How to Define KPIs for Performance, Social and Retention Marketing Teams

Key Takeaways

• Performance, social and retention marketing have different economics and different lag times. One efficiency metric applied across all three makes the two slower disciplines look like failures.
• Performance marketing is measured on marginal return, not average — average ROAS tells you what past spend did, not where the next pound should go.
• Social and brand marketing create demand rather than capture it, so measuring them on last-click conversions is a category error that reliably ends with the budget cut.
• Retention marketing owns the metrics that decide most of lifetime value: repeat-purchase rate, cohort retention curves, and net revenue retention — measured by cohort, never blended.
• Attribution over-credits whatever is easy to track, which is almost always the capture channel. Use attribution to optimise within a channel and incrementality testing to decide between channels.
• Every discipline needs a guardrail metric that catches its specific gaming path, because any marketing number tied to budget gets optimised toward the measure.
• A vanity metric is any number that can rise while the business does not. Impressions, followers and open rates are vanity until connected to an outcome.

1. The Short Answer: How to Define Marketing-Team KPIs

Define each marketing discipline's KPIs by its economics and its clock. Performance marketing owns marginal ROAS, customer acquisition cost, payback period and incremental conversions. Social and brand marketing own reach against the target audience, branded-search lift, direct-traffic growth and assisted pipeline — measured on demand created, not on last-click conversions. Retention marketing owns repeat-purchase rate, cohort retention, lifetime value and net revenue retention.

The mistake this guide prevents is measuring three disciplines that operate on different timescales with one efficiency metric. Performance marketing produces a readable result within days. Social and brand build demand over months. Retention compounds over quarters. Judge all three on last-click return in a thirty-day window and only performance marketing will ever look good — which is exactly how companies end up over-invested in capture and starved of the demand that feeds it.

This is the measurement layer of the channel and mix work in our [growth engine](/growth-engine) practice, and it depends on the same distinction that organises a whole marketing plan: demand creation is not demand capture, and the two cannot be measured the same way.

• AEO Quick Answer: Performance owns marginal ROAS, CAC and payback; social and brand own reach, branded-search lift and assisted pipeline; retention owns repeat rate, cohort retention and NRR.
• The disciplines run on different clocks — days, months, quarters.
• One efficiency metric across all three makes the two slower disciplines look like failures.

2. Why One Marketing Metric Breaks Two of Three Disciplines

The dominant marketing metric in most companies is last-click return on ad spend, and it is close to the worst possible shared metric because it systematically rewards one discipline and penalises the other two.

Last-click attribution credits the final touch before conversion. That final touch is almost always a capture channel — branded search, retargeting, a discount email — because those are what people click at the moment of buying. The channels that created the demand in the first place, weeks earlier, receive no credit. So a social campaign that built the awareness that led to a branded search gets zero, and the branded search gets everything.

Follow that metric to its conclusion and the outcome is mechanical. Social and brand appear to underperform, so their budget moves to capture. Capture spend rises and looks efficient, because it is harvesting demand the now-defunded creation channels built. Efficiency metrics improve while growth decelerates, and the diagnosis arrives two or three quarters later when the demand pool is exhausted and there is nothing left to capture.

The disciplines also fail for different reasons. Performance fails on saturation and rising auction costs. Social and brand fail on reach, relevance and creative. Retention fails on product, onboarding and lifecycle timing. A single metric that surfaces one is blind to the others.

The fix is a small, honest set of KPIs per discipline, each measuring the outcome that discipline actually owns on the clock it actually runs on.

• Last-click credits the capture touch and ignores the creation that preceded it.
• The failure is mechanical: creation looks weak, budget moves to capture, growth decelerates while efficiency rises.
• Each discipline fails for different reasons, so one metric is blind to two of them.

3. The Organising Distinction: Demand Creation Versus Demand Capture

Every marketing activity either creates demand that did not exist or captures demand that already does, and this single distinction determines how it must be measured.

Capture intercepts existing intent. Search advertising on high-intent terms, retargeting, comparison listings, branded search. It is efficient, measurable with short attribution windows, and bounded — you cannot capture more demand than exists in the market this moment. Performance marketing lives mostly here, and short-window efficiency metrics are appropriate for it.

Creation manufactures intent that was not there. Broad-reach social, video, content, PR, influencer and community work. It is less efficient per impression, operates on lags measured in months, and raises the ceiling that capture operates under. Social and brand marketing live mostly here, and short-window efficiency metrics are actively wrong for it.

Retention is a third category — it does not create or capture new demand, it extends the value of demand already converted. It runs on the longest clock of all, and its metrics are cohort-based and cumulative rather than per-campaign.

Classify every activity into one of the three before choosing its metric. Measuring a creation activity on capture metrics is the category error that ends with the budget cut; measuring a retention activity on acquisition metrics misses where most lifetime value is made. Getting the classification right is most of getting the measurement right, and it is the backbone of the [marketing mix plan guide](/guides/marketing-mix-plan-guide).

• Capture: existing intent, efficient, bounded, short-window metrics fine — performance marketing.
• Creation: new intent, lagged, raises the ceiling — social and brand; short-window metrics wrong.
• Retention: extends converted demand, longest clock, cohort-based metrics.
• Classify the activity first; it decides the metric.

4. Performance Marketing KPIs: Measuring at the Margin

Performance marketing — paid search, paid social direct-response, shopping, programmatic acquisition — is the most measurable discipline, which is both its strength and the source of its most common measurement error: optimising on average return rather than marginal return.

Marginal ROAS. What the next unit of spend returns, not the blended average across all spend. This is the metric that should drive budget decisions, because average return describes committed spend while marginal return describes the decision in front of you. A channel with a strong average and a collapsing marginal is one you should stop scaling, and average [ROAS](/glossary/roas) alone will not tell you that. The full method is in the [channel budget allocation guide](/guides/channel-budget-allocation-guide).

Customer acquisition cost, by channel and segment. The [CAC](/glossary/cac) for genuinely new customers, kept distinct from cost per lead or cost per purchase, because a blended CAC that mixes new and returning buyers flatters the number and hides the true cost of growth.

Payback period. How long until an acquired customer's contribution repays their acquisition cost. This governs how aggressively you can spend independently of eventual return, because payback is a cash constraint — you can estimate yours with the [CAC payback calculator](/tools/cac-payback-calculator).

Incremental conversions. The conversions that would not have happened without the spend, established by holdout or geo testing rather than by platform-reported numbers. Branded search is the standard cautionary case: it reports superb returns because the buyers were already looking for you, and only a holdout reveals how much of it is genuinely incremental.

Contribution-margin ROAS, not revenue ROAS. A 4x revenue ROAS on a product with a 20% margin is losing money; measuring on contribution after cost of goods, shipping and returns is the honest version.

What not to measure performance marketing on alone: impressions, clicks, click-through rate, or average ROAS. Impressions and clicks are inputs, CTR is a creative diagnostic, and average ROAS drives the over-scaling that saturates channels. These are useful diagnostics and dangerous primary KPIs.

• Marginal ROAS — drives budget decisions; average return does not.
• CAC for genuinely new customers, kept separate from cost per lead.
• Payback period — the cash constraint on how fast you can spend.
• Incremental conversions via holdout, not platform-reported numbers.
• Contribution-margin ROAS, not revenue ROAS — a high revenue ROAS can still lose money.

5. Social and Brand Marketing KPIs: Measuring Demand Created

Social and brand marketing is the discipline most often measured wrongly, because its output — demand that shows up later, elsewhere, attributed to another channel — is exactly what last-click cannot see. Its KPIs must measure demand created, on the lag it actually operates on.

Reach against the target audience, not raw reach. The number of genuine target-audience people reached at a viable frequency. Raw impressions are a vanity number; reach filtered to the people who could actually buy is the real input to demand creation. Frequency matters — reaching the right people too few times builds nothing, and too many wastes budget on fatigue.

Branded-search lift. The rise in searches for your brand and product names following creation activity. This is one of the cleanest available signals that demand is being manufactured, because a person searching your brand by name did not previously know to, and something taught them. Rising branded search is creation working.

Direct-traffic and dark-social growth. Increases in people arriving directly or through untracked shares, which is where much word-of-mouth and social-created demand actually lands, invisible to attribution.

Assisted pipeline and assisted conversions. The deals and purchases that a creation channel touched earlier in the journey, even when it was not the last click. This requires multi-touch measurement and is imperfect, but it is far closer to the truth than the zero that last-click assigns.

Share of voice and share of search — your presence in the category relative to competitors, which is a leading indicator of future market share.

Engagement rate as a creative diagnostic, not an outcome. Engagement tells you whether the creative resonates; it is not itself demand. Optimising for engagement produces engaging content that may sell nothing.

The honest way to prove social and brand work is incrementality testing — geo holdouts or matched-market tests that measure what happened to the whole funnel when the spend was present versus absent. This is the only method that captures lagged, cross-channel demand creation, and it is why creation budgets should carry a ring-fenced testing allocation. What not to measure on: follower count, likes and impressions in isolation — the definitional vanity metrics.

• Reach against the target audience at viable frequency — not raw impressions.
• Branded-search lift — one of the cleanest signals that demand is being created.
• Direct-traffic and dark-social growth — where word-of-mouth demand lands, unattributed.
• Assisted pipeline — imperfect, but far closer than the zero last-click assigns.
• Prove it with incrementality testing; never with follower count or likes.

6. Retention Marketing KPIs: Measuring Extended Value

Retention and lifecycle marketing — email, SMS, loyalty, onboarding, winback — governs the majority of lifetime value in any business with repeat purchase, and it runs on the longest clock, so its metrics are cohort-based and cumulative rather than per-campaign.

Repeat-purchase rate, and specifically first-to-second purchase rate. The proportion of customers who buy again, with the first-to-second transition tracked separately because it is the strongest early predictor of cohort value. A customer who buys a second time is dramatically more likely to become high-value, so moving this one transition moves lifetime value more than almost any other retention lever.

Cohort retention curves. The proportion of each acquisition cohort still active or purchasing over time, plotted as a curve rather than reduced to a single number. The shape reveals whether retention is improving across cohorts — the only honest way to know if lifecycle work is landing, because a blended retention number mixes cohorts of different ages and hides the trend.

Customer lifetime value, and the [LTV](/glossary/ltv) to CAC relationship, computed from actual cohort behaviour rather than extrapolated optimistically from a few months of a young cohort.

Net revenue retention for subscription models, and its components — churn, contraction and expansion — separated, because a healthy [NRR](/glossary/nrr) can hide serious logo churn beneath strong expansion in a few accounts.

Winback rate and reactivation — the proportion of lapsed customers recovered, which is often cheaper than new acquisition and rarely measured.

Revenue per email or per message sent, as the efficiency measure for lifecycle channels, in place of the vanity metrics those channels default to.

What not to measure retention marketing on: open rate and click rate in isolation. These are diagnostics for deliverability and subject lines; they are not retention. An email program optimised for opens produces clickbait subject lines and no incremental revenue. This discipline is central to our [retention](/glossary/retention) and lifecycle work.

• First-to-second purchase rate — the strongest early predictor of cohort value.
• Cohort retention curves — the shape reveals whether lifecycle work is landing.
• LTV and LTV:CAC from real cohort behaviour, not optimistic extrapolation.
• NRR with churn, contraction and expansion separated.
• Never: open rate and click rate as the outcome — they are deliverability diagnostics.

7. The Attribution Trap and How to Escape It

Attribution — assigning credit for a conversion to the touches that preceded it — is the single largest source of measurement error in marketing, because every attribution model is biased and the bias runs the same direction: toward whatever is easy to track and close to the conversion.

Last-click over-credits capture. First-click over-credits whatever started the journey and ignores everything that closed it. Linear and time-decay models spread credit by rules that have no relationship to actual causation. Data-driven and algorithmic models are better but still see only tracked touches, which excludes most brand, social and offline influence. There is no attribution model that is simply correct.

The practical consequence: attribution is a reasonable tool for optimising within a channel — which ad, which audience, which creative — and an unreliable tool for deciding between channels, especially between capture and creation. Using last-click to decide the split between performance and brand is how brand loses every argument it should win.

The escape is incrementality testing. Instead of asking which touch to credit, ask what would have happened without this spend. Geo holdouts, matched-market tests and platform-level lift studies answer that directly, and they are the only evidence that captures lagged, cross-channel, untracked demand creation. They cost design effort and a genuine holdout that someone will be uncomfortable turning off, which is the real obstacle rather than the method.

The working pattern for a mature marketing org: attribution for within-channel optimisation, incrementality testing for between-channel decisions, and marketing mix modelling where data volume supports it. State explicitly which decisions rest on which evidence, because deciding budget splits on last-click is the default path to a capture-heavy portfolio that looks efficient while it stops growing. We treat the limits of attribution in more depth in [first-click versus last-click attribution](/resource/blogs/first-click-vs-last-click-attribution) and in our [attribution modelling](/solutions/attribution-modeling) work.

• Every attribution model is biased toward what is easy to track and close to conversion.
• Use attribution to optimise within a channel; never to decide between capture and creation.
• Incrementality testing asks the right question: what would have happened without this spend.
• State which decisions rest on attribution, which on incrementality, which on mix modelling.

8. Vanity Metrics Versus Actionable Metrics

A vanity metric is any number that can rise while the business does not. It is not that these numbers are meaningless — it is that they are inputs or diagnostics being mistaken for outcomes, and reporting them as outcomes creates the illusion of progress.

Impressions can rise with no effect on demand. Followers can grow with no effect on revenue. Email opens can climb with better subject lines and no additional purchases. Clicks can increase with cheaper, lower-intent traffic. Each is useful in its place and misleading as a headline.

The test for whether a metric is actionable: can it rise while the business outcome falls? If yes, it is a diagnostic, not an outcome, and it must be paired with the outcome it is supposed to serve. Impressions paired with branded-search lift. Opens paired with revenue per send. Clicks paired with conversion and CAC.

The reason this matters beyond tidiness is incentives. A team reporting a vanity metric will optimise it, because that is what it is measured on, and optimising a vanity metric actively diverts effort from the outcome. A social team paid on follower growth buys followers; a team measured on branded-search lift builds demand. The metric chooses the behaviour.

Every discipline has its characteristic vanity trap: impressions and CTR for performance, followers and engagement for social, opens for retention. Naming them explicitly, and pairing each with its real outcome, is a short exercise that prevents a long list of expensive misdirections.

• Vanity metric test: can it rise while the business outcome falls?
• If yes, it is a diagnostic — pair it with the outcome it serves.
• Teams optimise what they are measured on, so a vanity KPI diverts real effort.
• Each discipline has a characteristic trap: impressions, followers, opens.

9. Leading and Lagging Metrics by Discipline

As with any measurement system, each discipline needs its lagging outcomes paired with the leading indicators that predict them, so you can manage forward rather than only report backward.

For performance marketing, the lagging outcome is CAC and payback; the leading indicators are marginal ROAS, auction competitiveness and creative fatigue signals (rising frequency with falling click-through). When marginal return starts bending, CAC will rise next — the leading indicator gives you the warning.

For social and brand, the lagging outcome is market share and eventually captured demand; the leading indicators are reach against target, branded-search lift and share of search. These move months before the demand they create shows up in the capture channels, which is precisely why they must be watched — the outcome is too far downstream to steer on.

For retention, the lagging outcome is lifetime value and net revenue retention; the leading indicators are first-to-second purchase rate, adoption or engagement depth, and early cohort retention shape. A cohort's first-month behaviour predicts its lifetime value well enough to act on long before the lifetime is complete.

The cross-discipline point: performance's leading indicators resolve in days, social's in months, retention's over the first cohort weeks predicting quarters. A dashboard that refreshes all of them monthly is too slow for performance and too fast to conclude anything about brand. Match the review cadence to each discipline's clock.

• Performance: marginal ROAS and fatigue signals lead CAC and payback.
• Social and brand: reach, branded-search lift and share of search lead market share.
• Retention: first-to-second purchase and early cohort shape lead lifetime value.
• Match review cadence to each discipline's clock — days, months, quarters.

10. Defining Marketing KPIs Precisely

A marketing metric name hides more disagreement than almost any other, because the same word means different things to different tools and teams. 'Conversion rate' over what denominator — sessions, users, or qualified visitors? 'CAC' including or excluding the retention team's cost, brand spend, and salaries? 'ROAS' on revenue or contribution? Each choice changes the number by a large factor.

For every marketing KPI, pin down the numerator, the denominator, the attribution window and basis, and the cost inclusions. A CAC that includes only media spend is a different, smaller number than one that includes creative, tools and team cost, and comparing the two across periods or channels is meaningless.

The attribution basis is the marketing-specific definitional landmine. State the model and the window explicitly — 'last non-direct click within a 30-day window' — because a metric computed on one basis and compared to a metric computed on another produces confident nonsense, and marketing tools each default to a different basis silently.

Cohort versus period is as consequential here as in sales. A blended monthly retention rate mixes cohorts of every age; a cohort retention curve follows one acquisition group through time. The blended number moves whenever acquisition volume changes even if retention is constant, which is a trap specific to fast-growing businesses.

Agree these definitions across the disciplines and treat a change as a versioned decision. When performance, social and retention each compute CAC differently, the marketing leader cannot compare them, and every budget conversation restarts from an argument about the numbers.

• Pin numerator, denominator, attribution basis and cost inclusions for every metric.
• State the attribution model and window explicitly — tools default silently and differently.
• Cohort beats blended, especially in fast-growing businesses where volume distorts period rates.
• Version definition changes so budget conversations do not restart as arguments about numbers.

11. The Gaming Traps in Marketing Measurement

Every marketing metric tied to budget or evaluation creates pressure to satisfy the measure. The defence is the same as everywhere: pair each outcome with a guardrail that catches its specific gaming path.

The cheap-traffic trap. A team measured on ROAS or CPL can improve the number by buying cheaper, lower-intent traffic that converts worse downstream. Guardrail: measure CAC for genuinely qualified customers and track downstream conversion and retention of acquired cohorts, not just the front-end ratio.

The branded-search trap. A performance team can inflate reported ROAS by shifting budget to branded search, which harvests demand it did not create. Guardrail: separate branded from non-branded reporting and run a branded-search holdout to size the genuinely incremental portion.

The engagement-bait trap. A social team measured on engagement produces provocative content that engages and sells nothing. Guardrail: treat engagement as a creative diagnostic and measure the discipline on reach against target and branded-search lift.

The discount-retention trap. A retention team measured on repeat-purchase rate can buy repeats with margin-destroying discounts. Guardrail: measure retention alongside contribution margin per retained customer, so a repeat bought at a loss is visible.

The attribution-shopping trap. Any team can choose the attribution model that flatters its channel. Guardrail: a single agreed attribution basis for shared reporting, and incrementality tests for the decisions that matter, so no team can pick its own scoreboard.

The general rule holds across all three disciplines: never evaluate on a single number, always pair the outcome with a guardrail that makes the gaming path visible, and reserve the important between-channel decisions for incrementality evidence rather than attribution.

• Cheap-traffic trap — guardrail with qualified CAC and downstream cohort behaviour.
• Branded-search trap — separate branded reporting and run a holdout.
• Engagement-bait trap — measure reach and branded-search lift, not engagement.
• Discount-retention trap — pair repeat rate with contribution margin per retained customer.
• Attribution-shopping trap — one agreed basis plus incrementality for real decisions.

12. Building the Marketing Metric System

Implementing this is a few weeks of work that front-loads the classification and definition steps most teams skip straight past.

Week one: classify every marketing activity as creation, capture or retention. This single exercise resolves most measurement disputes before they start, because it establishes which metrics are even appropriate for each activity.

Week two: for each discipline, define the primary owned outcome and two or three leading indicators, each with exact numerator, denominator, attribution basis and cost inclusions. Get the definitions agreed across disciplines and written in one place.

Week three: design the incrementality-testing plan. Decide which channels get a holdout, how often, and what the ring-fenced test budget is. This is the step that lets creation and retention prove their value, and it is the one most likely to be skipped because a genuine holdout feels like turning off working spend.

Week four: instrument. Connect the platforms, the analytics layer and, where volume justifies it, a warehouse — because most cross-channel marketing questions cannot be answered inside a single ad platform's reporting. Build the discipline dashboards on their respective clocks.

Ongoing: review performance metrics weekly, social and brand monthly, retention by cohort as cohorts mature, and revalidate that leading indicators still predict their lagging outcomes. Retire vanity metrics from reporting entirely, keeping them only as diagnostics where they earn it.

• Week 1: classify every activity as creation, capture or retention.
• Week 2: define primary and leading metrics with exact bases, agreed across disciplines.
• Week 3: design the incrementality-testing plan and ring-fence the test budget.
• Week 4: instrument across platforms and a warehouse; build per-discipline dashboards.

13. A Worked Example (Illustrative Model)

The figures below are an illustrative model to show how the layers interact. They are not client data.

A consumer brand sees blended ROAS improving quarter over quarter — from 3.1x to 3.8x — while total revenue growth slows from 40% to 12%. On the headline metric everything looks better; the business is decelerating. The layered metrics explain why.

Performance marketing: marginal ROAS on the scaled channels had fallen to near 1.0 even as average ROAS rose, because budget had shifted toward branded search and retargeting — capture channels harvesting existing demand. The rising average was an artefact of the mix moving toward cheap capture, not of improving efficiency.

Social and brand: reach against target had fallen 30% after the budget moved to performance, and branded-search volume — the demand social had been creating — was declining with a lag. The engine feeding the capture channels was being starved.

Retention: first-to-second purchase rate was flat, so the base was neither helping nor hurting; the problem was purely at the top.

The diagnosis: the company had optimised its way into a capture-heavy portfolio that looked efficient while the demand-creation engine that fed it was defunded. Revenue growth slowed because there was progressively less new demand to capture. A team watching only blended ROAS would have concluded things were improving and cut brand further, accelerating the decline. The layered metrics point at exactly the opposite action — restore demand creation — and the branded-search leading indicator had been signalling the slowdown a quarter before it reached revenue.

14. Putting It Together

Defining marketing KPIs well begins with one act of classification — creation, capture or retention — because that decides which metrics are even valid for each discipline. Get that wrong and no amount of dashboard sophistication recovers it; get it right and the rest follows.

Performance marketing is measured at the margin, on contribution, against incrementality. Social and brand are measured on demand created — reach, branded-search lift, assisted pipeline — never on last-click. Retention is measured by cohort, on repeat rate, retention curves and net revenue retention.

The thread through all of it is escaping last-click attribution for the decisions that matter. Last-click is a fine optimiser within a channel and a destroyer of brand and retention budgets when used to allocate between channels. Incrementality testing is the discipline that lets the slower, harder-to-track disciplines prove what they are worth.

If you want the marketing metric layer built against your actual channel mix and the incrementality tests designed to prove your creation spend, that is where our [growth engine](/solutions/roas-optimization) engagements start — and the planning layer above it is the [marketing mix plan guide](/guides/marketing-mix-plan-guide).