Outputs on this page come from a live run on September 10, 2026. Company names are masked to initials. No contacts were previewed or revealed.
Headcount decline is one of 24 company signals in the Lusha plugin for Claude, available over one, three, six and twelve month windows. Like IT spend decrease, it means opposite things for a customer and a prospect: a churn warning for one, a consolidation opening for the other. The two signals together are stronger than either alone.
Check my account list for headcount decline and tell me which accounts are contracting for real, which are normal attrition, and how to treat each.
Before anything else, run account_usage and tell me my remaining credits. Do not run any signal call if fewer than 50 credits remain; stop and tell me.
Here is my list (domain, one per line):
[PASTE UP TO 25 DOMAINS]
Customer domains (or write "none"):
[PASTE]
Steps:
1. Run signals_companies_search on the list with signalTypes ["headcountDecrease1m", "headcountDecrease3m"], startDate 90 days ago, maxResultsPerSignal 2. Batch in groups of 25 if the list is longer.
2. For every hit, pull the baseline employee count, the new count, the change rate and the reading date for each window. Compute people lost as baseline minus new. Keep accounts with no signal in the output as "no signal."
3. Classify: CONTRACTING if the 1 month and 3 month windows both show a decline and the 3 month rate is 5% or worse; TRENDING if both windows agree but the rate is under 5%; SPIKE if only the 1 month window fired at 5% or worse; NOISE if the decline is under 2% in every window. Normal attrition is 1 to 2% a month; do not treat it as a signal.
4. Show percentage and people side by side. Flag VERIFY on any account where the baseline looks larger than the company's real payroll (marketplaces, franchises and agencies carry contractors and agents in their LinkedIn count) or where a single month shows a drop of 15% or more.
5. Tell me the most recent reading date in the results. Readings are monthly and can lag; if the newest is more than 40 days old, say so in one line at the top.
6. Tag customers as CHURN RISK and route them to the account owner. Tag everyone else as CONSOLIDATION.
7. For the top 5 CONSOLIDATION accounts that are CONTRACTING or TRENDING, write one opening line that offers to do the same work with fewer seats or fewer tools. Do not use the words layoffs, struggling, downsizing or cuts, and do not quote a figure.
Optional step 8, if credits allow: rerun the same list with signalTypes ["itSpendDecrease"] and mark accounts that fire on both as STACKED. Those are the accounts to work first.
Output: one table with columns Account | 1 month | 3 month | People lost | Class | Reading dates | Segment | Opening line. Then two lines: what the classification is based on, and credits used. Do not preview or reveal any contacts in this run.The situation. The same 25 companies used for the IT spend decrease play: software, marketplaces, real estate tech and infrastructure, 700 to 36,000 LinkedIn employees. No customer domains supplied.
The run. account_usage first, at zero credits. Then one batch call on 25 domains for headcountDecrease1m and headcountDecrease3m, 90 day window, two readings per signal. 13 of 25 fired, 23 events, 23 credits. Readings were dated July 1 and August 1; the newest was 40 days old on the day of the run, and the prompt’s freshness check would have said so.
What came back.
| Account | 1 month | 3 month | People lost | Class | Reading dates | Also on IT spend decrease | Segment | Opening line |
|---|---|---|---|---|---|---|---|---|
| O.D. (real estate, ~1,500 emp) | -11% | -10% | ~190 | CONTRACTING | Jul 1 | Yes (-54%, -55%) | CONSOLIDATION, STACKED | When a team gets smaller, the tools it pays for rarely do. Worth twenty minutes on which data lines can become one. |
| Z.L. (real estate marketplace, ~10,000 emp) | -20%, then -22% | -12% | ~6,100 over two months | CONTRACTING, VERIFY | Jul 1, Aug 1 | No | CONSOLIDATION | Held for verification: the baseline includes agents, and a fall of six thousand in two months on a marketplace count is a change in who lists the company as employer, not necessarily in payroll. |
| V.M. (video platform, ~900 emp) | -3% | -6%, then -4% | ~90 over three months | TRENDING | Jul 1, Aug 1 | No | CONSOLIDATION | Three readings in the same direction is a plan, not a month. If the stack is being reviewed line by line, verified data is the line that gets cheaper as it feeds more of the rest. |
| E.B. (communications, ~3,000 emp) | -5% | -4% | ~135 | TRENDING | Jul 1 | No | CONSOLIDATION | Fewer people carrying the same number means every rep needs a fuller book on day one. That is the conversation. |
| C.G. (edtech, ~6,500 emp) | -2% | -3% | ~150 | TRENDING | Jul 1 | Yes (-77%, -78%) | CONSOLIDATION, STACKED | Headcount alone is modest here; the stacked IT spend reading is what moves this account up. Lead with the consolidation offer from the IT spend play. |
| D.M. (analytics, ~1,350 emp) | -1% | -2% | ~30 | NOISE | Jul 1 | Yes (-32%) | STACKED | Headcount is attrition. Work it on the IT spend signal, not this one. |
| C.P. (real estate tech, ~35,000 emp) | -2% | -2% | ~800 | NOISE, VERIFY | Jul 1 | Yes (-63%, -51%) | STACKED | Agent heavy baseline; 800 people is 2%. Work it on the IT spend signal. |
| C.R. (online learning, ~6,600 emp) | -4% | no reading | ~275 | SPIKE | Jul 1 | No | CONSOLIDATION | Watch next month. |
| E.V. (events, ~950 emp) | -4% | no reading | ~35 | SPIKE | Jul 1 | No | CONSOLIDATION | Watch next month. |
| R.S., B.C., Y.X., S.S. | -1% to -2% | no reading | 12 to 130 | NOISE | Jul 1 | R.S. and B.C. yes | Attrition. R.S. and B.C. move on the IT spend signal. | |
| 12 accounts | no signal | 3 fired on IT spend decrease only |
Classification based on agreement between the 1 and 3 month windows and a 5% threshold, with people shown against percentage. Credits used: 23 (plus 15 for the IT spend decrease rerun in play 2). Confirmed live via Lusha connector, September 10, 2026. Company names masked to initials.
What happens next. The STACKED rows are the list: six accounts where both people and technology budget fell in the same window. Take those to the decision maker play with the budget owner as the target and the consolidation line as the opener. The CHURN RISK rows (none here, no customers supplied) go to the account owner for a business review with a smaller plan on the table, not to a sequence. The SPIKE rows get one more month.
Sales teams already know when a big company announces layoffs; the news does that. What they do not see is the slow contraction with no announcement, where a team loses five percent a quarter and its tooling stays the size it was. This signal reads that from employee counts over four windows, and the prompt does the arithmetic a rep would never do by hand: agreement between windows, people against percentage, and a sanity check on baselines that include contractors. Running it inside Claude alongside IT spend decrease turns two mediocre signals into one good one, in two calls, without touching a contact until the account has earned a reveal. Every signal is sourced under Lusha’s data methodology and compliance framework.
A fall in the company’s LinkedIn employee count between a baseline and a new reading, over a one, three, six or twelve month window. The signal carries both counts, the rate and the reading date. It measures profiles, not payroll.
Because they measure different baselines. A company can be flat over three months and down over one, or down steadily over three months at a rate too small to register in any single month. Agreement between them is what turns a reading into a trend.
Because monthly attrition at most companies runs between one and two percent, and replacement hiring hides it. Five percent in a quarter with no offsetting hires is a decision, not turnover. Raise it if your accounts are large and stable; lower it to 3% for companies under 500 people, where 5% is a handful.
Marketplaces, franchises, real estate brokerages and agencies, where people who are not employees list the company on their profile. Also any account showing a single month drop of 15% or more. The prompt flags both.
One credit per event returned. Two signals with two readings each is up to four credits per account that fires; in the live run, 25 domains cost 23 credits. The optional IT spend rerun cost 15 more. account_usage runs first and the prompt stops under 50. No contact preview or reveal runs in this play.
People News reports announced reductions from published articles. This signal reads the employee count directly, so it catches the unannounced ones, which are most of them. Use both if you want the quiet and the loud.
One data connection. Works in Claude, ChatGPT, your CRM, or any agent you build.
