How We Benchmark Company Footprint Data

The first step in any benchmark is defining each firmographic metric clearly.

Take employee count. Is a contractor an employee? What about temporary staff in a seasonal tourism business? Should part-time employees count? Should the count apply to the legal entity, the operating brand, the parent company, or the whole corporate group?

Revenue has the same problem. Which year are we using? Is it the last completed financial year, current run-rate, gross revenue, net revenue, marketplace GMV, or recognised accounting revenue? For an insurance business, are we looking at total revenue or net premium? For a marketplace, are we measuring transaction volume or reported accounting revenue?

For Footprint benchmarks, we use definitions from financial reporting, accounting, and compliance wherever possible. These tend to be the most heavily vetted definitions available. They are not perfect for every commercial use case, but they give us a consistent baseline for fair comparison.

A benchmark is only useful if the test set reflects the companies customers actually care about.

It is easy to benchmark on large public companies. They have annual reports, audited filings, investor presentations, and a lot of available information. But that is not what most CRMs look like.

Real account universes contain private companies, subsidiaries, local businesses, franchises, marketplaces, holding companies, regional offices, public companies, government-linked entities, and businesses with limited public information.

Footprint benchmark datasets contain thousands of companies, human-rated and stratified across industry, geography, company size, ownership structure, and source availability.

Once the metric is defined and the sample has been built, the hardest question remains: what is the true value?

That depends on two things: source quality and entity definition.

Different people could reasonably draw the boundary differently depending on the question they are trying to answer.

For Footprint benchmarks, we only include cases where there is high confidence in both the source and the entity definition. If the answer is too ambiguous, it is not a good benchmark case. Benchmarks should not pretend uncertainty does not exist. They should control for it.

Once the benchmark set is defined, each company is run through the providers and models being evaluated.

Footprint-Accuracy measures whether the number is meaningfully wrong.

For metrics like employee count and revenue, we use a geometric threshold. A provider passes if its value is within 2x of the human-rated reference value.

If the reference value is 100 employees, then 50 to 200 employees is considered acceptable. Below 50 or above 200 fails.

This may sound generous, but it reflects how account data is usually used.

If a public company reports $1.5B in revenue and a provider returns $1.2B, that is technically wrong. But for most GTM workflows, it probably does not matter. The account is still in the same broad segment. It would likely be routed the same way, assigned to the same team, and included in the same TAM analysis.

Footprint-Account-Impact measures whether the error would change decisions based on a segmentation base model.

Most GTM systems do not use raw employee and revenue values directly. They use bands, thresholds, and segments. For example, a company with $125M in revenue and a company with $450M in revenue may both be treated as the same segment.

So we ask whether the provider’s error would change the decision using common segmentation rules based on revenue, location, size, and industry.

This is why providers usually score better on Footprint-Account-Impact than on Footprint-Accuracy. The benchmark is more forgiving because real-world motions are more forgiving. Not every numerical error changes a business decision unless it is an order of magnitude error.

The value of this approach is that benchmark results translate directly into operational impact.

If an enterprise has 500,000 accounts in its CRM, then a 1% improvement in the Account Impact benchmark means 5,000 accounts would receive a different GTM motion, segmentation, routing decision, or TAM assessment.

That is the standard we care about: how many decisions get better?

This makes benchmark results easier to interpret. Instead of asking whether one provider is abstractly more accurate, teams can ask how many accounts would be treated differently and whether those differences improve commercial execution.

Footprint findings are also cross-validated against OECD and national-level datasets where possible. This helps answer a different question: does the dataset resemble the actual composition of the economy?

In our experience, most company datasets do not come close. They often overrepresent certain geographies, industries, company sizes, and types of firms. They may look comprehensive, but when compared with national or international economic datasets, the distribution can be far from reality.

That is a strong signal that there is still room for improvement for the industry in general. Account-level benchmarking tells us whether a provider is right about specific companies. Macro-level validation tells us whether the dataset reflects the real structure of the market.