Intent Signal Decay and Recency Windows for B2B Buyers

Intent without recency is a history file. SDR and ABM teams need documented decay curves and windows, not only topic taxonomies. This pillar covers how clickstream and related intent signals age, how to set windows, and how to test vendors. Pair with clickstream intent, seed match testing, and ABM stacking.

Why do buyer-intent signals decay?

Intent Decay Is Not Device-Graph Decay

MAID and HEM freshness is a different clock. See device graph decay for identity. Intent decay is about topic timing on accounts you already care about. Do not reuse identity churn percentages as intent windows.

Vendor Questions That Matter

• What timestamp defines an intent event?
• How are bots and non-human traffic filtered?
• What is the maximum age of events in the delivered file?
• How fast do suppressions and deletions propagate?
• Can we re-run the same holdout in 90 days?

Operations: Routing and Suppression

Route only accounts that pass firmographic and technographic gates. Apply intent as a rank, not a universe builder. Suppress customers and open opportunities. Document layer ownership in CRM so marketing and sales see the same freshness badge.

Measure the Decay Curve Instead of Assuming One

Most teams inherit a recency window from a vendor deck and never test it against their own outcomes. The measurement is not complicated, and after one quarter of outreach you already hold the data. Bucket every touched account by the age of the intent event at the moment a rep actually worked it: zero to three days, four to seven, eight to fourteen, fifteen to twenty eight, and older. Then compute reply rate and meeting rate per bucket. What you are looking for is the age at which a bucket stops beating your no-intent baseline. That crossover, not the vendor’s stated window, is where the signal stops paying for a rep’s time.

1. Record event age at touch, not at file delivery. The two differ by your ingestion lag, and that gap is where most of the window quietly disappears.
2. Hold a no-intent control group drawn from the same ICP, so every bucket has something to beat. A bucket’s conversion rate means nothing on its own until you know the rate it has to beat.
3. Require a few hundred touches per bucket before trusting a difference. Intent cohorts are small, weekly noise is large, and a two-point swing on ninety accounts is not a finding.
4. Recompute quarterly. Decay curves move with buying season, category maturity, and your own list fatigue.

Fit a half-life if you want a single number for the scorecard. If your fourteen-day bucket performs at roughly half the lift of your three-day bucket, the practical half-life is about a week, and a twenty-eight-day window is carrying context rather than urgency. One number like that survives a procurement conversation far better than a taxonomy diagram does.

Delivery Lag Eats the Window Before You Do

A seven-day window is only seven days if the file arrives instantly, and it never does. Three clocks sit between a research event and a rep’s screen: the event timestamp, the vendor’s processing and delivery cadence, and your own ingestion and routing schedule. A vendor that observes an event on Monday, batches weekly on Thursday, and lands in your warehouse for a Monday morning sync has consumed the entire window before anyone sees a row. The signal is not stale because intent decays quickly. It is stale because the pipeline is slow, and those two diagnoses lead to completely different fixes.

Ask for the distribution of delivery lag, not the average. A median hides the tail, and the tail is what routes a three-week-old spike into a call that opens on the wrong assumption. Put a maximum event age in the contract next to the refresh cadence, and instrument your own side well enough to prove which of the three clocks is the slow one when lift drops. Vendors are usually willing to fix a lag they can see measured; almost nobody fixes one described as a general complaint about freshness.

One Window Does Not Fit Every Segment

A single global window is a compromise that fits nobody in particular. Enterprise evaluations that run through committees and procurement stages move over months, so a twenty-eight-day signal can still be live. Self-serve and SMB motions turn over in days, where a two-week-old spike is often a competitor’s closed deal. Split the window by segment first, then by motion. The same account can legitimately deserve a seven-day window for a product-led trial nudge and a ninety-day window for an executive briefing invitation.

The failure mode worth watching for is a window tuned during your best quarter. Windows fitted in a hot category cycle overrate freshness once the cycle cools, and the SDR team experiences that as a sudden collapse in a list they were told was improving. Re-fit on a rolling basis and keep the previous window’s numbers visible, so a drop reads as a market change rather than a data-quality incident.

Deletion and Suppression Ride the Same Clock

Recency is a compliance surface as well as a performance one. When a deletion request lands or a supplier removes a panel contributor, that change has to travel the same pipeline that delivers freshness, and it usually travels slower. Ask how long a suppression takes to reach the delivered file, whether it applies retroactively to events already sitting in your warehouse, and what the vendor expects you to delete on your side. A window you cannot honor in reverse is a window you cannot defend. See non-FCRA use cases for where these obligations bite hardest.

Test Recency on a Seed

Use B2B prospecting and clickstream web intent with the pilot process. Request a timed sample via contact.