Headcount is the most accessible signal for private company size. It is visible in filing disclosures, deducible from job advertisement density, and occasionally disclosed directly on company websites. The problem is that it is a poor proxy for revenue unless it is corrected for the labour composition of the business.
A 50-person cleaning services company and a 50-person technical software testing firm are not similar-sized businesses in any commercially meaningful sense. The first might generate 2.5 million GBP in annual revenue; the second, 6 to 9 million GBP. Using raw headcount to rank them would misorder their positions in any target list. This note describes the correction approach we use and where the method reaches its limits.
Why revenue per employee varies so widely
Revenue per employee is principally a function of three things: capital intensity, labour type mix, and business model structure. A capital-intensive industrial manufacturer with significant plant and machinery can generate substantial revenue from relatively few production workers because the machine does most of the revenue-generating work. A labour-intensive field services firm needs many more hands to produce the same revenue because the output is time measured in person-hours. A professional services firm sits between the two, but its revenue per employee varies further depending on whether it primarily employs junior staff or senior specialists.
None of these distinctions are visible in a raw headcount figure. To use headcount as a meaningful size proxy, you need to establish which labour model the company operates, and that requires a different input source than the one that gave you the headcount number in the first place.
Labour-type decomposition as a correction variable
The most reliable public signal for labour type composition is the job advertisement record. When a company is actively hiring, the titles and seniority mix of its open roles describe the composition of the workforce it is trying to maintain or grow. A company that consistently advertises for drivers, warehouse operatives, and site supervisors has a very different labour model than one advertising primarily for software engineers and product managers.
Job advertisement data has known limitations for this purpose. It only reflects the hiring activity that companies make public, which systematically underrepresents businesses that recruit through word of mouth, agencies, or direct contact. It is episodic rather than continuous, so a company that last advertised publicly three years ago will appear to have no hiring signal even if its workforce has grown. And it captures the margin of the workforce, not the total stock.
Despite those limitations, the job category distribution across advertisements over a multi-year window is a stable signal for labour model type. A firm that has advertised operational roles at a four-to-one ratio relative to commercial roles over several years is reliably different in its labour composition from one that has consistently advertised the reverse. The ratio is more informative than the absolute volume.
Mapping role categories to revenue multipliers
Our approach to headcount correction uses a set of sector-specific role-category calibrations. For each sector, we define the expected revenue per employee for three labour model archetypes: high-operational (workforce dominated by field, production, and service delivery roles), balanced (mix of operational, technical, and commercial), and high-knowledge (workforce dominated by professional, technical, and commercial roles).
These calibrations are derived from the subset of companies in each sector that file detailed accounts, where we can observe both headcount (from the directors' report or note disclosures) and revenue directly. That disclosed subset is smaller than the total company population but large enough to establish stable calibration ranges. We then apply those ranges to companies where only headcount can be inferred and direct revenue observation is not possible.
The output is not a precise revenue figure. It is a probability-weighted band, typically with a range of two to three times. A company assessed as 50 employees with a high-operational labour model in a cleaning services sub-sector might be assigned a band of 2 to 5 million GBP. That band is wide, but it is more useful than a single point estimate that ignores the uncertainty, and it correctly signals that the company is probably not in the 10 to 20 million GBP range that a particular deal mandate might specify.
Where the method is weakest
Group structures are the primary failure mode. A parent company may have a small UK registered headcount because most of its operational workforce is employed in subsidiaries. The filing record for the parent will show the parent's directly employed staff only, not the consolidated group headcount. If the subsidiaries also file as UK companies, the headcount may appear multiple times in the registry, once per entity. Neither the understated parent figure nor the duplicated subsidiary figures are accurate representations of the operating workforce.
We handle this partially through entity resolution: identifying which companies in a filing set share directors, addresses, or group parent designations, and consolidating their headcount signals where the evidence supports it. But for complex group structures with foreign-domiciled parents, the consolidation is incomplete. A company that looks like a 15-person UK operations office in the filing record may be the national face of a much larger organisation. We flag these cases in the output rather than assigning an estimated band, because the uncertainty is too large for the band to be reliable.
Seasonal businesses are a second limitation. A tourism or hospitality operator may have a permanent workforce of 30 and a peak seasonal workforce of 150. The filing disclosure captures the employee count at the year-end date, which may land at any point in the seasonal cycle. The job advertisement record may capture peak-season hiring volumes that inflate the apparent workforce size. Without knowing where the year-end falls relative to the seasonal peak, headcount-based size estimation for seasonal businesses carries substantial additional error.
Staffing cost disclosures as a cross-check
Where companies file detailed accounts rather than abbreviated ones, the notes often include a staff cost disclosure: the total wages and salaries charged in the period, sometimes broken down by category. This is a more direct input for revenue estimation than headcount alone, because it captures both the number of employees and the average compensation level, which correlates with seniority and therefore with revenue-generating capacity per person.
A company disclosing 1.8 million GBP in staff costs across a workforce of approximately 40 is paying an average of around 45,000 GBP per employee. That compensation level is consistent with a mixed-skill workforce in a field services context. Combined with sector benchmarks for the ratio of staff costs to revenue in that activity type, you can derive a revenue band estimate from the wages figure alone, without needing to use the headcount directly.
This approach is only available for companies that file detailed accounts, which is a minority of small company filers. But where it is available, it provides a materially more precise band estimate than the headcount correction alone, and serves as a useful cross-check when the two methods yield different bands for the same company.
What this means for target list quality
The practical effect of applying labour-type correction to headcount estimates is that the target list reorders meaningfully relative to a raw headcount sort. Companies with high-knowledge workforces move up; companies with high-operational workforces that look similar in raw headcount move down or are flagged as below-threshold for a particular revenue mandate.
Not all of these reorderings are correct. The calibrations are averages across a sector, and individual companies can deviate significantly from the sector average. A cleaning services company that has expanded into high-margin specialist decontamination work will have a higher revenue per employee than the sector average suggests. We do not claim to eliminate this error; we claim to reduce it. The output is more reliable than a raw headcount ranking, not perfect.
What we do not do is present a single point estimate as if it were precise. The band format in the output is intentional: it communicates the uncertainty honestly, and it is the right input for an analyst who needs to decide whether a company is worth a call to find out more, rather than for one who needs to model it to the nearest ten thousand pounds.