Strategic skills gap analysis

A strategic skills gap is only usable if it names two things: which skills file it compared, and which roadmap vintage it compared against. If either is missing, the gap stays empty. Talent strategy still chooses build versus buy. The analysis does not open a requisition and it does not invent a market skill percent.

That is the quality bar. Everything else on this page is how to reach it without dressing a missing input as a finding.

Embeddings can cluster current employee skills against roadmap requirements and against external market language. Clustering is a comparison. A comparison without dated sources is a slide, not a gap.

Start by loading the current skills file, not a live org chart and not a manager's memory of who is strong on data. The skills file is the extract you will cite: inferred skills from work, assessed skills, or a certified inventory, with a file date and a coverage note for who is in and who is out. If you do not yet have that extract, stop here and run skills inference from work artifacts first. Inferring inside the gap workflow mixes two jobs and makes the vintage impossible to defend.

Load the product or capability roadmap next, as a dated artifact. The platform roadmap is not a vintage. Name the roadmap, the fiscal window, and the sign-off date. Capture the time horizon the roadmap actually covers. A one-year skills cluster against a three-year architecture story will look like a crisis the business has not yet funded.

Do not cluster until both loads succeed. If the skills file is stale, incomplete for a function, or unlabeled, leave those people or skills blank rather than filling from last year's HRIS dump. If the roadmap is a deck without a date, a verbal direction of travel, or a mix of funded and unfunded bets with no flag, leave the gap empty for that bet. Empty is the correct output. A filled cell that cannot cite both sources is the failure.

Talent systems in the same class as Workday, Eightfold, Gloat, and LinkedIn Recruiter often already hold a skills taxonomy, a profile, or a market view. Treat them as sources you may pull from, not as the gap itself. Pull, date, and cite. Do not treat a vendor's default skill graph as your roadmap, and do not let an undated export stand in for the skills file.

Once both sources are loaded, cluster current skills against the capabilities the roadmap names. The useful output is a short list of capability gaps, each with the capability or skill cluster from the roadmap, the matching or unmatched clusters in the skills file, a citation that names the skills file identifier and date plus the roadmap identifier and vintage, a blank where match quality is too weak to claim, and an explicit non-decision: this row is not a hire, a transfer, or a course assignment.

Talent strategy then chooses build versus buy. Build can mean targeted learning, rotational exposure, or a redesign of the role. Buy can mean a hire, a contractor, an acquisition of a team, or a vendor for that capability. The analysis does not rank those options. It does not auto-hire. It does not write a requisition identifier into the gap row.

Here is the shape of the work, not a result. A payments product line publishes a dated first-half roadmap that names real-time reconciliation on a new ledger as a funded capability. The skills file from the prior month clusters current engineers and operations staff around batch reconciliation, SQL, and existing ledger APIs. The embedding neighborhood shows little overlap with the real-time and new-ledger language on the roadmap. The gap row cites both artifacts and stops. It does not say to hire streaming engineers. L&D may propose a build path for the people already closest to the cluster. Talent strategy may still buy if the time horizon is shorter than any credible build. Either choice sits downstream of the citation, not inside it.

If you treat the gap as a requisition, you skip the choice the page exists to inform. Headcount workflows will turn a red cell into a req without asking whether the capability is funded, whether coverage in the skills file was incomplete, or whether a build path already exists. That is a process failure wearing analytical clothing.

External market language (job ads, professional-network skill phrases, competitor role titles) can help you name a cluster you would otherwise miss. Use it as a signal that a capability is talked about outside the firm, not as a percentage of the labor market that holds the skill.

Do not invent a market skill percent. You do not have a census of the market. Dashboards in the same vendor class as Workday, Eightfold, Gloat, and LinkedIn Recruiter may show supply or demand indices. If you display those, quote the vendor's own label and date, and do not convert them into a share of the market. If you cannot quote them, omit them. A qualitative note that the capability appears under several names in external role language is enough to keep the cluster honest.

Market signals also drift faster than your roadmap vintage. A cluster that is loud in job ads this quarter can still be unfunded on your signed roadmap. The roadmap vintage wins for planning. Market language is a naming aid, not a second roadmap.

A gap with no roadmap vintage is the most common poison. Someone clusters skills against where we are going and publishes a heat map. Months later nobody can say which bets were in scope. If you cannot print the roadmap date on the gap, delete the row. The same rule applies if the skills file has no date or coverage note. One vintage is not half a gap.

Treating the gap as a req is the second failure. The moment the gap system writes into an applicant-tracking workflow, you have skipped build versus buy. Keep the gap in the planning artifact. Route hires only after talent strategy records a buy decision. Internal moves belong to internal opportunity matcher after someone chooses to look inside. Learning assignments belong to role-based learning auto-assignment after someone chooses to build.

Inventing a market percent is the third. A slide that claims only a small share of the market has this skill is almost never sourced. It pressures a buy decision with a number you cannot defend. Drop the percent. Keep the qualitative signal, or keep silence.

Catch these in review as well: clustering against last year's org chart instead of the skills file; mixing inferred and self-asserted skills without saying so; filling blanks for people outside the skills-file coverage; using an undated vendor taxonomy as if it were your capability model; letting a single embedding neighbor imply proficiency rather than related language.

This analysis tells you where the current skills file and a dated roadmap disagree. It does not forecast how many people you will have in a cluster next year; that is internal mobility supply forecaster. It does not match a person to an open role. It does not enroll anyone in a curriculum.

Keep those systems loosely coupled. The gap cites sources. Mobility, matching, and learning consume a decision, not a red cell. If a blank remains because a source was missing, do not let a downstream system treat the blank as zero supply or as a mandate to hire. Empty stays empty until the skills file and the roadmap vintage both exist.

The practitioner test is simple. Hand the gap table to someone who was not in the clustering meeting. They should be able to name the two artifacts, see which cells were left blank on purpose, and still have to make a build-versus-buy call. If they can skip the call because the table already hired, assigned, or quoted a market percent, the quality bar failed.