# Methodology: Do recruiter hires predict company growth?

By Cristian Frunze, Founder of Kenso · Nov 7, 2026 · [Kenso Research](/research) No. 20 · Hiring and org design, Sales teams

The full methodology behind [Do recruiter hires predict company growth?](/research/recruiters-predict-growth.md): the data, every method, every number and every limitation. The same text as the [methodology PDF](/research/recruiters-predict-growth.pdf).

Data: Kenso internal data, Kenso's dataset of public professional profiles, work histories and company records. Every number is rounded once, half-up, from the computed value.

## 1. Question and short answer

When a company adds recruiters, does headcount growth follow two to four quarters later? When it cuts recruiters, do cuts follow? How strong and how early is the signal?

- **Adding recruiters goes with growth, mostly at the same time.** When recruiters rise by at least one and 25% in a quarter, the rest of the company's headcount is 10% or more above its level at some point two to four quarters later in 43.1% of cases, against a base rate of 28.1% (lift 1.53; 1.62 against the same quarter and size band). Momentum (the company-quarters with the highest growth in the signal quarter, as many as the recruiter rule picks) is right 63.3% of the time.
- **The link is short.** Per extra recruiter per 100 staff, growth in the same quarter is 1.27 log points higher, in the next 1.06, three and four quarters later 0.40 and 0.35, six quarters later 0.01.
- **Lead times vary.** After a recruiter increase, the median wait until headcount is 10% above its level at the signal is 3 quarters (quartiles 2 and 5), among the 61.5% that get there within 8 quarters; 47.9% of increases follow a year of 10%+ growth (all company-quarters: 33.7%).
- **Cutting recruiters warns of cuts, weakly.** 22.8% of recruiter cuts (down by at least one and 25%) are followed by headcount 10% or more lower two to four quarters later, against 12.5% (lift 1.82); momentum down: 50.6%.
- **Recruiters add to momentum.** With the company's own growth in the signal quarter and the quarter before, and its change in salespeople, held fixed, the 4-quarter regression coefficient falls from 2.43 to 1.76 log points per recruiter per 100 staff (72.7% of it remains).

These are associations in rebuilt company histories, measured on staff in the US, UK, Canada, Australia, Ireland and New Zealand. They are not causal effects.

## 2. Data

- **Source.** Kenso internal data, through Kenso's shared headcount history extract (built for the Kenso Research study *Did AI kill the SDR role?* on 2026-10-16/17), read-only: company records, quarterly total headcount per company, quarterly headcount per company, function and level (the function-level headcount, labelled with the shared sales title taxonomy of the Kenso Research study *Who are GTM engineers, and where do they come from?*), and the person-level role spells behind them (kept on Kenso's servers).
- **Coverage.** The people data holds only people whose current country is the US, UK, Canada, Australia, Ireland or New Zealand, so all headcounts are staff in these six countries. Profiles were last refreshed between 2025-11 and 2026-04; the snapshot is 2026-04-30 and the last complete quarter 2026 Q1. People add new jobs months late (the SDR study measured that 41.1% of job changes Kenso detected in June 2026 had started in 2026 Q1), so recent quarters undercount.
- **How a person is counted.** A person counts at a company at a quarter end when one of their spells there covers that date: started on or before it, and ended on or after it (the end month counted in full) or still current (current spells run to the snapshot). Spells with no start date, or no end date and not current, are dropped. Quarters run from 2015 Q1 (q = 0) to 2026 Q1 (q = 44).
- **Spells read.** 555,053,752 spells, 342,882,318 kept under these rules, 258,807,098 of them at a known company; 4,232,766 carry a title the shared taxonomy labels recruiting or HR.

## 3. Recruiters, HR and the rest of the company

- **Recruiters (R).** Distinct people at the company at the quarter end whose normalized title (lower case, non-word characters to spaces) (a) gets the shared recruiting-or-HR label, (b) matches the recruiter rule of the Kenso Research study *Who gets cut first in a B2B layoff? Recruiters, then SDRs* (recruit, recruiter, recruiting, recruitment, talent acquisition, talent partner, talent advisor, talent lead, talent sourcer, talent sourcing, talent associate, sourcer, headhunter), (c) is not at the shared intern level (intern, internship, trainee, apprentice, working student), and (d) does not name recruiting of people who are not staff (students, patients, study or trial participants, donors, volunteers, members, franchisees, athletes and sports coaches, admissions and enrolment, committee and chair roles, market-research panels). Each person counts once per company and quarter.
- **Title audit.** Rule (d) was fixed from an audit of the titles before any outcome was computed (Decision D7). Version 1 classed 2,199,940 spells as recruiters; version 2 moved staff titles such as "recruitment team member" back in and sports, committee and market-research titles out, leaving 2,194,556 recruiter spells, 79,431 non-staff recruiting spells, 36,721 intern-level spells and 1,922,058 HR spells. 304,199 recruiter spells have titles that name contract, freelance, agency or consulting work (robustness b). The most common recruiter titles are recruiter, recruitment consultant, talent acquisition specialist and technical recruiter; procurement titles such as "strategic sourcing manager" and scheduling titles such as "staffing coordinator" stay in HR. The audit tables list every title held by 50 or more people.
- **HR (contrast).** Distinct people with a non-intern recruiting-or-HR title who are not recruiters at that quarter end.
- **Check.** The same pass counts every recruiting-or-HR title once per person and level; it equals the shared function-level count in all 10,593,118 company-quarter cells (0 differ).
- **Rest of the company (N).** N = H - R, where H is total headcount: everyone at the company at the quarter end without a recruiter title then. Growth is measured on N so that recruiters added cannot count as the growth they predict; total headcount is robustness (a).
- **Salespeople (S).** Function-level headcount summed over the sales labels of the layoffs study and the Kenso Research study *Sales and marketing draw level with engineering at about 50 people* (SDR and BDR, account executives, account management, general sales, partnerships and business development, sales engineering, revenue operations, enablement, GTM engineering and other GTM roles), every level except intern.
- **Growth.** y = ln(1 + N); growth over k quarters g(t, k) = 100 x (y(t + k) - y(t)), in log points (100 log points is a factor of e; small values are close to percent).

## 4. Sample

- 23,338,222 company records; 23,277,621 with HQ in the six countries; staffing and recruiting (103,355), executive search (1,524) and human resources services (63,293) dropped: 23,109,449 in the frame.
- Panel companies: frame companies with H of 50 or more at some quarter end from 2015 Q4 to 2024 Q4: 161,801 (7,281,045 company-quarter rows, q 0-44). By industry group (the map of the layoffs study): consumer, public sector and the rest 73,793, manufacturing, industrial and energy 32,352, business and professional services 18,687, software and IT 14,175, financial services 11,863, other technology, telecom and media 6,276, healthcare and life sciences 3,964, no industry on the record 691.
- Eligible company-quarters: H(t - 1) >= 50 and N(t) >= 1, signal quarter t from 2016 Q1 to 2025 Q1 (every 4-quarter window ends by 2026 Q1): 4,004,937 at 161,801 companies; median size 104 people; by size band 50-199 2,920,777, 200-999 860,384, 1,000-4,999 187,669, 5,000+ 36,107.
- Recruiters per 100 staff (pooled): 0.58; 64.9% of eligible company-quarters had no recruiter in the quarter before; the recruiter count changed in 10.9%.

## 5. Methods

1. **Panel regression.** For k = 1 to 4: g(t, k) = b_k x r(t) + company effect + quarter effect + error, with r(t) = 100 x (R(t) - R(t - 1)) / H(t - 1) (recruiters added per 100 staff), on all eligible company-quarters. Each outcome is winsorized at its 0.5th and 99.5th percentiles. Model A has r(t) only; Model B adds growth in the signal quarter g(t - 1, 1), in the quarter before g(t - 2, 1), and salespeople added per 100 staff s(t); Model C splits r(t) into additions and cuts. Both sets of effects are absorbed by alternating projections (stopping when the largest mean removed is below 1e-10); errors are clustered by company (CR1). A check computes every slope a second way (company demeaning plus explicit quarter dummies); the two agree to within 3e-13.
2. **Timing profile.** Model A with growth in a single quarter, g(t + j - 1, 1), as the outcome, j = -3 to +6 (j = 0 is the quarter of the recruiter change), on signal quarters 2016 Q1-2024 Q3.
3. **Signal scoring.** Rules at t: recruiters up (R up by at least 1 and at least 25%: R(t) - R(t - 1) >= 1 and 4 R(t) >= 5 R(t - 1)); recruiters down (down by at least 1 and 25%); the same rules on salespeople and on HR; momentum up and down (the company-quarters with the highest or lowest growth in the signal quarter, exactly as many as the recruiter rule picks, ties by a fixed company key); salespeople added per 100 staff at the same count. Targets: growth if N at t + 2, t + 3 or t + 4 is at least 1.10 x N(t); fall if it is at most 0.90 x N(t) (compared in integers). Scores: precision (share of fired company-quarters that hit the target), base rate, lift (precision over base rate), adjusted lift (hits over the hits expected from the base rate of each fired company-quarter's quarter and size band), recall (share of target company-quarters where the rule fired). 95% intervals from 500 company bootstrap draws (Poisson weights).
4. **Lead times.** For signals 2016 Q1-2024 Q1 (8 quarters observed after each): the first k from 1 to 8 with N(t + k) >= 1.10 x N(t) (falls: <= 0.90 x N(t)), or none; median and quartiles are the smallest k whose cumulative share among those with a lead time reaches 50% (25%, 75%). Already growing: N(t) >= 1.10 x N(t - 4).
5. **Event paths.** Events: recruiters up (or down) at t from 2017 Q1 to 2024 Q1. Controls: eligible company-quarters in the same cell where the rule did not fire, cell = quarter x size band x recruiters before (0, 1-2, 3-9, 10+) x quintile of growth over the 4 quarters up to t. Path = y(t + j) - y(t), j = -8 to +8; excess = event path minus the mean path of its cell's controls, averaged over events, 500 bootstrap draws. The window j = -8 to -4 is outside the matching and is a pre-trend check.
6. **Robustness.** (a) total headcount; (b) recruiters without contract or consulting titles; (c) signals up to 2024 Q2, so every window ends by 2025 Q2; (d) software and IT; (e) size bands; (f) recruiters before; (g) 5% and 20% targets; (h) the rules over two quarters; (i) log change of recruiters as the regressor; (j) outcomes not winsorized; (k) industry groups. Momentum and sales comparisons are rebuilt at the recruiter rule's count in each sample.
7. **Privacy.** Only aggregate tables leave Kenso's servers; published cells rest on at least 50 companies; counts under 50 are shown as "<50".

## 6. Results

### 6.1 Panel regression (log points of growth per recruiter added per 100 staff)

| Model | k = 1 | k = 2 | k = 3 | k = 4 (95% interval) |
|---|---|---|---|---|
| A | 1.07 | 1.70 | 2.08 | 2.43 (2.29 to 2.56) |
| B | 0.88 | 1.47 | 1.73 | 1.76 (1.65 to 1.87) |
| C, additions | 1.34 | 2.12 | 2.54 | 2.96 (2.77 to 3.15) |
| C, cuts | 0.67 | 1.07 | 1.40 | 1.63 (1.45 to 1.80) |

Model B controls at k = 4: growth in the signal quarter 0.255, the quarter before 0.215, salespeople added per 100 staff 0.339. Model B keeps 0.727 of Model A's 4-quarter coefficient. 4,004,937 company-quarters, 161,801 companies. Model C's cut coefficient multiplies negative values: each recruiter cut per 100 staff goes with 1.63 log points less growth over 4 quarters.

### 6.2 Timing profile

| Quarter of growth (j) | -3 | -2 | -1 | 0 | 1 | 2 | 3 | 4 | 5 | 6 |
|---|---|---|---|---|---|---|---|---|---|---|
| Coefficient | 0.32 | 0.44 | 0.71 | 1.27 | 1.06 | 0.66 | 0.40 | 0.35 | 0.12 | 0.01 |
| 95% interval | 0.29 to 0.35 | 0.41 to 0.48 | 0.67 to 0.74 | 1.20 to 1.34 | 1.00 to 1.12 | 0.62 to 0.70 | 0.36 to 0.43 | 0.32 to 0.38 | 0.10 to 0.15 | -0.02 to 0.03 |
| Share of the peak | 0.25 | 0.35 | 0.56 | 1.00 | 0.83 | 0.52 | 0.31 | 0.28 | 0.09 | 0.01 |

3,739,901 company-quarters at 158,539 companies. The peak among j = 0 to 4 is at j = 0.

### 6.3 Signal scores (headline sample)

| Rule | Target | Fired | Companies | Firing rate | Precision (95% interval) | Base rate | Lift (95% interval) | Adjusted lift | Recall |
|---|---|---|---|---|---|---|---|---|---|
| Recruiters up 1+ and 25%+ | growth | 167,725 | 62,725 | 4.19% | 43.1% (42.7 to 43.4) | 28.1% | 1.53 (1.52 to 1.54) | 1.62 | 6.4% |
| Momentum up (same count) | growth | 167,725 | 66,086 | 4.19% | 63.3% (63.0 to 63.7) | 28.1% | 2.25 (2.24 to 2.26) | 2.07 | 9.4% |
| Salespeople up 1+ and 25%+ | growth | 208,241 | 79,957 | 5.20% | 42.1% (41.8 to 42.4) | 28.1% | 1.50 (1.49 to 1.51) | 1.46 | 7.8% |
| Most salespeople added (same count) | growth | 167,725 | 51,847 | 4.19% | 47.8% (47.5 to 48.1) | 28.1% | 1.70 (1.69 to 1.71) | 1.58 | 7.1% |
| HR up 1+ and 25%+ | growth | 227,565 | 85,660 | 5.68% | 35.0% (34.7 to 35.2) | 28.1% | 1.24 (1.24 to 1.25) | 1.27 | 7.1% |
| Recruiters down 1+ and 25%+ | fall | 112,698 | 53,946 | 2.81% | 22.8% (22.5 to 23.2) | 12.5% | 1.82 (1.80 to 1.85) | 1.99 | 5.1% |
| Momentum down (same count) | fall | 112,698 | 53,157 | 2.81% | 50.6% (50.2 to 51.1) | 12.5% | 4.04 (4.00 to 4.08) | 3.71 | 11.4% |
| Salespeople down 1+ and 25%+ | fall | 132,521 | 64,434 | 3.31% | 25.6% (25.3 to 25.9) | 12.5% | 2.04 (2.02 to 2.06) | 1.98 | 6.8% |
| HR down 1+ and 25%+ | fall | 157,037 | 74,767 | 3.92% | 17.6% (17.3 to 17.8) | 12.5% | 1.40 (1.39 to 1.42) | 1.45 | 5.5% |

Hit rates within momentum groups (headline):

| Side | Momentum rule fired | Recruiter rule fired | Company-quarters | Companies | Hit rate (95% interval) |
|---|---|---|---|---|---|
| up | yes | yes | 19,100 | 12,048 | 79.8% (79.1 to 80.3) |
| up | yes | no | 148,625 | 64,251 | 61.2% (60.9 to 61.6) |
| up | yes | difference (fired minus not) | 167,725 | 66,086 | 18.5 (17.8 to 19.1) |
| up | no | yes | 148,625 | 59,431 | 38.3% (38.0 to 38.7) |
| up | no | no | 3,688,587 | 161,309 | 26.1% (26.0 to 26.3) |
| up | no | difference (fired minus not) | 3,837,212 | 161,412 | 12.2 (11.9 to 12.5) |
| down | yes | yes | 10,557 | 8,303 | 69.5% (68.6 to 70.5) |
| down | yes | no | 102,141 | 50,521 | 48.7% (48.2 to 49.1) |
| down | yes | difference (fired minus not) | 112,698 | 53,157 | 20.8 (19.9 to 21.8) |
| down | no | yes | 102,141 | 51,129 | 18.0% (17.7 to 18.3) |
| down | no | no | 3,790,098 | 159,615 | 11.2% (11.2 to 11.3) |
| down | no | difference (fired minus not) | 3,892,239 | 159,743 | 6.8 (6.5 to 7.0) |

Paired precision differences (percentage points, same bootstrap draws): recruiters up minus momentum up: -20.3 (-20.7 to -19.9); recruiters up minus most salespeople added: -4.7 (-5.1 to -4.3); recruiters down minus momentum down: -27.8 (-28.2 to -27.4).

### 6.4 Lead times

Share of company-quarters by quarters until headcount (recruiters left out) is 10% above (growth side) or below (fall side) its level at the signal:

| Side | Group | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | none |
|---|---|---|---|---|---|---|---|---|---|---|
| up | recruiters up | 13.1% | 14.0% | 10.1% | 7.9% | 5.7% | 4.3% | 3.5% | 2.9% | 38.5% |
| up | all eligible | 5.8% | 9.3% | 8.5% | 6.9% | 5.7% | 4.6% | 3.8% | 3.2% | 52.1% |
| up | momentum up | 27.2% | 18.6% | 8.7% | 11.6% | 3.1% | 2.2% | 1.5% | 3.6% | 23.6% |
| down | recruiters down | 7.3% | 6.9% | 5.1% | 3.9% | 3.0% | 2.6% | 2.2% | 1.9% | 67.2% |
| down | all eligible | 2.8% | 3.7% | 3.3% | 2.9% | 2.6% | 2.3% | 2.0% | 1.8% | 78.6% |
| down | momentum down | 24.9% | 13.6% | 6.7% | 6.8% | 3.5% | 2.0% | 1.5% | 2.7% | 38.4% |

| Side | Group | Company-quarters | Companies | With a lead time | Median (quartiles) | Mean | Within 1 quarter | 2-4 | 5-8 | Already growing or falling |
|---|---|---|---|---|---|---|---|---|---|---|
| up | recruiters up | 152,075 | 58,851 | 61.5% | 3 (2 to 5) | 3.34 | 21.3% | 52.0% | 26.7% | 47.9% (47.6 to 48.3) |
| up | all eligible | 3,476,595 | 154,239 | 47.9% | 4 (2 to 5) | 3.89 | 12.2% | 51.7% | 36.1% | 33.7% (33.6 to 33.9) |
| up | momentum up | 153,074 | 61,667 | 76.4% | 2 (1 to 4) | 2.68 | 35.6% | 50.9% | 13.6% | 87.3% (87.1 to 87.6) |
| down | recruiters down | 98,611 | 49,328 | 32.8% | 3 (2 to 5) | 3.44 | 22.1% | 48.4% | 29.5% | 15.0% (14.7 to 15.3) |
| down | all eligible | 3,476,595 | 154,239 | 21.4% | 4 (2 to 6) | 4.06 | 13.2% | 46.2% | 40.6% | 7.3% (7.2 to 7.4) |
| down | momentum down | 96,709 | 47,405 | 61.6% | 2 (1 to 4) | 2.61 | 40.4% | 43.9% | 15.7% | 55.5% (55.1 to 56.0) |

The lead-time shares among those with a lead time (columns "within 1 quarter" to "5-8") answer the study's question of how early the growth comes; the median and quartiles describe the same distribution.

### 6.5 Event paths against matched companies

| j | -8 | -6 | -4 | -2 | 0 | 2 | 4 | 6 | 8 |
|---|---|---|---|---|---|---|---|---|---|
| recruiters up: excess, log points | -6.05 | -4.41 | -2.92 | -2.15 | 0.00 | 3.00 | 4.45 | 4.98 | 5.03 |
| recruiters up: companies with the change, % | -26.6 | -20.4 | -13.9 | -7.2 | 0.0 | 6.0 | 9.5 | 11.2 | 12.0 |
| recruiters up: matched companies, % | -22.0 | -16.8 | -11.4 | -5.2 | 0.0 | 2.8 | 4.7 | 5.8 | 6.5 |
| recruiters down: excess, log points | -0.75 | 1.03 | 2.37 | 3.13 | 0.00 | -3.10 | -5.19 | -6.46 | -7.51 |
| recruiters down: companies with the change, % | -15.5 | -9.2 | -3.5 | 0.7 | 0.0 | -1.7 | -3.0 | -4.0 | -5.0 |
| recruiters down: matched companies, % | -14.9 | -10.1 | -5.8 | -2.4 | 0.0 | 1.4 | 2.2 | 2.4 | 2.4 |

- recruiters up: 138,664 events, 138,654 matched at 56,650 companies (10 without a control). Excess at j = +4: 4.45 log points (4.33 to 4.58); at j = +8: 5.03. Pre-trend (excess change from j = -8 to j = -4): 3.13 (2.99 to 3.29). The pre-trend interval excludes 0, so the gap is not cleanly identified.
- recruiters down: 90,462 events, 90,461 matched at 47,364 companies (1 without a control). Excess at j = +4: -5.19 log points (-5.35 to -4.98); at j = +8: -7.51. Pre-trend (excess change from j = -8 to j = -4): 3.12 (2.94 to 3.31). The pre-trend interval excludes 0, so the gap is not cleanly identified.

### 6.6 Robustness

Regression, 4 quarters (log points per recruiter per 100 staff):

| Variant | Model A (95% interval) | Model B (95% interval) | B over A | Companies |
|---|---|---|---|---|
| Headline | 2.43 (2.29 to 2.56) | 1.76 (1.65 to 1.87) | 0.727 | 161,801 |
| (a) total headcount | 2.26 (2.13 to 2.39) | 1.36 (1.25 to 1.47) | 0.603 | 161,801 |
| (b) no contract or consulting titles | 2.52 (2.39 to 2.65) | 1.82 (1.71 to 1.92) | 0.721 | 161,801 |
| (c) signals to 2024 Q2 | 2.39 (2.25 to 2.53) | 1.76 (1.65 to 1.88) | 0.737 | 156,285 |
| (d) software and IT | 2.22 (1.95 to 2.49) | 1.33 (1.13 to 1.53) | 0.598 | 14,175 |
| (e) 50-199 people | 1.72 (1.61 to 1.82) | 1.32 (1.22 to 1.41) | 0.769 | 141,944 |
| (e) 200-999 people | 4.83 (4.45 to 5.20) | 3.00 (2.69 to 3.31) | 0.621 | 36,672 |
| (e) 1,000-4,999 people | 6.33 (3.56 to 9.11) | 3.40 (1.60 to 5.21) | 0.537 | 7,434 |
| (e) 5,000+ people | 9.90 (4.81 to 14.99) | 6.16 (2.63 to 9.68) | 0.622 | 1,278 |
| (i) log change of recruiters | 6.30 (6.16 to 6.44) | 4.65 (4.53 to 4.76) | 0.737 | 161,801 |
| (j) outcomes not winsorized | 2.73 (2.57 to 2.89) | 2.04 (1.91 to 2.17) | 0.747 | 161,801 |

The log-change variant (i) uses ln(1 + R(t)) - ln(1 + R(t - 1)) as the regressor, so its coefficients are on another scale. Per 100 staff, one recruiter is a larger change at a large company, which is why coefficients rise with size.

Signal scores by variant (recruiter rules and their momentum comparisons):

| Variant | Recruiters up: precision, lift, adjusted lift | Momentum up precision | Salespeople at same count precision | Recruiters down: precision, lift, adjusted lift | Momentum down precision |
|---|---|---|---|---|---|
| Headline | 43.1%, 1.53, 1.62 | 63.3% | 47.8% | 22.8%, 1.82, 1.99 | 50.6% |
| (a) total headcount | 42.7%, 1.52, 1.60 | 63.5% | 47.7% | 22.8%, 1.82, 1.98 | 50.8% |
| (b) no contract or consulting titles | 42.9%, 1.52, 1.62 | 63.5% | 48.0% | 22.8%, 1.82, 1.99 | 51.0% |
| (c) signals to 2024 Q2 | 44.1%, 1.49, 1.60 | 64.1% | 48.9% | 22.7%, 1.83, 2.00 | 50.0% |
| (d) software and IT | 58.9%, 1.72, 1.63 | 76.3% | 63.5% | 39.3%, 1.76, 1.74 | 69.6% |
| (e) 50-199 people | 51.9%, 1.68, 1.66 | 65.6% | 51.1% | 26.7%, 1.89, 1.86 | 55.4% |
| (e) 200-999 people | 37.3%, 1.65, 1.59 | 59.2% | 41.3% | 19.6%, 2.19, 2.16 | 42.7% |
| (e) 1,000-4,999 people | 23.2%, 1.50, 1.40 | 47.3% | 33.2% | 14.7%, 2.65, 2.63 | 31.4% |
| (e) 5,000+ people | 19.8%, 1.88, 1.68 | 37.9% | 30.8% | 14.7%, 4.50, 4.34 | 26.7% |
| (f) no recruiter before | 44.2%, 1.55, 1.60 | 61.1% | 48.4% | suppressed or not defined | n/a |
| (f) 1-2 recruiters before | 42.8%, 1.50, 1.55 | 64.9% | 48.7% | 22.9%, 1.79, 1.76 | 43.9% |
| (f) 3-9 recruiters before | 41.2%, 1.54, 1.48 | 65.3% | 48.5% | 21.8%, 1.95, 1.79 | 37.8% |
| (f) 10+ recruiters before | 51.2%, 2.24, 1.85 | 74.4% | 60.0% | 36.4%, 4.66, 3.67 | 51.4% |
| (g) 5% targets | 63.0%, 1.32, 1.33 | 75.1% | 64.1% | 35.0%, 1.43, 1.53 | 63.0% |
| (g) 20% targets | 22.2%, 2.07, 2.30 | 43.4% | 26.8% | 11.7%, 2.57, 2.81 | 31.4% |
| (h) rules over two quarters | 41.7%, 1.48, 1.58 | 59.8% | 43.8% | 22.8%, 1.82, 1.99 | 45.0% |
| (k) software and IT | 58.9%, 1.72, 1.63 | 76.3% | 63.5% | 39.3%, 1.76, 1.74 | 69.6% |
| (k) other technology, telecom and media | 45.0%, 1.78, 1.82 | 62.2% | 45.4% | 32.0%, 1.62, 1.74 | 58.1% |
| (k) financial services | 49.2%, 1.63, 1.70 | 72.5% | 54.2% | 24.6%, 2.33, 2.41 | 60.6% |
| (k) business and professional services | 45.2%, 1.46, 1.48 | 60.7% | 47.6% | 25.3%, 1.76, 1.89 | 53.5% |
| (k) manufacturing, industrial and energy | 39.3%, 1.54, 1.68 | 64.5% | 40.9% | 19.2%, 1.80, 1.97 | 51.8% |
| (k) healthcare and life sciences | 58.7%, 1.71, 1.71 | 79.3% | 57.0% | 34.6%, 1.90, 1.92 | 73.6% |
| (k) consumer, public sector and the rest | 37.6%, 1.38, 1.51 | 57.7% | 43.2% | 16.6%, 1.54, 1.74 | 40.2% |

## 7. Hypotheses, as pre-registered

| Hypothesis | Test | Result | Holds |
|---|---|---|---|
| H1 | Model A b_4 > 0 and its 95% interval excludes 0 | b_4 = 2.4253 (2.2913 to 2.5593) | yes |
| H2 | largest coefficient among j = 0-4 at j = 0 or 1; j = 3 and j = 4 each below half of it | peak j = 0; j=3 0.311, j=4 0.276 of the peak | yes |
| H3 | Model B b_4 > 0, interval excludes 0, and at most half of Model A's b_4 | b_4 = 1.7627 (1.6543 to 1.8710); 0.727 of Model A (wrong: Model B keeps more than half of Model A's coefficient, so momentum explains less of the link than expected and recruiters carry more information of their own) | no |
| H4 | lift >= 1.3, adjusted lift >= 1.2, momentum precision >= recruiters-up precision, recruiters up raises the hit rate in both momentum groups | lift 1.5301, adjusted 1.6188; precision 0.4305 vs momentum 0.6335; difference within momentum fired 0.1852, not fired 0.1221 | yes |
| H5 | median lead time <= 2 quarters and already-growing share above all eligible | median 3; already growing 0.4793 vs 0.3374 (wrong on the median (3 quarters); the already-growing part holds) | no |
| H6 | lift >= 1.5, adjusted lift >= 1.3, precision < 30% | lift 1.8243, adjusted 1.9889, precision 0.2284 | yes |

## 8. Verification

An independent re-analysis, written from the study plan without seeing our code or working notes, got only the question, study brief, hypothesis, plan and pre-result logged decisions (D1-D13) and the extract, and recomputed 152 numbers with its own code, including a recount of the recruiter column from the person-level spells (its recruiter count equals the extract's in all 7,281,045 company-quarter cells, and its title classes reproduce the audit counts exactly).

- First comparison: 142 of 150 comparable numbers matched at the reported precision; 8 did not (six timing-profile coefficients and two lead-time shares of the momentum-down group). The study's timing profile had stopped one quarter short of the plan (q 37 instead of q 38), and its lead-time momentum groups had been re-ranked inside the shorter window instead of taken from the signal-scoring rules, as the plan reads. Both were fixed and the two steps re-run.
- After the fix: 150 of 150 match at the reported precision and 0 differ by more than one part in a million; the bootstrap interval ends agree within a quarter of the interval width. No hypothesis verdict changed.
- The verifier also measured how much the conventions matter: the precision of "most salespeople added" lies between 0.4703 and 0.4869 under any order of the tied company-quarters at its cut-off (momentum up: 0.6328 to 0.6339), and drawing the bootstrap weights in another layout moves interval ends by at most 0.0038.

Match rules were set before the verifier's numbers were read: counts equal; point estimates equal at the precision shown; interval ends within a quarter of the interval width or 0.01.

## 9. Alternative explanations and limitations

1. **Recruiters follow growth that has already started.** This is the study's main finding, not a confounder to remove: growth in the quarters before a recruiter change is also linked with it (timing profile, j = -1: 0.71), 47.9% of recruiter increases follow a year of 10%+ growth against 33.7% of all company-quarters, and companies adding recruiters had outgrown their matched companies before the matching window (pre-trend 3.13 log points). The recruiter signal is reported as coincident with growth, not as a lead.
2. **Recruiters are part of headcount.** Growth is measured without them (N = H - R). With total headcount instead, Model A's 4-quarter coefficient is 2.26 and the recruiters-up lift 1.52.
3. **It is only momentum.** Controls for the company's own growth in the signal quarter and the quarter before, and its change in salespeople, keep 72.7% of the coefficient; within both momentum groups the recruiter rule raises the hit rate. Momentum is still the better single rule in every variant.
4. **Growing companies simply have more recruiters.** Company effects absorb each company's average growth and quarter effects absorb booms and busts such as 2021-2022; the adjusted lift compares each signal with the same quarter and size band (1.62).
5. **Jobs added to profiles late.** With signals up to 2024 Q2, so that every outcome window ends by 2025 Q2, Model A gives 2.39 and the lift 1.49. Series end at 2026 Q1 and every scored signal has its full window.
6. **Agency recruiters and contractors.** They are not on the company's books, and staffing firms are left out of the sample. Dropping recruiter titles that name contract, freelance, agency or consulting work gives Model A 2.52 and a lift of 1.52. Companies that hire only through agencies show no recruiter signal at all; that is a limit of what this measure can see.
7. **Title errors.** Rule (d) was fixed by an audit of every title held by 50 or more people before any outcome was computed. HR staff rising by the same rule have a lift of only 1.24, so recruiters carry more of the signal than HR hiring in general.
8. **One recruiter is noise.** Lift by recruiters before: 1.55 for a first recruiter, 1.50 with 1-2 before, 1.54 with 3-9 and 2.24 with 10 or more; over two quarters instead of one, 1.48.
9. **Industry and size mix.** Across the seven industry groups and four size bands the lift runs from 1.38 to 1.88, and momentum has the higher precision in each.
10. **Staff outside the six countries.** Companies are headquartered in the six countries, but staff they have elsewhere are missing, so a company growing abroad looks flat and its recruiters abroad are not counted. This adds noise and can bias the level of every score.
11. **Missing and self-reported profiles.** Only currently active profiles exist, so older years undercount and short jobs are often missing; people edit old titles after a promotion. Company and quarter effects, and comparisons within the same quarter, limit how this moves the comparisons, but the growth rates themselves are not comparable with official statistics.
12. **Cause.** Nothing here shows that hiring recruiters causes growth. The design is a panel association and a matched comparison whose pre-trend check failed.

## 10. Sources and reuse

- Kenso internal data (Kenso's dataset of public professional profiles, work histories and company records), through Kenso's shared headcount history extract (built for the SDR study, build of 2026-10-17) and the shared sales title taxonomy of the GTM engineer study.
- The Kenso Research study *Who gets cut first in a B2B layoff? Recruiters, then SDRs*: recruiter title rule, industry groups, size bands; its finding that recruiting had the highest raw cut index in 2022-2024 software and IT contractions.
- The Kenso Research study *Sales and marketing draw level with engineering at about 50 people*: sales label group, tie-break key. The Kenso Research study *The first 20 hires: engineers first, a seller by hire 7* (its logged decision D3): reuse of the headcount-history spells. The Kenso Research study *Did AI kill the SDR role?*: software and IT labels; reporting lag.