The email you sent last Thursday — the one that said “we’d love to, but we can’t start you until late Q2” — was not a demand problem. The client had budget, a signed scope and a date. You had the expertise. What you did not have was a named senior engineer free on the 14th. That is a hiring-speed problem wearing a delivery-capacity costume, and it is now the single most expensive line item in data, AI and GTM services businesses that never appears on a P&L.
The “no” was a staffing decision
Services leaders are trained to read a declined project as a sequencing issue: the work will come back, the client will wait, the pipeline holds. Increasingly it does not. When a client’s platform roadmap is tied to a quarter-end milestone, “late Q2” is a referral to your competitor.
And the volume of work on the table keeps climbing. Over a million Genie Agents were created on Databricks in 2026 alone — a million separate little surfaces that someone has to scope, govern, evaluate and own after launch. Every one of those is a potential statement of work for a firm that can show up with a specialist who has done it before. None of them wait four months for your bench to clear.
You are not losing deals on capability or on price. You are losing them on availability.
Spend is growing faster than the team
The clearest tell that this is a capacity squeeze rather than a soft market is the gap between what firms spend on platforms and what they spend on people. In dbt Labs’ 2026 analytics engineering research, 57% of teams reported increased warehouse and compute spend against just 36% reporting an increased team budget.
Read that as a services leader rather than a practitioner. Your clients are spending more on the platform every quarter while their internal headcount budget barely moves — which means the implementation, migration and enablement work they cannot absorb internally lands on external delivery. Yours, if you can staff it. That 21-point gap is your addressable market, and it is also the reason your clients cannot simply hire their way out and stop calling you.
It is the same dynamic covered in more depth in capacity versus adoption on data teams: platform adoption curves and human capacity curves have come apart, and the space between them is filled by whoever can mobilise fastest.
AI accelerated build, not operate
The tempting internal answer is leverage: if AI-assisted coding makes each engineer meaningfully faster, you do not need more engineers, you need better tooling. That holds for exactly one phase of delivery.
The same research found that while 72% of data teams prioritise AI-assisted coding, only 24% prioritise AI-assisted pipeline management — testing, observability and quality controls. Build got faster. Operate did not.
What that does to a services P&L is specific and unpleasant:
| Delivery phase | Effect of AI assistance | Who can do it |
|---|---|---|
| Scaffolding models, transformations, boilerplate | Materially faster | Mid-level, supervised |
| Migration cutover and reconciliation | Marginal | Senior, named |
| Governance, lineage, access design | Marginal | Senior, named |
| Testing, observability, quality gates | Largely unchanged — only 24% prioritise it | Senior, named |
| Client-facing enablement and handover | Unchanged | Senior, named |
The phases AI compresses are the ones you could already staff flexibly. The phases that gate a go-live are the ones that still need a specific human with a specific platform history. So your throughput ceiling is set by how many senior, named specialists you can put on a project in the month the client wants to start — not by how productive each one is once they are on it.
A req is not an answer to a Q2 start
When delivery is full, the instinct is to open headcount. The labour market data says that is a plan for next year’s pipeline, not this quarter’s.
44% of US hiring managers say their company currently has open positions it cannot fill — up from 36% in autumn 2025 and the highest reading in three years. Nearly half of employers are already carrying unfilled roles. An approved req, in that environment, is a statement of intent rather than a resource.
Even when it works, it is slow. 43% of hiring managers say time-to-hire has increased, and more than one in four — 26% — say it now takes at least four weeks once resume reviews, interviews and the final decision are counted. Four weeks is the fast path, and it excludes approval, sourcing at the senior end, notice periods and ramp.
Stack it against the client’s clock:
| Route to capacity | Realistic time to billable | Commitment |
|---|---|---|
| New permanent req | Approval, then four weeks minimum of process, then notice and ramp | Permanent cost base |
| Internal reassignment | Immediate — but you have just destaffed another project | None |
| Subcontract to a partner firm | Fast, variable quality, margin compression | Per project |
| Pre-vetted flexible specialist | Days to shortlist, engaged on project terms | Project only |
| Decline the work | Immediate | Loses the account |
Nor does the macro picture suggest approvals are about to loosen. US nonfarm payrolls rose by just 29,000 jobs in September with unemployment ticking up to 4.2%, below expectations. A cautious labour market is one where finance holds permanent headcount while your pipeline keeps signing — the precise condition under which delivery leaders start declining work.
Flexible capacity is the only lever that moves this quarter
The one part of the market that is loosening is the contract one. Flexible staffing revenue in the US is forecast to grow 1% in 2026 and 2% in 2027 after years of contraction — modest in absolute terms, but directionally it means senior specialists are once again available on project terms rather than only through a permanent offer.
That matters because the people you actually need for a Q2 go-live are, almost by definition, employed. The platform architect who has run three Unity Catalog migrations is not on a job board. The GTM engineer who has rebuilt a lead routing stack in HubSpot and Snowflake is mid-contract somewhere. The only way to reach that cohort inside a client timeline is through a channel where they are already identified, already screened and already open to project work — which is the whole premise of an anonymised specialist network like the data engineering side of Lakehouse or the GTM engineering side. Searching costs nothing and needs no account, so scoping availability before you sign the SOW does not commit you to anything.
What to put on flexible capacity and what to hire
The answer is not to flex everything. The useful split is between the roles that carry your methodology and the roles that carry a client’s platform specifics.
| Keep permanent | Flex per engagement |
|---|---|
| Engagement leads and client owners | Platform-specific migration specialists |
| Solution architects who define your delivery model | Surge capacity for fixed-date cutovers |
| QA and delivery assurance | Niche connector, governance or ML-ops depth |
| Practice leads who train the bench | Vertical or regional expertise you sell once a year |
If you find yourself hiring permanently for a skill you will need on two engagements this year, you have converted a project cost into a fixed one — and in a market where 44% of employers cannot fill the roles they already have, that permanent req will also sit open for months while the project slips.
Say yes first, then solve the staffing
The operational change is small and the commercial effect is not. Today, most services firms qualify a deal against the bench: can we staff this? The alternative is to qualify against the network: can we staff this if we add one or two named specialists on project terms? The second question is answerable in days rather than quarters.
Three practical moves:
Shortlist before you scope. When a deal reaches proposal, check specialist availability for the two or three skills most likely to gate the start date. You are not hiring — you are pricing in known availability so the start date in your SOW is one you can actually hold.
Price flexible capacity into the proposal. A blended rate that assumes some project-based specialists is honest, defensible and far better than a lower rate with a start date four months out. Clients buy certainty on dates more often than they buy the cheapest day rate.
Decide your blend deliberately. The choice between subcontracting to a large integrator, building a permanent bench and engaging specialists directly is a margin and control decision, and it is worth making once rather than deal by deal — the tradeoffs are laid out in GSI or in-house blended delivery.
The market conditions are not ambiguous. Client platform spend is rising while their team budgets are not. AI has compressed build and left operate alone. Permanent hiring is slower and less reliable than at any point in three years, and flexible specialist capacity is the only supply curve pointing up. A firm that treats “delivery is full” as a fixed constraint will spend 2026 referring its own pipeline elsewhere. A firm that treats it as a sourcing question will not.