How do you track the decay rate of Marketing Qualified Leads by cohort week?

The Tuesday morning that made this a real problem

A mid-market SaaS RevOps lead opens the marketing dashboard and sees a comfortable number: 380 MQLs last month, up from 350 the month before. Volume is fine. Then sales leadership pushes back in the pipeline review — the SDR team says the leads are "dead on arrival," and the SDR-to-SQL conversion rate has slid from 18% to 11% over the quarter without anyone being able to name a cause.

Both parties are looking at the same data and reaching opposite conclusions, because both are looking at an aggregate. The monthly MQL count is a snapshot of arrivals. It says nothing about what happened to any particular lead after it arrived. When you blend all leads together into a single monthly bucket, a batch of high-intent demo requests from week 1 and a batch of low-intent ebook downloads from week 4 average into an undifferentiated "380." The average conceals the fact that the week-4 batch stopped opening emails within nine days.

Cohort-week tracking dissolves that ambiguity. You freeze a group — every lead that crossed the MQL threshold between, say, Monday March 3 and Sunday March 9 — and then you follow only that frozen group forward. New leads never enter the March 3 cohort. Leads that go cold never leave it. The denominator is fixed for the life of the measurement, which is the entire methodological point: a fixed denominator lets you see decay as a curve rather than as a static ratio.

What that RevOps lead actually found when they rebuilt the report by cohort week: the two cohorts from the first half of the month were converting normally, holding around 45% still-active at week 4. The two cohorts from the second half were collapsing — under 20% still-active at week 4. The proximate cause turned out to be a paid campaign that launched mid-month and drove volume from a much colder audience segment, plus an SDR who went on leave without their queue getting reassigned for eight business days. Neither of those was visible in the monthly rollup. Both were unmissable in the cohort view.

That is the scenario this whole practice exists for. Not academic curiosity about lead lifecycles, but the concrete moment when marketing and sales disagree about lead quality and neither side has a way to prove it. A decay curve by cohort week gives you a shared artifact both teams can read the same way. It converts an argument about vibes into an argument about a line on a chart, which is a much shorter argument.

How the mechanism actually works

The mechanism has four moving parts: a stamped entry date, a definition of "still active," a fixed observation window, and a repeatable weekly snapshot. Get all four right and the report builds itself. Get any one wrong and you produce a chart that looks authoritative and means nothing.

Part one: the immutable entry stamp. Create a custom date field — call it MQL_Entry_Date — that gets written exactly once, the first time a record crosses your MQL threshold, and is never overwritten. This is the single most common implementation failure. If your automation re-stamps the field every time a lead re-qualifies, every cohort silently reshuffles itself and your historical curves change retroactively. Lock the field with a workflow condition that only fires when the field is null, or use a "first touch" style field type if your platform offers one. From that date, derive a MQL_Cohort_Week value — the ISO week number, or better, the Monday date of that week stored as a date so it sorts correctly and survives year boundaries.

Part two: the activity definition. "Still active" is a decision you make, not a fact you discover, and you must write it down before you build anything. A workable starting definition: any of an email open plus a click, a visit to a page beyond the homepage or blog, a form submission, a meeting held, or an inbound reply. Note the deliberate exclusion of a bare email open — with privacy-protection features on modern mail clients pre-fetching images, raw opens are heavily inflated and will make your decay curve look far healthier than reality. Requiring a click, a visit, or a reply keeps the signal honest.

Part three: the observation window. Twelve weeks is the practical default for most B2B motions. Eight is too short to see the tail on considered purchases; twenty-six produces a chart nobody reads. If your average sales cycle exceeds 90 days, extend to sixteen or twenty weeks — the window should comfortably exceed the cycle so you can distinguish "gone cold" from "still deliberating."

Part four: the weekly snapshot. Because most CRMs make it hard to reconstruct "was this lead active during week 5 of its life" retroactively, the durable approach is to write the answer down each week rather than compute it later. A scheduled job runs every Monday, walks every lead whose cohort is inside the observation window, evaluates the activity definition against the prior seven days, and writes a row to a snapshot table or object: cohort week, weeks-since-entry, lead ID, active yes/no, current lifecycle stage. That snapshot table is your source of truth. Everything downstream — the curve, the source breakdown, the alerting — is just a query against it.

The elegance of the snapshot approach is that it degrades gracefully. If your CRM reporting layer is weak, the snapshot table still exports cleanly to a spreadsheet or a warehouse. If you later move to a proper BI tool, the historical snapshots come with you. And because each row is immutable, you can audit any published number back to the week it was recorded — which matters the first time someone in a QBR challenges a decay figure.

Real numbers, ranges, and benchmarks

Benchmarks in this area deserve heavy caveats, because decay curves are shaped far more by your qualification threshold than by your industry. A team that calls every ebook download an MQL will show catastrophic decay; a team that requires a demo request will show a gentle slope. Neither is "better" — they are measuring different populations. Read the ranges below as a way to sanity-check your own curve's *shape*, not as targets to hit.

Typical shape by intent tier. High-intent sources — demo requests, pricing-page conversions, free-trial signups, inbound contact forms — commonly hold a majority of the cohort active through week 2 and something in the 40–60% band at week 4, tapering into the 15–25% range by week 8. Medium-intent sources like webinar attendance or a bottom-funnel comparison guide typically sit ten to twenty points below that at every checkpoint. Low-intent sources — top-of-funnel content downloads, newsletter signups, list purchases — often fall under 30% active by week 4 and under 10% by week 8. If your "high-intent" cohort is behaving like the low-intent band, your threshold is mislabeled, not your marketing.

The half-life framing. A more portable metric than any single-week percentage is the cohort half-life: the number of weeks until 50% of the original cohort is no longer active. It compresses the whole curve to one number, it is comparable across cohorts of wildly different sizes, and it is intuitive in a leadership meeting. Most B2B cohorts land somewhere between two and six weeks. Track half-life per source and per cohort week, and you have a single trend line that catches most problems.

Cohort size and statistical honesty. Below roughly 30 leads in a cohort, week-over-week percentages swing wildly on single-lead movements — one rep working four accounts moves the number eight points. If your weekly MQL volume is under 30, switch to biweekly or monthly cohorts rather than pretending weekly resolution exists. At 50–100 leads per cohort, weekly is reasonable. Above 200, you can safely segment within the cohort by source or segment and still have readable sub-curves. Always display the raw count next to the percentage; a chart showing "22% active" without the "of 9 leads" alongside it will get someone in trouble.

The velocity checkpoints that matter most. Three points on the curve carry most of the diagnostic weight. Week 1 retention tells you about speed-to-lead and routing — if you are losing a large share of the cohort in the first seven days, the problem is almost never content. Week 4 retention tells you about nurture quality and offer relevance. Week 8 retention tells you whether you have a genuine long-consideration audience or a graveyard. Watching those three numbers per cohort catches more than watching all twelve.

Speed-to-lead as the upstream lever. The relationship between response time and outcomes is one of the better-established patterns in lead management, and it compounds directly into the week-1 number. Whatever your specific figures, the directional finding holds broadly: response within minutes materially outperforms response within hours, and response after a day underperforms sharply. If your week-1 retention is weak, instrument your median time-to-first-touch per cohort before touching anything else — it is usually the cheapest fix available and it moves every downstream week.

Anchoring against conversion, not just activity. Activity decay is a leading indicator; MQL-to-SQL and MQL-to-opportunity conversion are the lagging ones. Track both on the same cohort so you can validate that your activity definition actually predicts revenue. If cohorts with strong week-4 activity do not convert better than cohorts with weak week-4 activity, your activity definition is measuring noise — most often because you left bare email opens in it, or because a nurture email blast is registering as "engagement" for everyone simultaneously. That correlation check, run once a quarter, is what keeps the whole system from drifting into theater.

Trade-offs, alternatives, and where this fits alongside other measurement

Cohort decay tracking is not free, and it is not the only lens. Choosing it means accepting specific costs and declining specific alternatives, so it is worth being explicit about the trade space.

Weekly versus monthly cohorts. Weekly gives you resolution to catch a campaign launch, a routing outage, or a rep's absence within days. It costs you statistical stability at low volume and produces twelve to fifty-two lines on a chart if you are not disciplined about which cohorts you display. Monthly is stable and readable but blurs exactly the within-month variation that caused the problem in the opening scenario. The pragmatic compromise many teams land on: compute weekly, display the last four weekly cohorts as individual lines plus a trailing-quarter average as a reference band.

Activity-based versus stage-based decay. Activity decay measures engagement; stage decay measures whether the lead is progressing through lifecycle stages. Activity is more sensitive and shows problems earlier. Stage progression is closer to revenue and much harder to game. They fail in different directions: activity decay can look healthy while nobody advances (a nurture program generating clicks and no meetings), and stage decay can look healthy while the population quietly narrows (a few big deals carrying the cohort). Run both if you can; if you must pick one, pick the one your organization will actually act on.

Build in the CRM versus build in the warehouse. In-CRM is faster to stand up, keeps the data next to the people who act on it, and requires no data engineering ticket. It hits walls on retroactive computation, cross-object joins, and retention of historical states. Warehouse-based is more durable, joins cleanly to product and billing data, and survives CRM migrations — at the cost of a longer build and a dependency on a data team's queue. A reasonable sequence: prove the concept with a CRM snapshot object and a spreadsheet export for one quarter, then port to the warehouse once the definitions have stopped changing. Porting a stable definition is a week of work; porting a definition that is still in flux is a month of rework.

The adjacent instrumentation this unlocks. Once you have a fixed-cohort snapshot table, several neighboring analyses become nearly free, which materially changes the cost-benefit. Cohort-based lead-source ROI: attach spend to the cohort and you get cost-per-still-active-lead-at-week-4, a far better channel metric than raw cost-per-MQL. Nurture campaign lift: hold out 10% of a cohort from a nurture sequence and compare decay curves — a clean A/B on the same fixed denominator. SDR queue health: the same snapshot pattern applied to owner rather than source will surface a rep whose leads decay faster than peers, weeks before it shows in their pipeline number. And the identical mechanism extends downstream to opportunity-stage aging and customer-health cohorts; the entry-stamp-plus-weekly-snapshot pattern is the reusable piece.

What you give up. Cohort tracking is retrospective by construction. It tells you what happened to a group that has already been acquired; it will not tell you which individual lead to call this afternoon — that is lead scoring's job, and the two are complements, not substitutes. Cohort analysis also struggles with leads that legitimately re-enter the funnel months later; you will need an explicit rule for whether a recycled lead joins a new cohort or stays in its original one. Most teams keep it in the original cohort for decay purposes and track recycles as a separate metric, because moving it corrupts the fixed denominator.

Common pitfalls and how to avoid them

The moving denominator. By far the most damaging error: building the report so that leads drop out of the cohort when they convert, get disqualified, or get merged. The moment the denominator changes, the percentage stops meaning anything and cohorts stop being comparable. Fix it by snapshotting the original cohort membership list once, at the end of the entry week, and treating that list as immutable. Converted leads are still cohort members — they are simply members whose outcome you record separately. A converted lead should never make your decay curve look worse.

Re-stamping the entry date. Discussed above, but it belongs in any pitfall list because it fails silently and retroactively. Symptom: last month's chart no longer matches the screenshot in last month's deck. Diagnosis: pull ten records and compare MQL_Entry_Date to the record's audit history. If the stamp has moved, your workflow is missing a null check.

Counting bare email opens as activity. Image pre-fetching by mail providers registers opens for recipients who never looked at the message. A decay curve built on raw opens will show implausibly healthy retention and will not correlate with conversion. Require a click, a page visit, a reply, or a meeting. If you must use opens, use them only as a tiebreaker, never as the sole qualifying signal.

Nurture blasts flattening the curve. If a mass email goes out to an entire cohort in week 5, the week-5 activity number spikes for everyone at once. This looks like a recovery and is actually an artifact of your own sending calendar. Two defenses: exclude marketing-initiated touches from the activity definition and count only prospect-initiated actions, or annotate the chart with send dates so readers can see the artifact for what it is. The first is cleaner.

Treating the curve as a scorecard instead of a diagnostic. The instant a decay percentage becomes a number someone is measured on, behavior bends toward the number. Reps log spurious activities; marketing loosens the activity definition. Keep decay as a diagnostic that triggers investigation and keep the accountable metrics downstream — pipeline created, opportunities, revenue. State this explicitly when you introduce the chart, because someone will try to make it a target within two quarters.

Comparing cohorts of different ages. A cohort that is three weeks old has no week-8 data. Plotting it against a twelve-week-old cohort and noting that it "looks better" is a category error. Enforce it structurally: the chart should render nothing beyond a cohort's actual age, and any comparison table should align on weeks-since-entry, never on calendar date.

Ignoring the seasonal and holiday distortions. Cohorts entering the week before a major holiday, at fiscal year-end, or during a large industry conference will decay differently for reasons that have nothing to do with your process. Annotate them. A recurring calendar of known distortion weeks, kept next to the report, prevents a quarterly cycle of investigating the same non-problem.

Building it and not putting it anywhere. The most common failure is organizational, not technical: the report exists, it is correct, and nobody looks at it. Give it a specific home — one slide in the weekly marketing-sales sync, with a standing question attached: "which cohort deviated from the trailing band, and what changed that week?" A decay chart with no ritual around it stops being maintained within a quarter, and then it is worse than nothing because it is stale and still trusted.

Skipping the definition document. Write one page: what qualifies as an MQL, what qualifies as active, the observation window, the cohort boundary (which day starts the week), and the recycle rule. Put the report URL on it. Without this, the third person to inherit the report will quietly redefine "active," and no one will notice until a curve breaks for no reason.

Related questions

How is cohort decay different from lead scoring?

Lead scoring ranks individuals right now to decide who gets called first. Cohort decay measures a fixed group over time to diagnose whether your process is working. Scoring is operational and forward-looking; decay is analytical and retrospective. They answer different questions and should coexist.

What if my MQL volume is too low for weekly cohorts?

Under roughly 30 MQLs per week, switch to biweekly or monthly cohorts. Weekly percentages on small groups swing on single-lead movements and will generate false alarms. You can keep the same snapshot mechanism — only the grouping interval changes.

Should converted leads be removed from the cohort?

No. Removing them changes the denominator and breaks comparability across cohorts. Keep them in the cohort and track conversion as a separate outcome line on the same chart. A cohort that converts well should show that as a win, not as decay.

Can I apply cohort decay tracking to opportunities instead of leads?

Yes, and the mechanism is identical: stamp the stage-entry date, define "still progressing," snapshot weekly. Opportunity-stage aging by cohort surfaces stalled-deal patterns the same way MQL decay surfaces nurture problems. The same snapshot table can hold both.

How long before the data is useful?

You get a readable week-1 and week-2 signal within a fortnight, but genuine pattern recognition needs four to six complete cohorts through at least week 8 — call it three months. Resist drawing conclusions from the first two cohorts; early curves are dominated by whatever was running that month.

FAQ

What exactly is an MQL decay rate?

It is the rate at which a group of Marketing Qualified Leads stops engaging or progressing after entering the funnel, expressed as the percentage of the original cohort still active at each subsequent week. A steep rate means interest evaporates quickly; a shallow rate means the audience stays warm longer. It is a curve, not a single number.

Which CRM field do I actually need to add?

One write-once date field for the MQL entry date, plus a derived cohort-week field (store the Monday date rather than a week number so it sorts correctly across year boundaries). Optionally a "last qualifying activity" date field to simplify the weekly evaluation. Everything else can be computed from those.

How do I define "still active" without gaming it?

Require prospect-initiated signals: a link click, a visit to a page beyond the homepage, a form submission, a reply, or a meeting held. Exclude bare email opens and exclude marketing-initiated sends. Then validate quarterly by checking that high-activity cohorts actually convert better than low-activity ones — if they don't, the definition is measuring noise.

How often should I review the curves?

Weekly during campaign pushes or after any process change, monthly during steady state. The weekly snapshot job should run regardless; the review cadence is separate from the collection cadence. Attach it to an existing meeting rather than creating a new one, or it will quietly stop happening.

What decay pattern should worry me most?

A sharp drop between week 1 and week 2 — it almost always points at speed-to-lead, routing, or queue coverage rather than content. Check median time-to-first-touch for that cohort and whether any rep's queue went unworked. This failure mode is common, cheap to fix, and moves every downstream week of the curve.

Can I fix a bad decay curve, or is it a lead-quality problem?

Both are usually in play. Tighten qualification criteria to stop passing cold leads as MQLs, compress response time on the leads you do pass, and add a genuine re-engagement touch around the point where your curve crosses its half-life. Change one variable at a time on one segment so you can attribute the improvement.

Sources

• https://blog.hubspot.com/marketing/marketing-qualified-lead
• https://knowledge.hubspot.com/reports/create-and-use-funnel-reports
• https://support.google.com/analytics/answer/9670133
• https://help.salesforce.com/s/articleView?id=sf.reports_builder_overview.htm
• https://hbr.org/2011/03/the-short-life-of-online-sales-leads
• https://www.gartner.com/en/marketing/topics/marketing-analytics
• https://www.marketingprofs.com/topic/all/lead-generation
• https://developers.google.com/search/docs/appearance/structured-data/faqpage

Related on PULSE

• [How do you define an MQL so sales and marketing stop arguing about lead quality?](/knowledge/q9780)
• [How does AI in the funnel change your definition of a qualified lead in 2027?](/knowledge/q16456)
• [What question reveals whether a salesperson has properly qualified a lead before entering the pipeline?](/knowledge/q14410)
• [How do you compensate SDRs in a 2027 model where AI books 80% of qualified meetings?](/knowledge/q16296)
• [What is Qualified (Piper the AI SDR) and why is it a hot RevOps tool for 2027?](/knowledge/q12144)
• [Should I pay SDRs on demos booked or only on demos held + qualified?](/knowledge/q08)