The eleven stages
Stages one and two happen before any fee is agreed.
Define the decision, not the role
- What we establish
- The decision the research supports, who takes it, what evidence would change it, and what is currently believed but unverified.
- Why it matters
- Research scoped to a job title answers a smaller question than the one you actually face.
Agree the market definition in writing
- What we establish
- Geography, employer set, levels, capability definition, language requirement, and the explicit exclusions.
- Why it matters
- Every number depends on this. Written and signed off before fieldwork, so no figure is later contested on scope.
Frame from official sources
- What we establish
- The statistical frame — national labour market data, company registers, regulatory registers and published filings.
- Why it matters
- It gives a defensible denominator and shows where the public record is blind.
Build the employer universe
- What we establish
- Every organisation that could hold the capability, including mid-sized and privately held employers with no brand recognition.
- Why it matters
- In markets like the German Mittelstand or Dutch high-tech supply chain, most of the capability sits outside the names you would think of.
Define capability criteria, not titles
- What we establish
- The requirement translated into observable, testable attributes.
- Why it matters
- Job titles are not comparable across employers or borders. Capability is.
Research the market directly
- What we establish
- Named employers, structures, populations and movement, researched by analysts rather than queried from a database.
- Why it matters
- This is the stage that produces the value, and the one a licence cannot replicate.
Verify and corroborate
- What we establish
- Each material finding checked against an independent second source.
- Why it matters
- A claim corroborated twice is evidence. A claim repeated twice is still a claim.
Reconstruct compensation
- What we establish
- Pay rebuilt into its local components — on-costs, equity, allowances, collective agreement position, holiday allowance, superannuation — with the basis of every figure stated.
- Why it matters
- Cross-border pay comparison without reconstruction is the single most common error in this field.
Grade confidence
- What we establish
- A confidence level and its basis attached to each material finding.
- Why it matters
- Findings get quoted out of context. The confidence level should travel with the number.
Analyse against the decision
- What we establish
- What the evidence supports, what it rules out, what it cannot tell you, and what we would do next in your position.
- Why it matters
- Data is not intelligence until someone has interpreted it against a question.
Deliver, walk through and hand over
- What we establish
- Written analysis, structured dataset, methodology appendix and a working session with the lead analyst.
- Why it matters
- You own all of it permanently, including the right to refresh it with someone else.
The source hierarchy
Three tiers. A report that stops at tier two is describing the past.
| Tier | What it is | What it gives | What it cannot give |
|---|---|---|---|
| Tier 1 | Official statistics, company and regulatory registers, mandatory filings | An authoritative, citable frame and a defensible denominator | A current picture; coverage of smaller employers; anything role-specific |
| Tier 2 | Published professional information, disclosures, postings, publications | Breadth, structure and direction of travel | Verification; self-reported data is uneven and often stale |
| Tier 3 | Primary research into named employers and populations | What is actually true now, corroborated | Infinite breadth — it is slow and has to be scoped deliberately |
The practical consequence is that breadth and verification trade against each other. A product that covers every market cheaply is built on tiers one and two. A finding you can put in front of an investment committee needs tier three.
The question worth asking any provider, including us, is which tier a specific number came from and when it was collected. If that cannot be answered per figure, the methodology is not what it claims.
The verification standard
Every material finding is corroborated against an independent second source. A job title from one source is a claim; the same title corroborated by a company disclosure, a regulatory register or a second independent record is evidence.
Independence is the operative word. Two sources that both derive from the same original feed are one source wearing two hats, and a great deal of aggregated workforce data has exactly that problem. We treat derived sources as a single source and say so.
Where corroboration is not possible — which happens routinely with privately held mid-sized employers — the finding is reported with its confidence level and the reason. We would rather hand over a smaller population we can stand behind than a larger one we cannot.
How confidence is graded
Three levels, recorded per material finding.
| Level | What it means | How it should be used |
|---|---|---|
| Corroborated | Two or more independent sources agree, and no credible contradicting evidence was found | Safe to quote in a board paper or investment committee |
| Single-sourced | One credible source supports it, nothing contradicts it, corroboration was not available | Usable as directional evidence; say so when quoting it |
| Estimated | Modelled rather than observed, with the model and its assumptions stated | Use for sizing and scenario work, never as a point fact |
If a provider will not tell you which of these three a number is, treat the number as estimated. That is not cynicism; it is the statistically correct default.
Lawful basis by market
Scoped per jurisdiction, documented before work begins.
| Market | Regime | Basis we rely on | Regulator |
|---|---|---|---|
| United Kingdom | UK GDPR and Data Protection Act 2018 | Legitimate interests, with a documented balancing test | Information Commissioner’s Office |
| Germany, Netherlands, Belgium, Austria | GDPR plus national law | Legitimate interests, with a documented balancing test | National and, in Germany, state authorities |
| United States | State privacy laws, principally the CCPA as amended | Scoped to the jurisdictions in play | State attorneys general and the California Privacy Protection Agency |
| United Arab Emirates | Federal PDPL; DIFC and ADGM run their own regimes | Scoped to the entity type involved | Federal and free zone regulators |
| Australia | Privacy Act 1988 and the Australian Privacy Principles | Applied alongside our own global policy, whichever is stricter | Office of the Australian Information Commissioner |
Where a client operates across several of these, we default to the strictest applicable standard across the whole dataset rather than maintaining two levels of rigour. It keeps a single deliverable usable everywhere, and it is simpler to audit.
See also compliance support for the wider documentation and audit services we provide alongside research.
What we will not do
Published because it is a commercial position, not only an ethical one.
- Misrepresent who we are. We do not call a competitor’s staff under a false identity or a false pretext to extract information.
- Solicit confidential information. We do not ask current or former employees to disclose information they are contractually bound to protect.
- Compile special-category data on individuals. No health, ethnicity, belief, trade union or sexual orientation data against named people, in any market.
- Infer protected characteristics. Not from names, photographs, institutions or any other proxy. It is unlawful as a basis for individual records and unreliable as evidence.
- Report a population so small it identifies individuals. We raise the threshold, suppress the cell, or tell you it cannot be produced responsibly.
- Present an estimate as a fact. Modelled figures are labelled as modelled, with the assumptions shown.
- Resell your research. Client datasets are handed over and are not recycled into anything we sell to anyone else.
- Take an engagement we cannot deliver. If the question is not answerable from observable evidence, we say so at scoping rather than at delivery.
Anatomy of a deliverable
Written analysis
Fifteen to forty pages. Market definition, method, findings, what the evidence rules out, limitations, and a recommended next step.
Structured dataset
Every record behind the analysis, with source and collection date per material field, in a format your systems can ingest.
Confidence register
Each material finding with its grade and the basis for it, so numbers travel with their certainty attached.
Methodology appendix
Sources, lawful basis per jurisdiction, exclusions and known limitations — written for a privacy reviewer as much as for you.
Analyst session
A working session with the lead analyst who built it, not an account manager.
Handover
Permanent ownership of everything above, including the right to refresh it with another provider.
Decay and refresh
Talent intelligence decays at different rates depending on what it describes. Compensation data in a fast-moving function can be materially out of date within two quarters. Individual records decay with every job move in the population. Structural findings — how a competitor is organised, where a capability is concentrated — usually hold for a year or more.
Because every field carries its collection date, you can see precisely what has aged rather than refreshing everything or trusting all of it. That is the practical argument for recording dates per field rather than stamping a report with a single publication date.
Refresh is priced separately and never bundled into a first project by default. A retained programme with quarterly refreshes suits organisations making continuous decisions; a one-off brief with an annual re-run suits most others.
Quality review
Every deliverable passes a review before it leaves, conducted by someone other than the analyst who produced it. The reviewer checks three things: that every material figure traces to a source and a date, that the confidence grades are justified by what is actually in the record, and that the analysis answers the question in the signed scope rather than an adjacent one.
Findings that fail the first test are removed rather than downgraded. A number nobody can trace has no place in a deliverable you are going to rely on, however interesting it is.
Methodology FAQs
What methodology do you use for talent intelligence?
Eleven stages, from defining the decision through to handover, with a written scope agreed before any fieldwork begins. Data is drawn from three tiers — official statistics and registers, published professional information, and primary research — and every material finding is corroborated against a second independent source or carries a recorded confidence level explaining why it could not be.
The full stage-by-stage method is set out on this page, along with the source hierarchy, the verification standard and the lawful basis applied in each market.
How do you verify the data?
By corroboration against an independent second source wherever the finding is material. A job title taken from one source is a claim; the same title corroborated by a company disclosure, a regulatory register or a second independent record is evidence.
Where corroboration is not possible — which happens, particularly with privately held mid-sized employers — the finding is still reported, but labelled as single-sourced with the basis stated. We do not quietly upgrade a claim into a fact.
What is confidence grading?
A stated level of certainty attached to each material finding, with the reason for it recorded. We use three levels: corroborated, where two or more independent sources agree; single-sourced, where one credible source supports it and no contradicting evidence was found; and estimated, where a figure is modelled, with the model and its assumptions stated.
It matters because findings get quoted out of context. A number that travels into a board paper should carry its confidence level with it.
How current is the data?
Every material field carries the date it was collected. Primary research findings are collected during the engagement, so they are current as at delivery. Official statistics carry the publication date of the release used, which in some markets lags by a year or more.
We never present a lagged statistic as a current one. Where the only available official figure is dated, the report says so and says what primary research found instead.
What is your lawful basis for processing personal data?
In the UK and EU, legitimate interests under GDPR, supported by a documented balancing test rather than individual consent. In the UAE, the federal PDPL or the DIFC and ADGM regimes depending on the entity. In Australia, the Privacy Act and the Australian Privacy Principles. In the US, the applicable state regimes, principally the CCPA as amended.
Across all of them we collect professional information only, record the source per field and transfer deliverables securely. Your privacy function can review the methodology before work begins.
Do you collect special-category or sensitive personal data?
No. We do not compile health, ethnicity, religious or political belief, trade union membership or sexual orientation data against named individuals, in any market.
Diversity and representation work is produced at aggregate population level only, from sources appropriate to that purpose. We also do not infer characteristics from names, photographs or institutions, which is both unlawful as a basis for individual records and unreliable as evidence.
How do you approach competitor research ethically?
We research publicly observable professional information and corroborate it. We do not misrepresent who we are or who we are working for in order to obtain information, we do not approach a competitor’s staff under a pretext, and we do not solicit confidential information from current or former employees.
This is a commercial position as much as an ethical one. A firm willing to do those things for you will do them about you, and the resulting dataset is a liability attached to your name rather than an asset.
What happens if you cannot answer the question?
We say so, as early as we know it. Occasionally a market is too opaque, a population too small to report without identifying individuals, or a question not answerable from observable evidence at all.
Where that becomes clear during scoping, we do not take the engagement. Where it becomes clear mid-project, we say what we can establish, what we cannot, and we adjust the scope and the fee rather than padding a deliverable to fill the gap.
How often should a dataset be refreshed?
Quarterly for compensation and supply in fast-moving functions; annually for structural findings such as how a competitor is organised or where capability is concentrated.
Because every field carries its collection date, you can see precisely what has aged rather than refreshing everything or trusting all of it. Refresh is priced separately and never bundled into a first project by default.
Who actually does the research?
Analysts, not a database. Each engagement has a named lead analyst who scopes it, runs the primary research and presents the findings, supported by a research team and a quality review before delivery.
The handover session is with that analyst rather than an account manager, because questions about method and edge cases need answering by the person who collected the data.
Can our legal or privacy team review the methodology first?
Yes, and we encourage it. We provide the methodology, the lawful basis applied in each jurisdiction, the categories of data collected and the security arrangements in writing before work begins.
Where a client has a works council or employee representative body — common in Germany, Austria, the Netherlands and Belgium — we recommend briefing it at the start of any study that could lead to a structural decision.
How do you handle small populations?
With care, and sometimes by declining. In a senior or specialist market a population may be small enough that aggregate reporting could identify individuals, which is both a privacy problem and a commercial one.
Where that risk exists we raise the reporting threshold, suppress cells that would be identifying, or tell you that the analysis you want cannot be produced responsibly at that population size.