The Headcount That Didn't Drop
A practical first step for capturing AI productivity without restructuring the whole organization.
The most widespread consulting guidance on AI tends to converge around organizational redesign: redefining roles, updating job descriptions for AI-enabled workflows and restructuring teams to realize productivity gains. This advice, however, is expensive, politically hot and for most mid-market companies, premature.
The better first move is smaller and quieter: audit where AI is actually changing throughput at the function level, then adjust the forward hiring curve accordingly. This avoids premature restructuring, internal communications challenges and labor-relations complexity. The result is simple and digestible for any company: a revised workforce assumption in next year’s plan.
HR doing its job might be the problem
The reason AI productivity gains are invisible is structural.
Headcount planning is owned by HR, built around business-vertical needs, with no input from AI productivity assumptions.
The HR function is doing exactly what it was designed to do, with the inputs the process is designed to take.
HR runs the annual cycle. The intake question to function leaders is some version of "what headcount do you need next year, given growth and attrition". The question is forward-looking on demand and backward-looking on productivity. It assumes the productivity floor is roughly where it was last year. Function leaders, answering it, aren’t being asked to factor in AI productivity gains, so they don’t.
AI strategy sits elsewhere. Often in IT, sometimes a CDO, occasionally inside the COO function. Whoever owns it has visibility on tools, vendors, and usage. They don’t have the hiring forecast, and they are not a part of HR’s planning conversation. The CFO sees both numbers in aggregate at the budget layer, but CFOs rarely intervene at the function-level hiring assumption layer. That’s not where their attention sits in the cycle.
So HC plans get built on last year’s productivity floor, year after year, while the productivity floor is shifting underneath. The integration would have to be designed in: CFO direction at the planning input layer, or a dedicated workforce planning function with cross-functional authority. Otherwise, nothing changes, and the gap quietly gets bigger.
Don’t redesign roles yet
The competing prescription is role redesign. The argument has its place eventually, but as a first move it’s wrong, and the main reason is data, not cost.
Role redesign assumes you already know which AI productivity gains are real and persistent. At this stage of adoption, most companies don’t have that data. They’re redesigning roles against productivity assumptions that haven’t stabilized. If the assumptions move (and they will, because tooling is still maturing and adoption patterns are still settling), the redesign locks the company into a structure that no longer fits the underlying reality.
The cost argument is real, but it is not the main point. Role redesign is expensive in change-management terms. It is slow, politically charged, and especially difficult in EU labor environments where works councils have consultation rights. More importantly, it can create organizational resistance and consume goodwill that future AI initiatives will depend on.
The financial upside that role redesign promises can be captured through forward hiring-curve adjustment instead, with a fraction of the disruption and none of the lock-in risk. Role redesign becomes the right move at year three or year five, after the productivity floor has stabilized and the data is clean. Earlier than that, you’re building structure on sand.
The audit starts at the task level
A "15% productivity gain" applied uniformly across a function is a misleading average. The variance hides where the adjustment actually lives.
Take a finance function. A junior FP&A analyst spending most of their time on reporting, variance commentary, and reconciliation work is in the 25-30% productivity gain range with current AI capability. A senior analyst doing scenario modeling, stakeholder management, and ad hoc strategic work is closer to 5-10%. Same function, same headline number, very different implications for hiring.
The audit method works in three steps:
Map the task composition of each role, by hours per week or by deliverable
Identify which tasks are affected by current AI capability and apply a productivity range to each, drawn from observed performance, not vendor claims
Aggregate back to role level to get a defensible per-role productivity gain
The output of that audit goes directly into the hiring forecast. Take a finance function with 8 analysts and a 10% year-on-year headcount growth assumption built into next year’s plan. Baseline plan: hire 1 net analyst, going from 8 to 9. With a 15% productivity gain at the analyst layer, the realistic capacity need is approximately 7.6 FTE-equivalents to handle the same workload, or 8.4 FTE-equivalents at the original demand growth. The plan adjusts: hire 0 net, or hire 1 and reassign existing capacity to absorb new demand. The intervention is the planning input. The existing roster doesn’t change.
The most important move in this whole approach is the attrition leverage. Most mid-market companies run 10-15% annual voluntary attrition. If your AI-driven productivity gain matches or exceeds attrition in a given function, the entire adjustment can happen through reduced backfill.
Nobody is fired, and there is no role restructuring.
You change the backfill ratio.
That’s a quieter conversation than "AI is displacing people", and it’s the conversation boards actually approve.
What it’s worth over three years
The year-one savings are modest. The compounding over three years is where the analysis earns its keep.
Take the same 8-analyst finance function. The original three-year plan, built on last year’s productivity floor, grows headcount in line with revenue. The audit-driven plan absorbs that demand growth through productivity gains and natural attrition.
Cumulative three-year delta: 5-6 FTE-years of cost avoidance. At a fully loaded cost of €120k per FTE in a Western European mid-market context, that’s €600-720k captured without restructuring, without a communications program, without works council friction, without anyone in the function feeling threatened.
In P&L terms, that’s function cost as a percentage of revenue declining quietly while the function’s output keeps pace with growth.
In a services business, it shows up as contribution margin improvement. In a product business, it shows up as operating leverage. In both cases, it shows up before any consultant has finished a role redesign deck.
The question worth sitting with
This analysis is invisible to most companies because it does not sit cleanly within a single function. HR sees the hiring process, but not the AI productivity signal. AI strategy sees the technology roadmap, but not the workforce forecast. The CFO sees the aggregate plan, but usually does not intervene at the function-level assumption layer. The work sits in a gap where four roles touch the edges and none owns the middle.
When was the last time your company tested a function-level hiring assumption against an AI productivity input?
If the answer is never, the cumulative overhire by year three is already on the books.
It will not appear as an obvious error. The plan will look defensible, because each individual assumption will seem reasonable in isolation.