Most channel reports answer a limited question: what happened this week? Spend, clicks, leads and first purchases are useful operating signals, but they rarely reveal whether a channel is acquiring customers who stay, expand and repay the cost of acquisition.
Marketing cohort analysis solves that problem by grouping customers who started in the same period, usually through the same acquisition channel, and following their behaviour over time. It turns a channel report from a record of immediate activity into an assessment of customer quality.
This matters whenever the buying decision is separated from the business outcome. A paid search campaign may look expensive in month one but attract customers with excellent renewal behaviour. An organic landing page may generate low-cost conversions but attract users who never reach activation. Neither conclusion is visible in a last-click dashboard alone.
The practical goal is not to create an elaborate data project. It is to build a repeatable view that helps a commercial team answer four questions: which channels retain customers, which produce durable revenue, how quickly does each channel repay acquisition cost, and where should the next unit of budget go?
Start with one defined business outcome
Before selecting a cohort period or building a spreadsheet, agree on the unit of value being measured. This is where many analyses lose credibility: marketing, sales and finance use similar terms to mean different things.
For this framework, use a subscription software business with one consistent worked example. The desired outcome is a new paying account on a monthly plan. A lead is someone who submits a form or starts a trial. An opportunity is a sales-qualified account with a recorded potential contract value. Neither is revenue. Revenue begins only when an account pays an invoice.
We will measure recognised revenue: the subscription revenue earned in a given month after payment, excluding taxes, refunds and one-off pass-through charges. This is more reliable than pipeline or opportunity value for cohort economics because it represents money actually earned. If your sales cycle is long, you can run a leading-indicator cohort for opportunities separately, but do not blend it with revenue cohorts.
In the example, the business spends £12,000 on paid search in January and acquires 60 new paying accounts. Its initial customer acquisition cost (CAC) is £200 per account. Those 60 accounts form the January paid-search cohort. Every future month, the business examines the same original 60 accounts: how many remain active, how much recognised revenue they generate, and whether their accumulated gross profit has repaid the £12,000 spend.
This discipline connects cleanly to a broader marketing KPI tree. Channel actions should ultimately support a business outcome, not merely improve a platform metric.
Choose a cohort definition that survives scrutiny
A cohort needs two stable dimensions: when a customer entered and how they were acquired. Monthly cohorts are usually the sensible default for B2B, higher-consideration services and products with modest customer volumes. Weekly cohorts work for high-volume ecommerce or consumer subscriptions. Quarterly cohorts often hide too much change to guide optimisation.
Use the date of first paid invoice, first order or first verified transaction as the acquisition date. Avoid grouping customers by lead date if sales conversion can take weeks or months. Doing so makes retention appear longer or shorter depending on pipeline speed rather than real customer behaviour.
For channel, store both first-touch and attributable acquisition fields where possible. First touch describes how the customer was initially introduced to the business. Attributable acquisition records the rule used for investment reporting, such as last non-direct touch or a defined multi-touch model. Select one as the primary cohort dimension and label it plainly.
My practical preference is to start with a deterministic acquisition field: paid search, paid social, organic search, referral, partner, email, direct or other. Do not begin with dozens of campaign-level cohorts. Small cells create noisy results, and frequent platform naming changes make them hard to maintain. Drill into campaign or landing-page cohorts only after a channel-level pattern is evident.
Build a minimal customer-level data table
A cohort table is only as trustworthy as its underlying grain. Maintain one row per customer or account, not one row per web session, invoice line or CRM contact. Where multiple people belong to one company, use the commercial account ID and retain a clear mapping to individual contacts.
| Field | Purpose |
|---|---|
| Customer or account ID | Joins billing, CRM and analytics records without double-counting. |
| First paid date | Assigns the acquisition month or week. |
| Primary acquisition channel | Supports the channel comparison. |
| Initial acquisition cost | Allocates media, agency and relevant onboarding costs consistently. |
| Monthly recognised revenue | Measures customer value over time. |
| Monthly gross profit | Supports payback after direct delivery costs. |
| Active status and cancellation date | Calculates retention and identifies churn timing. |
Attribution data will never be perfect. A prospect may discover a brand through organic search, return through a retargeting ad and convert after a branded query. The solution is not to pretend the answer is exact. Document the chosen rule, keep it unchanged for trend comparison and investigate material shifts in channel mix. A sound marketing measurement taxonomy makes that discipline much easier.
Measure retention before lifetime value
Retention is the clearest early test of channel quality. For each cohort month, divide active customers at the end of a later month by customers at the beginning of the cohort.
Month-three logo retention = active accounts from the original cohort at the end of month three ÷ accounts acquired in month zero.
If 48 of the 60 January paid-search accounts are active at the end of month three, month-three logo retention is 80%. Build this as a matrix: acquisition month down the side, months since acquisition across the top. The January cohort occupies one row; February is the next; the columns are month zero, month one, month two and so on.
For subscription businesses, track both logo retention and revenue retention. Logo retention shows whether accounts remain. Gross revenue retention shows how much starting subscription revenue remains after cancellations and downgrades, excluding expansion. Net revenue retention includes expansion revenue from the existing cohort. Define these once in your dashboard and do not use them interchangeably.
For ecommerce, replace active-account retention with repeat purchase rate or repeat revenue at defined intervals. For lead-generation businesses, a useful equivalent may be client renewal, repeat project purchase or retained contract value. The principle remains constant: compare the later behaviour of the original customers, not all current customers in the database.
Calculate revenue quality, payback and lifetime value
Retention tells you whether customers stay. Revenue quality explains what staying is worth. Begin with cumulative recognised revenue per acquired customer:
Cumulative revenue per customer at month n = total recognised revenue from the original cohort through month n ÷ original customers acquired.
In the January paid-search example, assume the cohort produces £6,000 of recognised revenue in month zero, £5,600 in month one, £5,100 in month two and £4,800 in month three. Cumulative revenue by month three is £21,500, or £358 per original acquired account. This calculation includes churned accounts in the denominator; that is intentional. It shows the value generated by each account the channel initially acquired.
Revenue is not profit, so payback should normally use gross profit. If the business retains 75% of subscription revenue after hosting, support and other direct service costs, the cohort has generated £16,125 in cumulative gross profit by month three. It has passed the £12,000 acquisition cost, so payback occurs during month three.
Payback period = first month in which cumulative cohort gross profit equals or exceeds total acquisition cost.
Include the costs you can apply consistently. Paid media is obvious. Agency fees, sales commission and onboarding labour may also be appropriate when they are incremental to acquisition. Do not load broad company overhead into one channel and omit it from another. For comparison, consistency is more valuable than false precision.
Lifetime value (LTV) is an estimate of the gross profit a customer will generate over its relationship with the business. Early cohorts do not yet have a full lifetime, so present observed LTV separately from forecast LTV. Observed LTV is the actual cumulative gross profit to date. Forecast LTV extends it using a documented retention and margin assumption. In my view, observed cohort value should carry more weight in budget discussions until enough maturity exists to support a forecast.
Compare channels using a practical scorecard
A channel should not win simply because it has the lowest CAC or the highest short-term return. Compare cohorts at the same maturity point. It is unfair to compare a January cohort at month six with a June cohort at month one.
| Metric at month three | Why it matters | Interpretation |
|---|---|---|
| CAC | Cost to acquire one new paying account | Low is useful, but not decisive alone. |
| Logo retention | Whether acquired accounts remain active | Weak retention can signal poor targeting or onboarding. |
| Cumulative revenue per customer | Realised customer value so far | Reveals quality beyond the first sale. |
| Cumulative gross profit | Contribution available to recover acquisition cost | Determines payback. |
| Payback status | Speed of capital recovery | Matters most when cash is constrained. |
Suppose organic search acquires fewer accounts than paid search but achieves higher month-three retention and pays back in two months. That does not automatically mean moving all budget to SEO. Organic growth can be constrained by demand, technical capacity and content production time. Paid search may still be scalable and strategically useful. The better decision is to identify why the quality differs: query intent, landing-page promise, pricing fit, sales follow-up or post-purchase activation.
For search-led programmes, separate branded and non-branded acquisition where volume allows. Branded search frequently captures demand created elsewhere, while non-branded search is more informative about new-demand capture. Guidance from Google Search Central and Bing Webmaster Tools can help teams understand search visibility and technical discovery, but cohort performance still requires customer and billing data.
Turn cohort findings into operating decisions
Refresh a monthly cohort view after billing closes, then hold a short review with marketing, sales, customer success and finance. Look for changes that are both commercially meaningful and repeatable across more than one cohort. One small cohort can be distorted by a large account, a seasonal promotion or a reporting error.
- High CAC, strong retention: test efficiency improvements before cutting spend. The offer may be attracting valuable customers.
- Low CAC, weak retention: inspect audience targeting, qualification, expectation setting and activation rather than celebrating cheap acquisition.
- Good retention, slow payback: examine pricing, gross margin, upfront discounting and onboarding cost.
- Fast payback, declining later retention: protect growth by investigating churn before scaling the same acquisition promise.
Pair cohort analysis with controlled tests where possible. Attribution describes how credited customers behave; it does not prove the channel caused every conversion. For major budget reallocations, use the principles in a marketing incrementality testing roadmap to test whether spend creates net-new outcomes.
FAQ and conclusion
How many customers does a cohort need?
There is no universal threshold. Small cohorts can still inform qualitative investigation, but avoid declaring a channel winner from a handful of accounts. Combine adjacent months only when channel strategy, offer and tracking remained comparable.
Should cohort analysis use first-touch or last-touch attribution?
Use the model that matches the decision. First touch is useful for evaluating demand creation; last non-direct touch can support conversion-budget management. Keep the model visible, and compare like with like over time.
What is the best cohort interval?
Monthly is the practical starting point for most B2B teams. Move to weekly only when volume is high enough and purchasing behaviour changes quickly.
What should agencies report to clients?
Report acquisition volume alongside retention, cumulative revenue and payback at matched cohort ages. Be explicit about data gaps, attribution rules and whether revenue is observed or forecast.
Conclusion: A useful marketing cohort analysis system does not need perfect attribution or advanced software on day one. It needs a stable customer ID, a defensible acquisition rule, clean revenue data and a regular review cadence. Start with monthly channel cohorts, compare them at equal maturity, and let retained gross profit—not cheap clicks or flattering dashboards—guide investment decisions.