics; the figures are placeholders you replace with yours. The foreword describes the old way: a B2B company judged success by white-paper requests, business cards in a fishbowl, and the 'thud factor' of a PR clip book. Those were counts of attention, and management accepted them. Now take a brand running a discount offer through blog posts, tweets, and a video. Layer 1 reports votes in a related poll (59), videos viewed (58), and an engagement rate (56) of, say, N interactions per audience reached. That shows people care, but per Chapter 5 you are 'only part of the way home.' Layer 2 asks what they did. Clicks (52) and time on site from social referrals (50) show arrival. Discount redemption rate (86) and visits to store locator pages (90) show action. Impact on online and offline sales (84, 85) shows revenue. Suppose the poll gets heavy participation but redemptions are flat. The program is entertaining without driving revenue, so you rebuild the call to action instead of celebrating the votes. If redemptions rise, you attribute them to source using traffic from earned media (53) and registrations by channel (69), then fund the channel that converts.

• If a metric cannot be connected back to income, cost savings, or customer satisfaction, then either drop it or state in writing why you are measuring it.
• If interactions are high but clicks, redemptions, or registrations are flat, then treat the program as an attention success and an action failure, and fix the call to action before adding reach.
• If the objective is lowered cost, then judge response by customers assisted (75) and savings per customer assisted compared with other channels (76), not by clicks.
• If you must show leadership value, then report the action layer (redemptions, leads, sales impact) first and the interaction layer second, in plain language rather than analytics jargon.
• If two channels show similar engagement, then fund the one with the better downstream action rate, using registrations by channel (69) and conversion change (91).

• Reporting engagement counts as the result. The tell: your report ends at likes, votes, and views, and nobody can say what changed in sales, cost, or satisfaction. This is the social-media version of the clip book's 'thud factor.'
• Measuring everything on the 100-metric list. The tell: a dashboard of dozens of numbers and no stated business objective. The source says to start with objectives and apply metrics accordingly.
• Using jargon in place of income. The tell: colleagues are baffled by terms like 'pass-along engagement velocity' and cannot evaluate whether the program is worthwhile.
• Counting clicks without tracing where they came from. The tell: you cannot say which post, channel, or discovery method produced the registrations or redemptions, so you cannot decide where to spend next.
• Ignoring the cost side. The tell: engagement is celebrated while the spend to produce it goes untracked, so you cannot say whether social is cheaper than the call center, store, or traditional research.

• ▸Business objective (Big Three): Name one: increased revenue, lowered costs, or improved customer satisfaction.
• ▸Program/content: Name the blog post, tweet series, video, poll, or contest being measured.
• ▸Layer 1 interaction metric(s): List one or two from interactions, engagement rate, video percentage viewed, votes, tags, or submissions, with the baseline.
• ▸Layer 2 action metric(s): List one or two from clicks, time on site from referrals, registrations, contest entries, redemptions, leads, samples, or store-locator visits.
• ▸Source tag: Record the channel or content tag that identifies where the action originated.
• ▸Outcome link: State how the action connects to online sales, offline sales, retention, lifetime value, or savings.
• ▸Chapter 5 check: Answer yes or no on click-through, new forms of engagement, and a more profitable relationship, with evidence.
• ▸Decision: Mark keep, change, or kill, and give the one-line reason tied to the objective.
• ▸Next-cycle KPI revision: Note any Key Performance Indicator to reevaluate based on what you learned.

1. 1Name the purpose before you pick a tool. Stage 1 of Owyang's eight-stage model is a listening program with no goals that uses the data for nothing. Write down the decision each listening feed must inform (brand attribute check, risk flare-up, campaign adjustment) before you set anything up.
2. 2Start with free alerting. Create a Google Alert on 'I would recommend (your product/company)' and a mirror alert on 'I would not recommend'. Treat it as a floor, not a program: the source says alerting alone without action on the data is useless.
3. 3Build three keyword lists, as Adam Greco did at Salesforce.com: (1) branded keywords that mention your company, (2) competitor keywords, (3) general industry keywords you care about (his examples: 'cloud computing', 'CRM'). Listening to competitors and the category shows you who is talking, about whom, and what they say.
4. 4Pull the data on a schedule and chart it. Greco uses the Twitter API to pull tweets matching those phrases hourly, charts them over time, and gets alerts by e-mail or cell phone on a spike. Run it for a period first to calibrate the background noise level, so you know what normal traffic looks like.
5. 5Set alarms in both directions. Alert on spikes, and also on a drop to near-silence. If the public stops talking about you altogether, PR and marketing need to be told their efforts are not keeping the brand in the marketplace's minds and voices.
6. 6Test your brand attributes. List the attributes you want associated with the brand (the source's examples: Wal-Mart=low prices, Apple=style, Coke=refreshing). Read the actual conversation to see whether those are the attributes people attach to you. Check which attributes people actually voice, not which ones you assume.
7. 7Review sentiment with a human in the loop. Tools such as Social Mention (positive-to-negative ratio) and Scout Labs (self-reported 70–80% accuracy) are approximate. Reclassify misread items so the system learns, and read a sample yourself, especially where humans are playful or sarcastic.
8. 8Move up the maturity ladder deliberately. From tracking mentions (Stage 2), add staff who seek out discussions and signal internal teams (Stage 3: reduce flare-ups, find prospects, win unhappy competitors' customers). Then add a dedicated resource to watch a campaign in flight and make course corrections near real time (Stage 4). Keep talking to customers directly, since surveys and conversation reveal things they won't say to followers.

Adam Greco, director of Web Analytics at Salesforce.com, is the source's model. He integrated Omniture's SiteCatalyst with Twitter and set up three keyword groups: branded terms, competitor terms, and industry terms such as 'cloud computing' and 'CRM'. Using the Twitter API, the system pulls every tweet matching those phrases once an hour. The results feed charts and graphs that track volume over time. E-mail or cell phone alerts fire when there is a spike. His stated goal is to find out who is talking about Salesforce.com, who is talking about competitors, and what they say. Separate staff handle outbound work (a YouTube channel and a Facebook channel), while Greco analyzes what people say about the company. Against Owyang's model, this sits at Stage 2 or Stage 3: tracking plus alerting. It becomes Stage 3 once a person acts on a spike or flare-up and signals an internal team.

• If you have a listening program but cannot name the decision it informs, then you are at Stage 1: define the goal before buying a platform.
• If mention volume spikes beyond the calibrated background level, then read the underlying posts and notify the owning team (PR, marketing, support) before the flare-up spreads.
• If mention volume drops well below normal, then alert PR and marketing that the brand is falling out of the conversation.
• If you need depth and tonality, not just counts, then add human review and reclassification on top of automated sentiment, since the source puts accuracy at roughly 70–80% for one vendor.
• If you want to know what drives satisfaction, then do not rely on what customers say is critical. Test it against drivers (the ForeSee point: price may matter less than better site search), and keep direct conversation going alongside social listening.

• Counting without acting: reports of mention volume that nobody uses. The tell is a dashboard with no named owner or decision attached (Stage 1–2).
• Trusting automated sentiment blindly: playful phrasing such as 'I recommend Hummers for those who are brainless about the environment' reads as positive to a keyword match. The tell is a sentiment ratio that doesn't match what you read when you sample posts.
• Exact-phrase retweet tracking: retweets morph like a game of Telephone, so strict string matching undercounts. The tell is propagation numbers far lower than your own sense of the spread.
• Listening only to your own brand: ignoring competitor and industry keywords leaves you without context. The tell is that you can't say who is talking about competitors or what they say.
• Assuming your brand attributes are the ones people hold: you take your intended positioning as fact. The tell is never having read the conversation to see whether the associations you want actually appear.

• ▸Decision to inform: the specific decision this listening feed supports (attribute check, risk flare-up, campaign tweak).
• ▸Branded keywords: phrases that mention your company and products.
• ▸Competitor keywords: phrases naming competitors and their products.
• ▸Industry keywords: general category terms you care about (e.g., 'cloud computing', 'CRM').
• ▸Pull frequency and source: how often and from where you collect (e.g., hourly via API, Google Alerts).
• ▸Baseline mention level: the normal background volume you calibrated; note the spike and drop thresholds.
• ▸Alert routing: who is notified by e-mail or phone on a spike or drop, and what they do.
• ▸Target brand attributes vs. observed: attributes you want vs. what the conversation shows.
• ▸Sentiment review log: sample size read by a human, misclassifications corrected, and current maturity stage (1–8).

1. 1Start with your own goals, the company's goals, and your budget limits for gathering and quantifying data. The source calls these 'the entrance fee' to any conversation about social media. Write them down before you approach anyone, and if you can't state them, find out first.
2. 2Agree on definitions before you argue for resources. Use the source's working definition: social media is consumer-generated content distributed through easy-to-access online tools, so anybody can communicate with everybody. Settle what 'engagement', 'buzz' and 'influence' mean with each stakeholder. Heuer's point is that perception is shaped by past experience, role requirements and organizational norms, so two executives can hear the same word differently.
3. 3Profile each decision-maker's measurement history. Executives raised on sales leads and the PR clip book's 'thud factor' accept those as proof of success. Their bosses probably did not grow up with Internet access. Assume they are slow to embrace new methods, not dumb, and plan to teach them rather than to win a debate.
4. 4Translate every social metric into one of the Big Three business outcomes: increased revenue, lowered costs, or improved customer satisfaction. Chapter 7's test is that social media does no good unless you can tell whether it produced one of these. Do not take a raw count to a senior manager as the case for funding.
5. 5Make the argument in three claims, matching the source's framing of Chapter 8: social media is inevitable, it is a vital part of the marketing mix, and it is a pathway to profits that deserves resources. Attach one piece of evidence to each. For example, cite the ENGAGEMENTdb finding that socially engaged companies are more financially successful, or the Cluetrain idea that the marketplace is a conversation.
6. 6Pick a small set of KPIs from the 100-metric list rather than presenting the list. State which of your goals each KPI serves, and say plainly which questions you cannot yet measure.
7. 7Build cross-functional support the way Best Buy did. Involve the groups that carry risk and reach, such as the Employee Customer Care Group and the legal team, to produce public participation guidelines. Describe them as 'lane bumpers with a compass north' rather than tight controls.
8. 8Revisit your goals and KPIs after the first results come in. Your 'Big Three' goals persist, but your KPIs are likely to need reevaluation once you see how social media really works. Report back to the same executives so support is sustained, not just granted once.

Best Buy's Twelpforce shows buy-in built through structure and honest limits. The initiative lets employees jump on Twitter and answer any customer question. Best Buy introduced it to many people through TV ads showing lone customers addressing thousands of service representatives in a stadium. Social media manager John Bernier says the company's path to social was 'fairly organic' and collaborative, with no senior director of social and layered hierarchy above it. The Employee Customer Care Group created the forums and worked with legal to publish blogging and social media guidelines. Those guidelines are deliberately not 'too constrictive', and they remind store employees that saying they work for Best Buy places part of the brand's reputation in their hands. Asked what metrics prove the program's value, Bernier admits there are things they would like to measure but cannot yet, such as how Twelpforce compares with Dell Outlet and how positive word-of-mouth affects national store traffic. The lesson for you is to secure legal and operational allies early, set guardrails that make sharing safe, and name the unmeasured questions up front instead of overclaiming.

• If a stakeholder judges success by older measures (leads, clip counts, booth traffic), then translate your results into revenue, cost or customer satisfaction before presenting. Do not ask them to value a new unit.
• If you cannot state your goals, company goals and budget limits, then stop pitching and get them first. Without them you have no basis for a resource request.
• If a term like 'engagement' or 'influence' draws different reactions from different people, then pin down the definition in writing before debating results.
• If you cannot yet measure an outcome executives care about, then say so and name the question, as Best Buy did. Do not invent a proxy to look complete.
• If your program touches employees or customer-facing staff, then bring in legal and the customer-care owners first and publish light guidelines. This builds internal support and lowers risk.

• Pitching social media as an inevitable trend with no link to profit. The tell: executives nod but ask 'so what does it get us?' and no budget moves.
• Presenting a long metric list as the case, for example all 100 metrics. The tell: eyes glaze, and you cannot say which number maps to revenue, cost or satisfaction.
• Treating slow executives as obstacles instead of people with different norms. The tell: you reference tools and jargon, and their questions keep returning to clip-book-style proof.
• Letting stakeholders use the same words with different meanings. The tell: after a report, the same figure is called 'good' by one person and 'meaningless' by another.
• Overclaiming to cover measurement gaps. The tell: a number is attributed to social media that you cannot defend when asked how it was derived.

• ▸Our goals: state the program goals in one sentence each.
• ▸Company goals and budget limit: list the company goals you serve and the ceiling on data-gathering spend.
• ▸Stakeholder and their current success measure: name each executive or colleague and what they accept as proof now (leads, clips, traffic, etc.).
• ▸Agreed definitions: write the working definition of each key term (social media, engagement, influence) and who signed off.
• ▸Big Three mapping: for each KPI, mark whether it supports revenue, lower costs, or customer satisfaction.
• ▸Three-claim evidence: one piece of support each for inevitable, vital to the marketing mix, and a pathway to profits.
• ▸Internal allies and guardrails: list the groups consulted (e.g., legal, customer care) and the guidelines drafted.
• ▸Known measurement gaps: list the questions you cannot yet answer and say so openly.
• ▸Review date: set when goals and KPIs will be reevaluated with these stakeholders.

1. 1Name the Big Three goal first. Write down which of increased revenue, lowered costs, or improved customer satisfaction the program serves. The source is blunt: work that improves none of them wastes time and money. Doing all three is fine, but you must name at least one.
2. 2Lay out the dependency chain backward from the bottom line: profits need income, income needs customers, customers need prospects, prospects need suspects, suspects need awareness. Mark where your social activity sits on that chain. This tells you which outcome it can credibly influence.
3. 3Pick outcomes from the measurable list, one per chain link you are actually influencing. The options are awareness, survey completions, subscriptions, registrations, blog comments, blog posts (mentions to others), leads, and purchases. Phrase each as a question, such as 'How many signed up for the newsletter, blog, or tweets?'
4. 4Run the connect-the-dots test on every metric. If it cannot be connected back to income, cost, or satisfaction, write down why you are spending time measuring it. If you have no reason, drop it. Exotic metrics such as 'sentiment volatility rate' or 'pass-along engagement velocity' earn their place only when you can trace them to a Big Three result.
5. 5Set clear, measurable objectives, per Katie Paine's step two. State the problem you are solving, and do nothing in social media that adds no value. 'You can't manage what you can't measure' means each outcome gets a measurable target.
6. 6Use outcomes as navigation, not as a quarterly report. The bottom line is the rearview mirror. Outcomes are the windshield, so review them on a cadence short enough to adjust mid-program instead of waiting for month-end or quarter-end results.
7. 7Re-examine your KPIs after you learn how the channel behaves. The Big Three stay fixed, but the indicators are likely to need re-evaluation, so revise them while keeping the goals.
8. 8Report in the language of money and satisfaction. Lead with the Big Three result, then show the outcome chain beneath it, so colleagues see income, cost, or satisfaction instead of analytics vocabulary.

The foreword describes a B2B marketing VP who judged success by two measures: sales leads (people who requested a white paper or dropped a business card in a fishbowl) and a PR clip book whose 'thud factor' signaled a good month. Run both through the outcome method. Sales leads sit in the middle of the chain, between suspects and prospects, and map to increased revenue. They are a legitimate outcome, but they stop short of purchases, so the missing link is how many leads actually bought. The clip book measures awareness at the base of the chain. Its thickness says nothing about whether awareness became subscriptions, registrations, leads, or purchases. Apply the same logic to social media: blog subscriptions and registrations are the 'suspects to prospects' links, leads are next, and purchases confirm revenue. Any metric that cannot be tied to a link in this chain, such as a fancy engagement ratio, must justify itself or be dropped. The result is a short ladder of outcomes, each answering 'are we still going in the right direction?' instead of just 'how big was the pile?'

• If a metric cannot be connected back to income, lower cost, or customer satisfaction, then either state explicitly why it is worth tracking or stop measuring it.
• If you only look at results at month-end or quarter-end, then add mid-stream outcome indicators (subscriptions, registrations, leads), because you run programs in real time and need to adjust along the way.
• If an activity adds no value toward a measurable objective, then do not do it in social media (Paine's step two).
• If you are tempted to present a coined analytics ratio to colleagues, then translate it into the Big Three result first; use the jargon only if you can connect the dots.
• If your outcome list stops at awareness or leads, then extend it down the chain to purchases, since you cannot have profits without income or income without customers.

• Measuring awareness and stopping there, the modern clip-book 'thud factor'. The tell: reports full of impressions, mentions, or comment counts with no leads or purchases next to them.
• Reporting only the bottom line. The tell: you learn whether you hit the goal only after the quarter closes and cannot say which activity moved it, because there are no intermediate indicators.
• Inventing impressive-sounding composite metrics that baffle colleagues. The tell: you cannot explain in one sentence how the metric ties to revenue, cost, or satisfaction.
• Doing social media activity with no defined problem to solve. The tell: lots of data and pretty charts, but nobody can state the question the data was meant to answer.
• Keeping outdated KPIs after learning how the channel really works. The tell: the goals still make sense, but the indicators were set before you understood the audience's behavior and have never been reviewed.

• ▸Big Three goal served: Write which one or more of increased revenue, lowered costs, improved customer satisfaction this program targets.
• ▸Problem to solve: State in one sentence the specific problem this social activity addresses.
• ▸Chain link influenced: Mark where it sits: awareness, suspects, prospects, customers, income, or profits.
• ▸Outcome metric: Choose from awareness, survey completions, subscriptions, registrations, blog comments, blog posts, leads, purchases.
• ▸Question the metric answers: Phrase it as 'How many...?' tied to the chain link.
• ▸Measurable target: Give the number or change that would count as success, set before you start.
• ▸Connect-the-dots statement: Explain in one sentence how this outcome leads back to income, cost, or satisfaction; if you cannot, justify or drop it.
• ▸Review cadence: Name how often you check it so you can adjust before month-end or quarter-end.
• ▸KPI re-evaluation note: Record what you learned about the channel that changes the indicator, while keeping the Big Three goal fixed.

1. 1Write the business goal before opening any tool. State it in one sentence tied to revenue, cost, or customer satisfaction. The source frames the job as 'measuring the business value of social media,' not its 'size and popularity.' A goal like 'be active on Twitter' fails that test.
2. 2Answer the source's own recurring question, 'How valuable?', in dollars or in a customer-satisfaction outcome. If you can only answer 'that is an excellent question,' you are not ready to measure.
3. 3Name the action you want from the audience. The source's action questions work as templates: do people click through to your web site, engage with the organization in new ways, or move into a profitable, sustainable relationship? Pick the one your goal depends on.
4. 4Treat long metric lists as a menu. The source reproduces a 100-item list (buzz volume, fans, followers, embeds, sentiment, share of voice, and more). Select only the 3-5 metrics that would move if the goal were being met, and write down why each one would move.
5. 5Sort each chosen metric by what it tells you: reach (audience size), resonance (message pass-along), sentiment (opinion shift), or response (clicks, new engagement). Confirm that at least one metric is a response metric. Reach and buzz alone never reach revenue, cost, or satisfaction.
6. 6Weight raw counts by audience quality. The source scores a subscriber or follower at 1, a reader at 1, a fan at 2, and a repeater at 4. Set a scoring rule up front, so a bigger number reflects more goal-relevant attention and not just more people.
7. 7Set a baseline and a before/during/after comparison. The source lists 'shift in sentiment before, during, and after social marketing' as a metric. Record the starting values and the time window before launch, or the results cannot be tied to the goal.
8. 8Schedule a review that can change the plan. The source says measurement should show 'how to tweak your tweets.' At each review, ask what the data says about the goal and what you will change. Keep or drop each metric on that basis.

The source's new-car announcement shows goal-driven reading of data. Suppose the goal is to build purchase interest in a newly launched vehicle (the goal here is illustrative; the source does not state one). You track online conversation by topic: style, horsepower, and mileage. The data shows strong interest in style that 'drops off pretty quickly' once people have seen photos and passed along 'Did you see this?' The horsepower and mileage discussions keep going. Because the goal is sustained interest, you stop treating the style spike as success and recognize that horsepower and mileage carry the lasting conversation. You now know what intrigues people most, which the source calls 'marketplace insight you can act on.' The mileage discussion also ignites 'a healthy amount of conversation in the environmental population,' an audience you had not targeted. With a defined goal, that group counts as a new audience to engage. The source adds that you should have known this community would care, and that reaching it through others is more powerful. Without a goal, the same charts are only activity counts.

• If you cannot state the goal as revenue, cost, or customer satisfaction, then stop and define it before choosing any metric.
• If a metric would not change your decision whichever way it moves (for example, follower count by itself), then drop it or weight it by audience type.
• If your metric set contains only reach and buzz measures, then add at least one response metric (click-through, new engagement, relationship) that connects to the goal.
• If conversation spikes on one topic and then fades, then judge it against the goal. Shift effort toward the topics that keep the conversation going.
• If an unplanned audience shows up, then decide whether it serves the goal and engage it deliberately. Do not discard it as noise.

• Starting from the metric list. The tell: your dashboard has dozens of items (buzz, fans, embeds, ratings) and no sentence says what business result they serve.
• Measuring size and popularity instead of business value. The tell: your reports lead with follower or fan totals and never mention revenue, cost, or satisfaction.
• Counting every subscriber the same. The tell: a large subscriber count sits beside flat reads and no repeaters, because someone subscribed once and has since ignored each post.
• Citing the correlation that 'socially engaged companies are more financially successful' as your own goal. The tell: the goal reads 'increase engagement' with no link to your own financial outcome.
• Declaring success on a short-lived spike. The tell: a launch spike (like the style discussion) gets reported as a win, but the topic has gone quiet and no response action was measured.

• ▸Business goal (one sentence): State the goal in terms of revenue, cost, or customer satisfaction.
• ▸Value question: Write what 'how valuable?' means here in dollars or a satisfaction outcome.
• ▸Desired audience action: Pick one: click through to the site, engage in new ways, or enter a profitable relationship.
• ▸Chosen metrics (3-5): List only metrics that would move if the goal were met, each with a one-line reason.
• ▸Metric type: Mark each as reach, resonance, sentiment, or response. Confirm at least one is response.
• ▸Audience weighting rule: Define the point values (such as subscriber 1, reader 1, fan 2, repeater 4) or your own.
• ▸Baseline and window: Record starting values and the before/during/after dates.
• ▸Review date and decision: Set the date to review and name the action you will take if the numbers are flat.
• ▸Unexpected audiences: Note any group you did not target that showed interest, and whether you will engage it.

1. 1Inventory every outbound asset by type: blog posts, tweets, videos, apps, widgets. Activity is what you publish and take part in, so list the assets before you count anything else.
2. 2Attach to each asset type the output counts that match it from the Berkowitz list: uploads (21), pageviews for blogs and microsites (29), subscriptions to RSS, podcasts or video series (28), downloads (20), embeds/installs (19), and polls taken/votes received (59). Counting posts alone measures volume and says nothing about quality.
3. 3Add quality ratios beside the raw volume: user-initiated views of videos (22), percentage of videos viewed (58), the ratio of embeds or favoriting to views (23), and asset popularity (10), meaning which of several embeddable assets gets used most. Review volume and ratios together.
4. 4For widgets, instrument the full life: Hosted, Viewed, Grabbed, Install, Active, Uninstall, Mouseover, Time on page, Event. Report Viewed and Active, not only Install. An installed widget that is never viewed is only an opportunity to see, 'as useless as a billboard in the dark.'
5. 5For tweets and blog posts, estimate reach by walking the retweet cascade: your followers, how many retweeted, how many followers they have, how many of those retweeted, and so on. For blog posts, use comments and trackbacks the same way. Treat the result as a back-of-the-envelope estimate.
6. 6Set up monitoring to catch the spread: Tweetbeep, trackbacks and Google Alerts keep a watch on where your content travels.
7. 7Filter bots before totaling 'people' who had a chance to see a message. Robots can scrape, follow, friend, tweet and retweet, so a node in the cascade may be software. Count humans only, or label the number as mixed.
8. 8Expect retweets to morph like a game of Telephone. Match on more than the exact phrase, and record the cascade figure as an estimate with its known fuzziness. The source paper warns that keeping track of who truly pays attention to whom is 'hopeless' to do exactly.
9. 9Publish the numbers with a note on their precision. Fuzzy numbers presented as exact are what wilt on the way to the CMO's office.

The source's widget lifecycle is the clearest case. A brand publishes a widget and tracks nine stages: Hosted, Viewed, Grabbed, Install, Active, Uninstall, Mouseover, Time on page, Event. Installs run into the thousands, which shows high affinity with the brand, so a report built on installs looks like success. But Active is defined as viewed within a set time frame, and the source notes a widget 'may get installed but never viewed.' The expert therefore leads the report with Active against Install, then adds Mouseover and Event (clicks on tabs, video plays) to show interaction. Uninstall goes in as the counter-signal. Each figure is labeled approximate, because the widget-measurement discussion warns that these numbers do not add up precisely. The headline moves from 'thousands grabbed' to 'how many are actually in front of eyes and being used,' which is activity measured on quality, not volume.

• If an asset can be installed, embedded or downloaded, then report the view/active count beside it, because a grab is only an opportunity to see.
• If you total reach from retweets or trackbacks, then strip out bots and label the figure an estimate, because a node may be software and retweets change wording.
• If you compare publishing channels, then use ratios (embeds or favoriting to views, percentage of videos viewed), not raw counts, because volume alone ignores quality.
• If a number will go to executives, then state its precision limits up front, because fuzzy numbers presented as exact lose the room.
• If an output metric moves, then check it against an audience-side measure (subscriptions, pageviews, comments) before calling it a result, since publishing is only the first half of the activity picture.

• Counting posts or tweets as success: the report shows output volume with no views, pageviews or subscriptions beside it.
• Celebrating installs or grabs: the widget count is high, but nobody can say how many were ever viewed or active.
• Treating the retweet cascade as exact reach: the figure is quoted to the person with no note on bots, morphing wording or estimation.
• Counting bots as audience: follower and retweet totals include scraping or auto-following accounts and nobody checked.
• Tracking only exact-phrase retweets: reach looks small because edited retweets never match the string.

• ▸Asset and type: name the blog post, tweet series, video, app or widget.
• ▸Publish date and channel: record when and where it went out.
• ▸Output counts: uploads, pageviews, downloads, subscriptions, embeds/installs, polls/votes as applicable.
• ▸Viewed/Active count: how many were actually viewed or active in the set time frame.
• ▸Quality ratio: embeds or favoriting to views, or percentage of videos viewed.
• ▸Interaction signals: mouseovers, events, comments, trackbacks.
• ▸Cascade reach estimate: followers, retweeters, their followers, and so on, labeled as an estimate.
• ▸Bot check: how you separated human from robot activity, and the share excluded.
• ▸Precision note: the known fuzziness to state when reporting this number.

1. 1Define the target audience before you count anything. Reach is a percentage of the people you want, not of everyone. Write the target as a size (the denominator) with demographics and geography. The source's metric list separates 'demographics of target audience engaged' (#41) from 'demographics of audience reached' (#42), and you need both to see whether the audience you reach is the audience you want.
2. 2Inventory the channels where the message lives. Use the source's categories of social media (forums and message boards, blogs, media sharing such as Flickr and YouTube, Twitter, social networks, and so on). Record 'buzz by social channel' (#8) so reach is never one blended number.
3. 3Count first-degree reach per channel with the source's direct-audience metrics: fans (#12), followers (#13), friends (#14), and subscriptions to RSS, podcasts, and video series (#28). Pageviews for blogs and microsites (#29) cover owned destinations. Keep each as its own line, because a fan, a follower, and an RSS subscriber are different kinds of opportunity to see a message.
4. 4Add second-degree reach (#18): the connections of your fans, followers, and friends who are exposed, counted either by people or by impressions. Record which unit you used, and never add a people figure to an impressions figure.
5. 5Measure the growth rate of fans, followers, and friends (#15) and the rate of virality or pass-along (#16), plus its change over time (#17). A static head-count says nothing about whether reach is expanding or about to flatten.
6. 6Convert to a percentage. Divide unique target-audience people reached (first-degree plus second-degree, de-duplicated as far as your tools allow) by the target-audience size from step 1. Show it next to the raw count.
7. 7Qualify the number with who is in it. Check influence of consumers reached (#38), influence of publishers reached (#39), and demographics of audience reached (#42). A large count made of non-target or low-influence people is not the reach you set out to buy.
8. 8Tie reach to cost and to what it feeds. Compute effective CPM from spend per impressions received (#30) so you can compare channels. Then pass the reach figure to the engagement measures (likes, comments, interactions) downstream as the denominator, so reach is not the only number reported.

The source's own contrast is the clearest anchor. The foreword's author, then a B2B marketing VP, measured reach with the number of people who stopped by the trade show booth and a PR clip book judged by its 'thud factor.' He calls those measures 'wholly inadequate online.' Now apply the reach method to a hypothetical brand; the numbers are illustrative, not from the source. Say the target audience is 200,000 people. The brand has 6,000 Twitter followers, 4,000 Facebook fans, and 1,500 RSS and video-series subscribers. Using the source's second-degree metric (#18), you estimate that the connections of those fans and followers who were actually exposed come to another 30,000 people. You label the unit 'people,' not 'impressions.' Counting only first-degree reach, you report 11,500 audience members, or 5.75% of the target. With the second-degree figure added and de-duplicated, the same data supports a higher percentage, and the report shows both lines. You then check the growth rate of fans and followers (#15) and the pass-along rate (#16), and run demographics of audience reached (#42) against the target. If most of the reached people sit outside the target, the headline percentage is overstated and you report the in-target subset instead. The result is a number with a stated denominator, a stated unit, and a stated quality check, which a clip-book thud never offered.

• If a number has no target-audience denominator, then report it as a raw count (fans, followers, friends) and do not call it reach.
• If you include second-degree exposure (#18), then state whether it is counted in people or in impressions and keep it on a separate line from first-degree reach.
• If demographics of audience reached (#42) diverge from your target (#41), then report reach for the in-target subset only and treat the rest as spillover.
• If follower or fan growth (#15) is flat while pass-along (#16) is falling, then treat reach as at risk even when the total looks large, and look at content or channel changes before adding spend.
• If you compare paid and earned channels, then use effective CPM (#30) on impressions received so spend is judged on exposure delivered, not on audience size claimed.

• Reporting follower or fan counts as reach. The tell: the slide shows a big number and no percentage or target definition. The source itself warns against an 'obsession with the number of followers.'
• Copying offline logic online, the 'thud factor' of volume for its own sake. The tell: success is declared on the size of a pile (mentions, clips, booth traffic) with no check on who saw it.
• Gating content behind registration to generate 'leads.' The foreword notes this blocked sharing. The tell: first-degree numbers look tidy but pass-along and second-degree reach collapse.
• Mixing units. The tell: second-degree reach in people is added to impression counts in one total, so the figure cannot be reproduced or compared period to period.
• Ignoring who was reached. The tell: reach is high but influence (#38, #39) and demographics (#42) of the audience reached were never checked, so the number cannot be defended against the target.

• ▸Target audience definition and size (denominator): Describe who counts as target (demographics, geography) and enter the total number of people.
• ▸Channel: Name the single channel measured (e.g., Twitter, Facebook, YouTube, blog, RSS/podcast).
• ▸First-degree audience: Enter fans, followers, friends, or subscribers for that channel, labeled by type.
• ▸Second-degree exposure: Enter connections of those fans/followers/friends exposed, and mark the unit (people or impressions).
• ▸Growth rate and pass-along rate: Enter the period-over-period growth of fans/followers/friends and the virality rate, plus the change from last period.
• ▸In-target share: Enter the percentage of those reached who match the target demographics (#42 against #41).
• ▸Influence check: Note the influence of consumers and publishers reached (#38, #39) in a line.
• ▸Reach percentage: Divide de-duplicated in-target people reached by the target size and write it beside the raw count.
• ▸Effective CPM: Enter spend divided by impressions received for the period, for channel-to-channel comparison.

1. 1Define who carries your message. List the individuals (bloggers, tweeters, repeaters, and people inside your own organization) and score each on three things: audience size (subscribers/readers), connectedness (who they link to and are linked by), and voice (how often others repeat or cite them). The source says a post with no readers or a tweet with no retweeters is 'just as useless'.
2. 2Separate three roles in your scoring: subscriber, reader, and repeater. The source's example is Mike, a repeater who saw something from Ryland and passed it along. Mike may be neither a subscriber nor a reader of the original. Score repeaters on how many people subscribe, read, and repeat them, and expect the numbers to get complex quickly.
3. 3Weight scores by channel. Tweets are time-dependent: people ignore or forget old ones, dip in and out of the stream, and old tweets are dropped from Twitter's database. Blogs are relatively permanent, indexed by Google, and their opinions and links stay findable, readable, and clickable. Give blog citations more weight than tweets, and record the weights you chose.
4. 4Code every link you publish so republished and re-retweeted copies can be traced. Turn www.example.com into www.example.com?1234, then count how often code 1234 shows up in your analytics database. That count shows how far the post or tweet traveled, and who carried it.
5. 5Compute your message multiplier from three readings: how many people thought the message worth repeating, how quickly it spread (velocity), and the scope of its dissemination (reach). Together they tell you whether the insight is popular or prosaic.
6. 6Track citations. Per Avinash Kaushik, a citation occurs when people comment on your blog posts on their own blogs, and Technorati rank can track it. Technorati Authority is the number of blogs linking to a site in the last six months. A smaller Technorati Rank means a stronger blog.
7. 7Rank carriers two ways: by how many people they help you reach, and by how many valuable people they help you reach. Address the second group directly, personally, and often. The source prefers exclusive 'insider' info over a gift basket, because it makes them more valuable to their own readers and repeaters.
8. 8Use the results to pick your spokespeople. Velocity and reach also show who within your organization is best equipped to be the face of the firm in social media. Compare those people against external carriers using the same scores.

The source's own example is a new-car launch. You publish coded links in posts and tweets, then count each code in your analytics database as the posts get re-re-retweeted (Figure 3.1). You then follow subject matter along the social graph. Figure 3.2 shows strong interest in styling that drops off quickly once people have seen photos and passed along 'Did you see this?'. Horsepower and mileage conversations keep growing. Figure 3.3 shows the mileage discussion igniting an environmental population you had not targeted, and that conversation caught fire elsewhere. The carriers in that unexpected group are the influence finding. You now know which subjects multiply your message and which unplanned audience is amplifying it. You can act on this by tweaking your tweets toward mileage and by contacting the influential environmental voices directly with insider information.

• If a carrier has a large audience but nobody repeats or cites them, then rank them low. Reach without repetition is the tree falling with nobody to hear it.
• If a carrier reaches fewer people but those people are more valuable to you, then place them in the top tier and contact them directly, personally, and frequently.
• If you are comparing tweet-based and blog-based influence, then weight blog citations higher. Blogs are permanent and indexed, while tweets are ephemeral.
• If the message multiplier is low (few repeaters, slow velocity, narrow scope), then treat the message as prosaic and change it, not the distribution.
• If coded links show traffic from a group you did not target, then add that group's carriers to your outreach list and track their subject.

• Counting audience size as influence. The tell: a high follower or subscriber count with no retweets, comments, or citations behind it.
• Ignoring repeaters who are not subscribers. The tell: your influence list contains only subscribers, yet traffic arrives through people you have never scored (the Mike case).
• Treating tweets and blogs as equivalent. The tell: a burst of tweets scores like a lasting blog citation, and the credit vanishes when old tweets age out of the database.
• Publishing uncoded links. The tell: you can see mentions but cannot trace clicks back to the original post or the carrier who spread it.
• Keeping score without acting. The tell: you hold a ranked list but send no personal contact or insider information to the top tier, and the talk about you fades.

• ▸Carrier name and channel: record who they are and whether they publish on blog, Twitter, or another venue.
• ▸Role: mark subscriber, reader, or repeater (a repeater may be neither of the others).
• ▸Audience size: record subscribers/readers for this carrier.
• ▸Connectedness: note who links to them and how many blogs cite them (e.g., Technorati Authority, last six months).
• ▸Channel weight: write the weight applied (tweet vs. blog) and why.
• ▸Coded link ID: list the code (e.g., ?1234) assigned to their shared posts and the click count from analytics.
• ▸Message multiplier: record how many repeated it, how fast it spread, and how widely.
• ▸Value of audience reached: note whether the people they reach are the valuable ones, yes/no and why.
• ▸Next action: note the personal contact or insider information you will send, and the date.

1. 1Define the measurement target first. Sentiment covers the brand, its products, and its people, and each needs its own tracking stream. Source and target are separate variables, so record both for every post: where it appeared (Twitter, blog, news, reviews) and what it is about.
2. 2Score every post on two axes. Polarity is the color chart: positive (green), negative (red), neutral (gray). Intensity is the thermometer: how hot the emotion runs. Never collapse them into one number. 'I hate this!' said quietly, laughing, petulantly, or vehemently carries the same words and different intensity.
3. 3Choose a counting basis and report both. The source list names 'sentiment by volume of posts' and 'sentiment by volume of impressions' as separate metrics. A hundred negative posts that nobody read differ from ten that were widely seen.
4. 4Pick an automated tool, then audit it by hand. Bag-of-words classifiers key on strong words, so 'poor' and 'losing' trigger negative. Sample the machine's classifications and reverse the wrong ones, as Twitter Sentiment allows humans to do.
5. 5Flag the known failure zones for manual review: sarcasm, irony, idioms, slang, and facts that read as opinions. 'The pot roast is inedible' might be a statement of fact. 'Sinful' is praise when applied to chocolate cake. 'Better than your mother's' may be damning with faint praise.
6. 6Favor longer text where you can. Once you are free of Twitter's 140-character limit, sentiment is easier to recognize mathematically. Weight your confidence toward blogs, news, and reviews over single tweets.
7. 7Track sentiment over time, not as a snapshot. Record the shift before, during, and after each social marketing program, since the value lies in attitudinal shifts that opinion polls, surveys, and satisfaction questionnaires miss.
8. 8Tie the trend back to a business outcome before you report it. Terms like 'sentiment volatility rate' baffle colleagues until they connect to income, cost, or customer satisfaction.

Sterne tests the Twitter Sentiment tool built by Stanford students. The first tweet on the list is color-coded negative. The machine reads the words 'poor' and 'losing' and scores a bag of strong negative words. But the tweet is about the Bears, not the Tigers, and it is a very positive tweet about the Tigers. Sterne reverses the polarity manually, and that one correction shifts the displayed communal sentiment about the Detroit Tigers by 6 percent. A second tweet ('Sox lose 5-3 to Tigers in finale... I LOVES MY TIGERS! GO DETROIT!') is also classified negative, and he is still puzzled why. The lesson: a single misread polarity moves the aggregate visibly, so a small audit sample can change the number you report. The developers call the tool 'strictly a school project,' so a business should move to the more serious sentiment services.

• If the text is a single tweet or a short snippet, then treat the automated polarity as low confidence and sample it by hand. If it is a longer blog, news item, or review, then trust the math more.
• If a post contains a word with a flipped or ambiguous meaning (sinful, inedible, 'better than your mother's'), then route it to a human reader rather than accepting the classifier's score.
• If the emotional words are strong but the target is a rival or a different entity (Bears vs. Tigers), then check the subject before counting the post against your brand.
• If negative volume rises, then check intensity and impressions before reacting. Quiet, low-reach dissatisfaction and vehement, widely seen outrage call for different responses.
• If a sentiment figure cannot be linked to revenue, cost, or customer satisfaction, then do not present it to colleagues alone. Attach the business outcome or drop it.

• Treating word counts as meaning. Tell: a tweet about your team's win gets scored negative because it contains 'poor' and 'losing'.
• Reporting polarity without intensity. Tell: a calm complaint and a furious one sit in the same red bucket, so every negative spike looks equally urgent.
• Trusting automation without human review. Tell: sarcasm, slang, and idioms are counted at face value and the aggregate drifts from what readers would conclude.
• Judging sentiment from low-volume feedback channels. Tell: the Sterne example of five to ten responses a day per article, across over 10,000 articles, gives no usable picture and leads to 'looking into textual analysis tools.'
• Reporting jargon metrics to the business. Tell: colleagues glaze over at 'sentiment volatility rate' and nobody can say what decision the number supports.

• ▸Target: brand, product, or person being scored; keep one sheet per target.
• ▸Source: Twitter, blog, news, or reviews; note that longer text scores more reliably.
• ▸Time window: dates for before, during, and after any social marketing program.
• ▸Polarity counts: positive, negative, neutral number of posts.
• ▸Intensity note: mild, moderate, or vehement for the highest-reach negative and positive items.
• ▸Counting basis: sentiment by volume of posts and by volume of impressions, side by side.
• ▸Human audit sample: number of posts hand-checked and number of polarity corrections made.
• ▸Ambiguity flags: posts with sarcasm, irony, slang, or fact-like opinions routed to manual review.
• ▸Shift vs. prior period: change in sentiment before, during, and after the program.
• ▸Business link: the income, cost, or customer-satisfaction outcome this trend is tied to.