Cohort Attribution Analysis: Retention by Acquisition Channel
A direct cohort retains at 2.3x the lifetime value of a paid one. Why cohorting by acquisition channel and month beats a single blended retention curve.
Muzahid Maruf, Founder · TrackRev.io & Contant.io
On this page
Explore with AI
Opens this article inside the chosen assistant with a ready-made prompt.
Blended retention hides the truth: a cohort acquired through direct retains at more than double the lifetime value of one acquired through paid — 2.3x against 0.8x — so a single company-wide retention curve averages your best and worst channels into a line that describes neither (TrackRev platform data, Q2 2026).
Cohort attribution analysis fixes that by slicing customers two ways at once — the channel that acquired them and the month they arrived — and tracking each slice’s retained revenue separately.
A cohort is a group of customers who share an acquisition channel and a start month; cohort attribution analysis measures how each group’s retained revenue behaves over the months that follow, so you can see which channels produce customers who stay. This guide explains why blended numbers mislead, how to build a channel cohort, and how to read the curves — so you can see which acquisition channels actually pay your MRR over the long run, not just at signup.
Key Takeaways
- A blended retention curve is the average of channels that behave nothing alike, so it describes a customer who does not exist and hides which channel is unprofitable.
- Cohort attribution analysis slices customers by acquisition channel and start month at once, which is the only way to ask whether a specific channel is getting better or worse at producing customers who stay.
- Channel lifetime-value multipliers already encode cohort behaviour: direct at 2.3x is flat and long-lived, while paid at 0.8x is front-loaded with an early-month churn cliff.
- Blended curves are vulnerable to survivorship and Simpson’s paradox, where overall retention can appear to rise in a month when every individual channel’s retention fell.
- Use blended retention for board headlines or when data is too thin to cohort; switch to channel cohorts once you have the volume and a real budget decision to make.
The one-line version
A single blended retention curve is the average of channels that behave nothing alike. Cohort attribution analysis splits retention by acquisition channel and start month, so a paid cohort’s month-two cliff and a direct cohort’s flat, long-lived curve stop cancelling each other out into a number that guides no decision.
Why this matters for your revenue
Budget decisions made on a blended retention curve are decisions made on an artefact.
If half your customers come from a channel that churns fast and half from one that never leaves, the blended curve sits in the middle — describing a customer who does not exist and hiding the fact that one of those channels is unprofitable.
You cannot fix a retention problem you have averaged away.
The money at stake is the gap between the channels.
At a 2.3x multiplier a direct cohort funds growth on its own; at 0.8x a paid cohort can churn before its acquisition cost is recovered (TrackRev platform data, Q2 2026, at /data/saas-attribution-benchmarks).
Cohorting by channel makes that gap visible and therefore actionable — you can cut the cliff-shaped cohort, double down on the flat one, and stop treating a company-wide retention number as if it were a strategy.
It is the retention-curve companion to our channel LTV guide.
Why blended numbers mislead
A single retention or LTV number for the whole business is comforting and almost always wrong in a way that matters.
The average of two channels describes neither
If a direct cohort retains most of its revenue at month twelve and a paid cohort has lost most of its by month three, the blended curve lands somewhere between — a shape neither cohort actually follows.
Decisions made on that middle line are calibrated to a fiction: you would neither cut the paid cohort you should nor protect the direct cohort you must. The average is not a summary; it is a distortion.
Survivorship in a single curve
A blended curve also flatters itself over time through survivorship: as the fast-churning customers drop out early, the remaining base is increasingly the loyal channels, so the curve appears to stabilise.
That late flattening is not your product getting stickier — it is the composition of the survivors shifting towards the channels that were always going to stay.
Cohorting by channel removes the illusion by never mixing the populations in the first place.
Simpson’s paradox in channel data
Channel data is a classic setting for Simpson’s paradox, where a trend that holds within every channel reverses when the channels are pooled.
Overall retention can appear to rise in a month when every individual channel’s retention fell, simply because the mix shifted towards higher-retaining channels.
If you only ever look at the blended number, you can congratulate yourself on an improvement that did not happen in any cohort you actually run.
Building a channel cohort
A useful cohort is defined on two axes at once. Miss either and the analysis loses its power.
Slice by acquisition channel
The first axis is the channel that sourced the customer — direct, organic search, newsletter, affiliate, paid.
This requires durable attribution: the customer record has to remember which channel acquired it for the whole life of the subscription, not just the first session.
Without a reliable sourcing channel on every customer, you cannot build a channel cohort at all, which is why first-party click capture underpins the whole analysis.
Slice by start month
The second axis is the month the customer started paying.
Start-month cohorts let you read retention as a curve — month one, month two, month three after acquisition — and compare cohorts acquired at different times to see whether recent cohorts retain better or worse than older ones.
A channel improving or degrading over time only shows up when you hold the start month constant.
Why you need both axes
Channel alone tells you which sources retain; start month alone tells you whether retention is trending. Only both together let you ask the decisive question: is this specific channel getting better or worse at producing customers who stay?
A paid channel whose recent cohorts churn faster than last quarter’s is a signal to act now — and it is invisible unless you cross the two axes.
Retention shape by channel
The published lifetime-value multipliers already encode how each channel’s cohort tends to behave, from flat-and-loyal to front-loaded-and-leaky.
| Channel | LTV multiplier | Typical cohort behaviour |
|---|---|---|
| Direct | 2.3x | Flat, long-lived — retains and expands |
| Organic search | 2.1x | Slow to convert, retains strongly |
| Newsletter | 1.9x | Steady, gradual attrition |
| Affiliate | 1.4x | Moderate, coupon-sensitive at the edges |
| Paid (avg) | 0.8x | Front-loaded — early-month churn cliff |
Lifetime-value multipliers from TrackRev platform data, Q2 2026 (4,217 workspaces). Cohort-behaviour descriptions are directional interpretations of those multipliers, not separately published retention curves. See /data/saas-attribution-benchmarks.
Blended vs cohorted, concretely
Suppose a month’s new customers split evenly between direct and paid. Blended, they might show a moderate retention curve that looks acceptable. Cohorted, the direct half at a 2.3x multiplier stays and expands while the paid half at 0.8x falls off a cliff by month three (TrackRev platform data, Q2 2026). The blended curve told you retention was fine; the cohorts tell you half your acquisition is unprofitable and the other half is carrying it.
Reading retention curves per channel
Once cohorts are built, the shape of each curve tells you what to do with the channel.
Flat curves vs cliff curves
A flat curve — revenue that holds steady month after month — marks a channel whose customers found real, lasting value; fund it.
A cliff curve — a sharp drop in the first few months — marks a channel acquiring customers who never activated or converted on a promotion they did not need.
The shape, not the signup count, is the verdict: a channel can post strong acquisition and still be a cliff.
When paid’s cliff shows up
Paid cohorts characteristically lose revenue early, because paid clicks skew towards lower-intent and promotion-driven buyers.
If your paid cohort’s curve drops steeply before the acquisition cost is recovered, the channel is unprofitable no matter how good its cost per signup looked.
Reading paid under a cohort curve rather than a signup total is how that unprofitability becomes undeniable. Our attribution models comparison covers how the sourcing credit is assigned in the first place.
The month-3 checkpoint
Month three is a useful universal checkpoint because it is far enough out to have exposed early churn but soon enough to act on.
Comparing each channel’s retained revenue at month three against its acquisition cost gives you a fast, honest read on which channels are paying for themselves and which are being carried.
If a channel has not recovered its cost by then and its curve is still falling, it is a candidate to cut.
Blended vs cohort analysis
The contrast between the two views is the contrast between a comforting number and an actionable one.
| Dimension | Blended retention | Channel-cohort retention |
|---|---|---|
| What it shows | One curve for the whole business | A curve per channel and start month |
| Hides channel differences | Yes | No |
| Reveals which channel to cut | No | Yes |
| Exposes the month of churn | No | Yes |
| Vulnerable to Simpson’s paradox | Yes | No |
Conceptual comparison of blended versus cohorted retention analysis. TrackRev cohort views as published at /products/channel-analytics.
When blended retention is fine
The blended number is not useless — it is just the wrong tool for a channel decision.
Board-level headline metrics
For a board slide or an investor update, a single net-revenue-retention figure is the right altitude — it summarises the business without drowning the audience in channel curves.
Blended retention communicates trajectory; it just should not be the number you use internally to decide where the next marketing pound goes.
Too little data to cohort
Early on, splitting a small customer base by both channel and start month leaves cohorts too tiny to read — a five-customer cohort tells you nothing reliable.
Until you have enough volume for each channel-month slice to be meaningful, a blended view is the honest one, and you graduate to cohorts as the numbers grow. Forcing cohort analysis on thin data manufactures noise, not insight.
When to cohort by channel
Once you have the volume and channels that plausibly differ, blended retention actively hides the decisions you most need to make.
Parity first, then the shared model
TrackRev attributes the sourcing channel on every customer with the same first-party capture any attribution tool uses, then — because it reads recurring revenue from billing — assembles those customers into channel-and-month cohorts automatically.
Because attribution, link tracking, and the affiliate programme share one data model, the cohort you see for organic is built on the same definition of a sale as the one you see for affiliates.
Parity on sourcing attribution, then the cohort curves on top.
The stack math
Cohort analysis by channel should not require stitching tools together.
Teams often run a link tracker like Bitly Growth (~$35/mo) beside an affiliate tool like Rewardful Starter (~$49/mo) — about $84/mo for two products that each hold half the data a cohort needs and never reconcile.
TrackRev is $39/mo for link tracking, revenue attribution, and affiliates on one shared model, so the sourcing channel and the recurring revenue live together. A free tier at 1,000 events/mo lets you try it; pricing is on the pricing page.
When NOT to use TrackRev
If you have too few customers to split by both channel and month, cohort attribution will produce noise rather than signal, and a blended view is the honest choice until you scale.
TrackRev is also not a full product-analytics suite — it does not do behavioural cohorting on in-app events like feature adoption or session depth; it cohorts revenue by acquisition channel.
It is built for SaaS and subscription teams that want to know which channels produce customers who stay and pay.
Found this useful? Share it.
Frequently asked questions
- Cohort attribution analysis groups customers by the channel that acquired them and the month they started paying, then tracks each group’s retained revenue over the following months. It replaces a single company-wide retention curve with a curve per channel and start month, so you can see which acquisition channels produce customers who stay and whether a channel is improving or degrading over time.
- A blended retention curve averages channels that behave nothing alike — a direct cohort that stays for years and a paid cohort that churns in months land as one middle line that neither actually follows. Decisions made on that average are calibrated to a fiction. Blended curves also flatter themselves through survivorship, appearing to stabilise as fast-churning customers drop out and the loyal channels dominate the remaining base.
- You need durable attribution on every customer and a read of their recurring revenue. Capture the click on your own domain, bind it to the customer so the sourcing channel is remembered for the life of the subscription, then group customers by that channel and by start month. Connect your billing provider so each cohort’s retained revenue can be tracked month by month rather than estimated.
- A retention curve plots how much of a cohort’s revenue remains in each month after acquisition. Flat curves mark channels whose customers found lasting value, such as direct and organic search at 2.1x to 2.3x lifetime-value multipliers. Cliff curves — sharp early drops — mark channels acquiring lower-intent or promotion-driven buyers, characteristic of paid at a 0.8x multiplier. The shape, not the signup count, is the verdict on a channel.
- Simpson’s paradox is when a trend that holds within every group reverses when the groups are pooled. In channel data, overall retention can appear to rise in a month when every individual channel’s retention actually fell, simply because the customer mix shifted towards higher-retaining channels. Looking only at the blended number can make you credit an improvement that did not happen in any cohort you run, which is why cohorting by channel matters.
- Use blended retention for board-level headlines, where a single net-revenue-retention figure is the right altitude, and when you have too few customers to split by both channel and month without cohorts becoming statistically meaningless. A five-customer cohort tells you nothing reliable. Graduate to channel cohorts once you have the volume for each channel-month slice to be readable and a real budget decision depends on the answer.
- By comparing each channel cohort’s retained revenue against its acquisition cost at a checkpoint such as month three. A channel whose cohort curve has dropped steeply and has not recovered its cost by then is unprofitable, however attractive its cost per signup looked at acquisition. Blended retention hides this because the unprofitable cohort’s cliff is averaged against loyal cohorts into a curve that looks acceptable.
Written by
Muzahid Maruf, Founder, TrackRev.io & Contant.io
Muzahid Maruf is the founder of TrackRev.io and Contant.io. He writes about marketing attribution, link tracking, and revenue analytics for SaaS teams.
Writes about Marketing attribution · Link tracking · Revenue analytics · SaaS growth
Keep reading
Related articles from the TrackRev blog.