Part of Lusha’s ongoing measurement of B2B data decay. First measured August 15, 2026. Re-run September 12, 2026, and the two year rate moved by three hundredths of a point. Both runs are below, with the method.
Every B2B data vendor cites the same statistic: contact data decays at roughly 30% a year. We have cited it ourselves. It appears in dozens of vendor blogs, in sales enablement decks, in pitch material across the category.
We went looking for the original study behind it. There isn’t one we could find. The figure appears without primary attribution wherever it is repeated, each source citing another source citing another.
So we measured it against our own database instead. Here is what came back.
The short version: among US sales leaders we track, 12.25% changed roles over twelve months and 25.67% over twenty four. The widely repeated 30% figure is close to a two year rate, not an annual one. The monthly rate is stable at about 1%, and it varies more by country than by anything else we measured: UK sales leaders changed roles at 1.75 times the US rate.
Figures in this summary are from the September 12, 2026 run. The August 15 run is shown alongside every table. Methodology, sample sizes and limitations are stated in full at the end. No individual contacts are named.
The curve: about 1% a month, and it holds
We took a single cohort, US sales leaders at VP and C suite level, and asked the same question over three time windows.
| Window | Changed role | % of cohort | Per month |
|---|---|---|---|
| 3 months | 2,101 | 1.5% | 0.50% |
| 12 months | 17,187 | 12.25% | 1.02% |
| 24 months | 36,015 | 25.67% | 1.07% |
Cohort 140,284, measured September 12, 2026. The three month figure is from the August 15 run against a cohort of 140,964; it has not been re-measured.
The twelve and twenty four month figures agree at just over 1% per month. That agreement is what makes the annual number trustworthy: it is not an artifact of where we drew the window.
And 25.67% over two years is close enough to the industry’s 30% that the likely explanation is simple. The number is roughly right. It is being quoted over the wrong period.
The three month figure is lower, and that is a finding in itself
At three months the rate drops to 0.50% per month, half the longer run figure. Job changes do not slow down in recent months, so this is a detection lag: a change takes time to surface in any dataset that observes employment rather than being told about it directly.
The practical consequence is worth sitting with. A record that looks current may already be wrong. Recency of a refresh is not the same as accuracy, because the newest changes are the ones least likely to have been captured yet. That is an argument for verifying at the point of use rather than trusting a refresh date.
Does it hold? The September re-run
A measurement published once is a claim. A measurement that survives being re-run is a rate.
So on September 12, four weeks after the first pull, we ran the same four cohorts again against the same filters. Six API calls. Six credits. Nothing else changed.
| Measure | August 15, 2026 | September 12, 2026 | Change |
|---|---|---|---|
| US sales VP and C suite, cohort size | 140,964 | 140,284 | −680 |
| Changed role, 12 months | 17,719 (12.6%) | 17,187 (12.25%) | −0.35 points |
| Changed role, 24 months | 36,281 (25.7%) | 36,015 (25.67%) | −0.03 points |
| UK sales VP and C suite, cohort size | 4,223 | 4,214 | −9 |
| UK changed role, 12 months | 918 (21.7%) | 902 (21.40%) | −0.3 points |
| UK against US ratio | 1.73x | 1.75x | +0.02 |
| US marketing VP and C suite, cohort size | 148,011 | 147,825 | −186 |
| Marketing changed role, 12 months | 20,839 (14.1%) | 20,298 (13.73%) | −0.37 points |
| Marketing against sales ratio | 1.12x | 1.12x | unchanged |
The twenty four month figure is the one to watch
25.7% against 25.67%. Three hundredths of a percentage point across four weeks and 140,000 people.
That is the result that makes the rest of this page usable. A rate drawn from a single pull can be an accident of the week you drew it. A rate that returns the same answer from a window that has slid forward by a month is measuring something real. The monthly figure behind it is 1.07% in both runs.
The twelve month figures all drifted down, and we do not know why
Three cohorts, three declines, all between 0.3 and 0.37 points. Three readings moving the same direction is a pattern rather than noise, and we are not going to pretend to know what caused it.
Two explanations fit the data equally well. Senior job changes may have slowed over the northern hemisphere summer, which would show up first in the shorter window. Or the twelve month window has rolled forward to include a quieter August and drop a busier one. A single re-run cannot separate those, and anyone telling you otherwise from two data points is guessing.
The detection lag documented above is a third possibility worth holding: the newest changes are the ones least likely to have surfaced yet, so a window ending today always under counts its own final weeks. That effect would push the twelve month number down slightly and leave the twenty four month number almost untouched, which is close to what we see.
We will keep re-running it. If the twelve month rate continues falling across the next two pulls, that is a trend worth a separate write up. If it returns to 12.6%, it was seasonality.
The two ratios did not move
The regional gap went from 1.73x to 1.75x. The function gap held at 1.12x. Absolute rates drifted; the relationships between them did not.
This is the practically useful half of the re-run. If you set refresh cadence on the ratio, a UK list needs checking roughly 1.75 times as often as a US one, and that instruction survives the rate moving around underneath it. Cadence built on the relationship is more durable than cadence built on the headline number.
Where you sell changes the answer more than what you sell
We ran the same twelve month measurement on sales leaders in four countries.
| Country | Cohort | Changed role | Rate |
|---|---|---|---|
| United Kingdom | 4,214 | 902 | 21.40% |
| Germany | 4,204 | 722 | 17.2% |
| India | 15,876 | 2,120 | 13.4% |
| United States | 140,284 | 17,187 | 12.25% |
UK and US re-measured September 12, 2026. Germany and India are from the August 15 run and have not been re-measured.
A UK sales list goes out of date 1.75 times faster than an equivalent US one. If you run the same quarterly refresh cadence across both territories, your UK data is meaningfully worse than your US data at every point in the cycle, and nothing in your process would tell you.
One detail is worth flagging because it addresses the obvious objection. The UK and Germany cohorts are almost identical in size, 4,214 and 4,204, and return different rates, 21.40% against 17.2%. If the regional variation were an artifact of how much data we hold per country, two cohorts of the same size would not diverge by four points. The variation appears to be in the underlying labour markets rather than in the measurement.
By function: marketing moves fastest, engineering slowest
| Function (US, VP and C suite) | Cohort | Changed role | Rate |
|---|---|---|---|
| Marketing | 147,825 | 20,298 | 13.73% |
| Sales | 140,284 | 17,187 | 12.25% |
| Engineering and technical | 94,692 | 9,397 | 9.9% |
Marketing and sales re-measured September 12, 2026. Engineering is from the August 15 run.
A 1.12x spread between marketing and sales, and roughly 1.4x between marketing and engineering. If you sell to marketing leaders your list decays faster than if you sell to engineering leaders, and a single blended refresh policy will over serve one and under serve the other.
What did not vary: seniority
We expected seniority to matter. It doesn’t.
| Seniority (US sales) | Cohort | Changed role | Rate |
|---|---|---|---|
| Vice president | 125,399 | 15,786 | 12.6% |
| C suite | 15,565 | 1,933 | 12.4% |
From the August 15 run; not re-measured in September.
Two tenths of a percentage point apart. We are publishing this because a null result is still a result, and because the assumption that senior people are more stable, or less stable, is one we have heard often enough to be worth testing. On this evidence, seniority is not a useful input to a refresh policy. Country and function are.
What to do with this
Stop budgeting refresh cadence off a single annual figure. On these numbers a US sales list is about 6% wrong at six months, 12.25% at twelve and 25.67% at twenty four. A UK list reaches that same 12.25% in roughly seven months.
Set cadence by territory before you set it by anything else. Region was the strongest variable we measured, at 1.75x between the fastest and slowest markets, and it is the variable that held steady when the absolute rates moved. A uniform global refresh policy is systematically under serving your fastest moving territory.
Verify at the point of use for anything high stakes. The three month detection lag means a recent refresh date is weaker evidence than it looks. For a list going into a sequence, a check at send time catches what a scheduled refresh has not yet seen.
Ignore seniority when planning cadence. It does not predict anything useful here.
Methodology
Every figure comes from Lusha’s contact database, measured through the same API our customers use. First run August 15, 2026. Second run September 12, 2026.
What we measured. For a defined cohort, a department, a seniority band, a country, we counted the total population, then counted how many of those contacts have an employment change recorded within a given window. The ratio is the job change rate.
Cohort definitions. Seniority is vice president and C suite unless stated otherwise. Function is the department classification on the contact record. Country is the contact’s location, not the company’s headquarters.
Sample sizes are stated in every table. The largest cohort is 147,825 contacts and the smallest is 4,204. Nothing here is extrapolated from a sample; these are counts of the full population within each cohort.
Cost. The August run was 15 API calls. The September re-run was 6 calls and 6 credits, about four minutes.
Re-run schedule. Monthly. Each re-run is added to this page rather than published as a new post, so the URL accumulates the history instead of fragmenting it. Next scheduled re-run: October 2026.
Reproduce it. The filters are prospecting contact search with departments, seniority, countries and jobChangedAfterDate. Take the total for the unfiltered cohort, then the total with the date filter applied, and divide. Any Lusha customer can run the same six calls against their own territory and function. If you get a materially different answer, we want to hear about it.
What this data cannot tell you
Four limitations, stated plainly, because a benchmark that hides them is not worth citing.
This is a job change rate, not a full decay rate. We measured employment changes. A record can also go stale through a title change within the same company, a phone number changing, or an email format migration. Those are not counted here, which means the true rate at which records become wrong is higher than the figures above.
It measures detected changes. A change we have not yet observed is not in the count. The three month figure demonstrates this directly. Every number here is a floor rather than a ceiling.
The denominator is the present day population, not a cohort fixed twelve months ago. A true longitudinal measurement would define the group at the start of the window and follow it forward. We measured the group as it stands today and looked backward. These are close but not identical, and the difference is worth knowing before anyone cites the figure as a decay rate.
Four countries and three functions is a narrow slice. The regional variation we found is striking enough that we would want a wider set before treating 1.75x as the outer bound. It may not be.
Why we published a lower number than the one we had been using
The figure we measured is well below the one this category repeats, including on our own pages, which we are now correcting.
A smaller number is less useful for creating urgency. It is more useful for planning, which is what the number is actually for. And a figure with a stated methodology, stated sample sizes and stated limitations is worth more than a larger one that nobody can trace to a source.
If you find an error in this, we would like to know. The method is in the section above and the cohorts are reproducible in six API calls.
And we will keep running it. A number published once and left alone becomes exactly the kind of unsourced statistic this page was written to replace. The re-run above is the first of those checks. The next is in October.
Go deeper
- How fast B2B contact data decays, and what it costs — every report we have published on this, in one place
- Refresh stale CRM records before you reach out — the prompt that applies this to a list you already hold
- Monitor CRM contacts for promotions and company changes — the continuous version
- Five checks before you trust a buying signal — job changes read as a trigger rather than a problem
- Which platforms refresh stale CRM data, and how fast does it need to be?: what this rate means for refresh cadence
- 10 best data enrichment tools in 2026: which tools do the refresh
- B2B Contact Change Report — the same question measured by volume rather than rate
Want to run this measurement against your own territory and function? Add the Lusha plugin to Claude
FAQ
How fast does B2B contact data actually decay?
We measured 12.25% of US sales leaders changing roles over twelve months and 25.67% over twenty four, which is about 1% per month, and the rate holds across both windows. It varies by market: UK sales leaders changed roles at 21.40% against 12.25% in the US. This counts employment changes only, so the rate at which records become wrong in some way is higher.
Where does the 30% data decay statistic come from?
We could not find a primary source for it. It appears across vendor blogs, sales decks and pitch material throughout this category, each citing another secondary source. Our measurement suggests the number is roughly right but attached to the wrong period: 25.67% over twenty four months is close to 30%, while the annual figure is about half that.
Has this measurement been repeated?
Yes. We re-ran all four cohorts on September 12, 2026, four weeks after the first pull. The twenty four month rate moved from 25.7% to 25.67%, the regional gap from 1.73x to 1.75x, and the function gap held at 1.12x. The twelve month rates each fell about a third of a point, which could be summer seasonality, a window effect or detection lag; two data points cannot tell them apart. Both runs are published above.
How often should I refresh my CRM contact data?
Set the cadence by territory before anything else, since region varied more than any other factor we tested and held steady across both runs. On our numbers a US sales list is around 6% wrong at six months and 12.25% at twelve; a UK list reaches that same point in roughly seven months. For lists going into a sequence, a check at send time catches what a scheduled refresh has not yet seen.
Does a recent refresh date mean the data is accurate?
Less than you would expect. Our three month measurement returned half the monthly rate of the longer windows, which is a detection lag rather than a real slowdown: the newest job changes are the ones least likely to have surfaced in any dataset that observes employment rather than being told about it. A record that looks current can already be wrong.
Which job functions have the highest turnover?
Among US VP and C suite contacts: marketing at 13.73%, sales at 12.25%, engineering and technical at 9.9%. The marketing to sales ratio held at 1.12x across both runs, which means a single blended refresh policy over serves one function and under serves another.
Do senior people change jobs less often?
No meaningful difference in our data. US sales VPs came in at 12.6% and C suite at 12.4%, two tenths of a point apart. Seniority is not a useful input to a refresh policy; country and function are.
How often do you update this page?
Monthly. Each re-run is added here rather than published as a new post, so the history stays in one place and the figures can be compared directly. Every run states its date, its cohort sizes and its cost.
