Before You Hire a Head of Data: The Team Structures Mid-Market Companies Actually Need
- Matt Lazarus
- 11 minutes ago
- 5 min read
The job ad is probably drafted already: Head of Data, six-figure package, a mandate covering everything from pipelines to AI strategy. It is one of the most expensive documents a mid-market company writes - because if the sequencing is wrong, the hire fails regardless of who fills it.
The classic versions are painful to watch: a senior data scientist hired into a company with no pipelines to work on, or a capable analyst quietly expected to architect an enterprise estate.
The fix is not a better candidate. It is a capability model that separates the roles your stage needs from the headcount you can justify.
Key Takeaways
Sequence capabilities, not titles: engineering first, meaning second, advanced AI value third.
Roles are not headcount: hybrid patterns - fractional seniors guiding internal staff - cover the stack without the payroll.
Each hire should land on a platform that exists: the maturity signals tell you when each transition is due.
What Capabilities Does a Data Function Actually Need - in What Order?
Three layers, in strict sequence: engineering and architecture build the foundations (pipelines, platforms, integration); analytics and semantic modelling create the meaning (governed metrics, reporting, self-service); and science and AI specialisation extract advanced value (forecasting, agents, machine learning). Each layer only performs on top of the one beneath it.
The sequence explains both classic mis-hires instantly. The data scientist failed because layer three was hired before layers one and two existed - brilliant skills with nothing to stand on. The overstretched analyst failed in the other direction - a layer-two professional carrying layer-one responsibilities no one had budgeted for.
The order is not ideology; it is dependency. Models need features, features need pipelines, pipelines need architecture. Skipping a layer does not skip the work - it just assigns it to someone unqualified for it, at the worst possible moment.
What Are the Maturity Signals for Each Transition?
You are ready for the next layer when the current one stops being the bottleneck. Foundations are done enough when data lands reliably, integrated and trusted, without heroics. Meaning is done enough when the organisation argues about decisions rather than numbers. Advanced value is due when governed data and settled definitions are waiting for harder questions than reporting asks.
The signals, made concrete:
One signal outranks the rest when budgets force a choice: trusted numbers. An organisation whose leadership argues about decisions rather than figures has banked the asset every later stage spends - and an organisation still arguing about figures should spend nothing on stage three until that argument ends.
Stage one complete: refreshes run unattended, new sources onboard in days, and no single person's resignation threatens the estate.
Stage two complete: metrics have one certified definition, leadership trusts the dashboards, and self-service works without chaos.
Stage three justified: use cases exist with measurable returns, the data slices they need are governed, and the organisation can evaluate AI output rather than admire it.
Most mid-market companies that feel "behind on AI" are actually mid-stage-two - which is excellent news, because stage two is cheaper than the AI ambitions waiting behind it.
Build, Rent or Hybrid - What Do the Economics Say?
Honestly stated: the layers need senior expertise intermittently and operational capacity continuously - which is the exact inverse of what a single full-time senior hire provides. A principal-level architect is essential for weeks at each stage transition and underused between them; an internal analyst is needed every day and cannot architect alone. The economics favour hybrids at mid-market scale.
The three honest structures:
Pure build: justified once data work is continuous across all three layers - usually later than founders expect, and always after the platform exists for hires to land on.
Pure rent: fine for bounded projects; risky as a permanent posture, because context and continuity live outside the business.
The hybrid most actually run: fractional senior architecture and engineering setting the foundations and standards, with internal analysts operating and extending them - capability compounding in-house while the scarce skills arrive on demand.
That hybrid is precisely the model behind a fractional data and AI team: principal-level capability across all three layers, scaled to the stage you are actually at, with the explicit goal of building your internal bench rather than replacing it.
What Should Happen Before Any Senior Hire?
Baseline the estate. A structured assessment of your platforms, pipelines, definitions and governance tells you which stage you are genuinely at, what the first ninety days of any senior hire would have to fix, and whether that work is better completed before the salary starts. Hires succeed when they land on a known platform with a sequenced roadmap - and that artefact is exactly what an AI Data Readiness Audit produces.
It also sharpens the job ad immeasurably: a mandate written against a scored baseline attracts the right seniority and repels the wrong promises - on both sides of the interview table.
What Should the First 90 Days of the Hybrid Model Look Like?
Ninety days is enough for the hybrid to prove itself: foundations stabilised, one visible reporting win shipped, and your internal analyst measurably more capable than they started. If those three are not landing by day ninety, the engagement is drifting - and a good fractional team will say so first.
Days 1-30 - baseline and stabilise: the estate documented, refresh failures and fragile pipelines triaged, and the quality register started. Unglamorous by design; this is the platform every later win stands on.
Days 31-60 - ship the visible win: one reporting or definitions problem leadership personally feels - the month-end pack, the revenue reconciliation - fixed end to end, with your analyst pairing on every step rather than observing.
Days 61-90 - codify and transfer: standards written down, the stewardship rhythm running, and the roadmap for the next two quarters sequenced by the maturity signals - with explicit markers for which capabilities move in-house, when.
The accountability artefact that keeps hybrids honest is the capability transfer plan: a one-page register of what your team can do unaided today versus ninety days ago. Senior talent that resists writing it is selling dependence; talent that volunteers it is building your bench - which is the entire point of renting seniority in the first place.
Measured that way, the hybrid is not a compromise between hiring and outsourcing. It is the sequencing tool that makes the eventual hire land on a platform, a roadmap and a team that already works.
What If You Have Already Made the Wrong Hire?
Salvage before you restart. A data scientist hired before the pipelines existed is rarely a bad hire - just a mis-sequenced one. Pair them with fractional senior architecture to build the foundations they need, and redirect their craft toward the analytical groundwork every later model depends on: definitions, evaluation sets and the first governed data slices.
The honest conversation is about sequence, not capability - most mis-hired specialists know exactly what is missing underneath them and are relieved when someone finally funds it. Six months of foundation-building with senior support typically converts a frustrated hire into the future leader of the function. Replacing them and repeating the sequencing mistake converts them into a competitor's head start.
Hire Into a System, Not a Vacuum
The Head of Data you eventually hire will be excellent or wasted depending almost entirely on what exists before their first day. Sequence the capabilities, run the hybrid until the maturity signals fire, and write the job ad against a baseline rather than a hope.
The right person, hired at the right stage, onto a platform that exists - that is the entire trick. Everything else is recruitment marketing.
