The Build, Buy, or Borrow Read for a Capability Gap

You have a named capability gap and three ways to close it: build the skill internally, buy it on the open market, or borrow it through contractors and consultants. This guide is for strategy, research, and talent-intelligence teams who have to make that call and defend it. It gives you a scoring instrument that converts six dimensions into supply, time, and cost evidence, so one gap resolves to one move you can put in front of a leadership review.

Most public writeups on this decision are HR essays. They describe the three options and then hand you back to your gut. That is not enough when you are staffing a roadmap or sizing a capability build. Below, each dimension gets a score, and each score names the number it rests on.

What build, buy, and borrow actually mean here

Build means developing current staff through training, job rotation, and stretch assignments. Buy means recruiting the skill externally as a full-time hire. Borrow means renting it from freelancers, consultants, or vendors outside the organization. That operational split comes from SHRM's HR framing; the underlying model traces to Laurence Capron and Will Mitchell's book "Build, Borrow, or Buy: Solving the Growth Dilemma."

The problem is that none of these sources tell you how to weigh the three against the talent market for the case in front of you. This guide scores six dimensions:

The first five are market questions with numbers behind them. The sixth is a judgement, but you make it last, after the evidence is on the table.

External supply depth: the dimension that flips buy viability

Supply depth is the addressable pool for your exact skill in your exact market, and it is market-specific by nearly an order of magnitude. The same skill can be a plausible buy in one country and a build or borrow in another, on geography alone.

The evidence is stark. In Refolk's index of professional profiles, there are about 10,542 Machine Learning Engineers in the United States against about 1,260 in Germany. That is 8.4x more supply for an identical role, driven by concentration in hubs. If your gap statement says Germany, "buy an ML engineer" is a much weaker move than the same sentence in the US, and no amount of framework theory tells you that. The pool count does.

Scarcity is also skill-specific inside a single market. In Refolk's index there are about 3,422 profiles listing Rust against about 138,726 listing Kubernetes in the United States. Rust is roughly 40x rarer. When the skill graph is that thin, hiring velocity collapses and borrowing becomes the rational cost of speed rather than a failure of nerve.

Dataset A - External supply depth by market

Pool counts are a single pull from Refolk's index; ratios are derived by division. No public source names a hard "not viable" pool-size threshold, so score supply as a relative signal: compare your pool to a deeper adjacent skill in the same market, then confirm it with velocity. A big number is a green light for buy only if the market also converts.

Hiring velocity: does the market clear before your deadline

Hiring velocity is the gap between how long the market takes to fill your role and how long you have. Published time-to-fill is the clearest public proxy for whether buy is realistic, and in tech that proxy runs long.

The US average time-to-fill across all positions is 44 days, but tech-specific searches run 48 to 89 days depending on role and seniority. One benchmark dataset put the average at 63.5 days in 2025, down slightly from 67.7 the year before. If your deadline is inside those windows, buy is already losing on time before you count cost.

Depth does not guarantee conversion. Tech candidates face a 0.7% offer rate, 45% below the all-industry average. A deep pool that almost never converts to an accepted offer is a slow, competitive buy no matter how large the count. That is why supply depth and velocity are separate dimensions: one measures how many exist, the other measures how many you can actually land in time.

There is a structural signal too. Tech fills 16% of roles through referrals, 129% above the global average, and uses internal hires 25% more than average. When a market leans that hard on referrals and internal moves, the open market is a weak buy channel, and the data is quietly telling you that build and borrow already dominate there.

Internal adjacency and time-to-productivity: whether build is fast enough

Adjacency is the degree of skill overlap between your current staff and the target role, and it is the lever that makes build competitive on time rather than just on cost. Two roles are adjacent when they share core competencies even if the titles suggest otherwise.

You can estimate adjacency without HRIS access. Gartner analyzed billions of job postings and found a company needing a natural-language-processing expert can look to employees with machine learning, Python, or TensorFlow experience, because those skills are closely related. Public skill tags, project histories, and certifications are the observable proxies. Illustratively, a data analyst may already hold 70% of the capabilities needed to move into a business intelligence role.

Adjacency converts directly into a ramp estimate, which is your time-to-productivity dimension.

Dataset B - Build timeline by adjacency and role band

Read the table against velocity. A reskill at 0.6 overlap or higher finishes in 2 to 3 months. A tech hire takes 48 to 89 days to fill and then a mid-level ramp runs to a Gallup median of 8.2 months, or 6 to 12 months for a senior technical role. So a high-adjacency internal pool can beat buy on time-to-productivity, not only on cost. That is the finding most essays miss.

Adjacency scoring behavior is a competitive habit, not a nice-to-have: 46% of high-performing organizations actively identify adjacent skill sets versus only 26% of others. If you are not scoring adjacency, you are leaving your cheapest, fastest option unmeasured.

Finding Refolk profiles with the adjacent skills already listed is what turns adjacency from a guess into a count. Instead of asking a line manager who "might be close," you pull the people whose public skill tags prove the overlap.

Cost delta: the borrow-versus-buy call is an hours calculation

Cost is where borrow and buy get compared wrongly most often. The decisive number is not the headline hourly rate; it is the number of hours the gap actually consumes in a year.

Start by loading the full-time hire correctly. The standard employer multiplier is 1.25 to 1.40x base salary, so a $100,000 employee actually costs $125,000 to $140,000 once taxes, benefits, and insurance are counted. If you source through an agency, add a one-time fee of 15 to 25% of first-year salary. Cost-per-hire itself runs about $5,475 for non-executive roles and $35,879 for executive roles.

Now the contractor. Contractors charge 25 to 100% more per hour than the equivalent employee rate, because they cover self-employment tax, health insurance, retirement, unpaid time off, equipment, and income variability. In one worked example, matching a $130K salaried package with full benefits landed at roughly a 47% premium over the full-time hourly equivalent.

Dataset C - Cost multipliers across the three moves

The break-even is the whole game. Below roughly 1,500 to 1,800 hours per year, contractors are often cheaper because you avoid benefits overhead. Above it, the loaded FTE wins. So a short, bursty, or uncertain gap favors borrow, and a durable full-time-equivalent need favors buy. Duration decides, not rate.

The common error is comparing a contractor's hourly rate straight to an employee's base salary. That overstates the borrow premium every time, because the employee's real cost is 1.25 to 1.40x that base. Load the FTE first, then compare.

The scoring procedure end to end

Run these eight steps in order. Each produces one input, and the final step totals them into a single move with its evidence attached. Budget about five working days across an analyst, a line manager, finance, and a strategy lead.

One point of honest disagreement in the sources: order of operations on step 4. SHRM-style guidance says start from an internal skills inventory; market-first practitioners score external supply first. I score supply first because it is the cheapest dimension to measure and the fastest to eliminate a move. If external supply is deep and fast, you may not need a full adjacency study at all. If it is thin, adjacency is where the answer lives, and you invest there.

Dimension            | Build | Buy | Borrow | Evidence cited
External supply depth |   _   |  _  |   _    | pool count (market)
Hiring velocity       |   _   |  _  |   _    | time-to-fill vs deadline
Internal adjacency    |   _   |  _  |   _    | share of staff at >=0.6 overlap
Time-to-productivity  |   _   |  _  |   _    | ramp weeks from Dataset B
Cost delta            |   _   |  _  |   _    | annualized cost, hours vs break-even
Strategic ownership   |   _   |  _  |   _    | core / peripheral call
TOTAL                 |   _   |  _  |   _    | highest total = the move

How this read goes wrong

The scoring instrument fails in predictable ways, and every failure is a specific number misread. Screen for these before you present a recommendation.

• Pool count mistaken for hireable supply. A large index count includes people not open to moving. Check it against offer-acceptance and time-to-fill; tech's 0.7% offer rate proves a deep pool can still be hard to convert. The count sizes the opportunity, velocity sizes the reality.
• Time-to-fill blended across role types. A 44-day company average makes hourly roles look fine while specialized roles are actually slow. Segment by role family before scoring buy, or you will greenlight a hire the market cannot deliver on your deadline.
• Adjacency inferred from title, not skills. A "Marketing Analyst" at one company may spend 80% of their time on data visualization; the same title elsewhere is mostly copywriting. Two identical titles doing different work is the classic false positive. Verify against actual skill tags and project histories.
• Contractor rate compared to base salary directly. This inflates the borrow premium every time. Load the FTE at 1.25 to 1.40x first, then compare like with like.
• Ignoring the hours break-even. Borrow scored "cheaper" for a full-time-equivalent need is simply wrong past 1,500 to 1,800 hours per year. Recheck against expected utilization before you trust a borrow recommendation.
• Build timeline guessed, not anchored. Assuming a 3-month ramp for a senior technical hire is a false positive; role-band data says 6 to 12 months. Anchor every timeline to the adjacency score and Dataset B.
• Stale internal skills data. Knowing which employees have adjacent skills gives a running start only if the documentation is current. Confirm the skill records are fresh before you trust an adjacency estimate.

The single most valuable check: when supply is deep but velocity and offer rate are poor, the market is telling you buy is theoretically possible but practically slow. That is exactly the case where a high-adjacency build or a fast borrow quietly wins, and the essay-level frameworks will still say "just hire."

Verify before you call the read done

Before the recommendation leaves your desk, confirm the evidence chain holds. This is the last gate.

Keeping the read current

A build, buy, or borrow read has a short shelf life because its inputs move. Supply counts shift as markets grow, time-to-fill drifts with the hiring cycle, and internal adjacency changes as staff learn and leave. Treat the recommendation as accurate on the date you ran it, and note that date on the sheet.

Re-pull the pool count and time-to-fill whenever the deadline shifts or the market changes, since those two dimensions decide buy viability and they are the fastest to go stale. Re-check adjacency whenever the skills documentation is updated or the team composition changes. For an emerging skill like Rust, where the pool is roughly 40x thinner than a mainstream skill in the same market, re-check supply more often, because thin markets swing faster in relative terms. The instrument stays defensible only as long as the numbers under each score are the current ones.