Quantifying AI Impact: From Productivity Claims to Dollars

Over the past few years I have had the privilege of speaking with CEOs, CFOs, CHROs and COOs across a range of industries as they work through one of the most consequential shifts in modern enterprise history: the rapid infusion of AI into how work gets done. The enthusiasm in those conversations is genuine. So is a recurring problem underneath it.

Almost every organisation can describe what AI is doing for them in qualitative terms. Very few can describe what it changes in their workforce and what that change is worth in financial terms. The business cases stay abstract — AI will improve productivity, AI will reduce manual work, AI will augment knowledge workers — and abstract cases do not survive contact with a CFO. Without that clarity, leaders either overestimate the impact and commit to savings they cannot deliver, or they leave real value unrealised because nobody could defend the number.

We spent the last twelve months studying how AI savings actually show up inside enterprises, and built a framework for quantifying them. What follows is the method. The intent is not to promote a single tool but to describe an approach that finance, operations and the board can all review and agree on.

The questions an AI business case has to answer

Three realities intersect in most large enterprises right now. AI adoption is accelerating across functions, from Sales and Marketing to HR, Finance, IT and Operations. Headcount is the largest controllable cost on the P&L. And leadership has no unified model that connects the first fact to the second.

A defensible model has to answer five specific questions rather than one general one. Which roles are impacted, and how? How many hours, FTEs or management levels can realistically be reduced? Which savings are structural and which are temporary? Over what time horizon do they materialise? And how confident are the assumptions behind each of those answers? A framework that cannot answer the fifth question is not a framework; it is an estimate with a spreadsheet around it.

The underlying principle is that AI impact has to be evaluated role by role, task by task and skill by skill, then translated into financial terms using conservative assumptions. That runs across five steps: establish a role and headcount baseline, decompose roles into tasks, map AI impact and automation potential, segment and monetise the savings, then evaluate ROI with confidence scoring.

Start with a baseline you can defend

The first step sounds trivial and is usually the hardest. You need total headcount by function, the number of unique roles inside it, the level distribution across ICs, managers, directors and VPs, and a weighted average fully loaded cost per head that includes salary, benefits and overhead. From that you get total workforce cost and span-of-control metrics.

Most enterprises underestimate how fragmented this data is across HR systems, finance systems and functional teams. Without a unified baseline, everything downstream is speculative.

Take a Sales Operations function of 120 people as a worked example: 80 ICs at a fully loaded cost of $120,000, 30 managers at $165,000, and 10 directors at $210,000. That is $9.6M, $4.95M and $2.1M respectively — a total function cost of $16.65M per year. Every savings claim that follows is measured against that number.

AI does not replace roles, it replaces tasks

This is the distinction that most internal analyses miss. Each role has to be broken into its core tasks, with the frequency and effort of each and the skills needed to execute it. For a Sales Operations IC, that might be CRM data hygiene, pipeline reporting, forecast analysis, territory planning and ad hoc analysis. Each task then carries a time allocation, skill intensity, repeatability and susceptibility to automation.

Only then can you assess AI impact, across three distinct types: full automation, where the task can be handled entirely by AI or agents; partial automation, where AI accelerates human execution; and augmentation, where AI improves quality or decision speed without materially reducing effort. That third category is where a lot of inflated business cases quietly hide.

• CRM data hygiene — % of time: 20% · Hours/year: 400 · Automation type: Full automation · Effort reduction: 80%
• Pipeline reporting — % of time: 25% · Hours/year: 500 · Automation type: Partial automation · Effort reduction: 60%
• Forecast analysis — % of time: 20% · Hours/year: 400 · Automation type: Augmentation · Effort reduction: 30%
• Territory planning — % of time: 15% · Hours/year: 300 · Automation type: Partial automation · Effort reduction: 40%
• Ad hoc analysis — % of time: 20% · Hours/year: 400 · Automation type: Augmentation · Effort reduction: 20%

The weighted result is roughly a 45% reduction in total role effort, or about 900 hours saved per IC per year. Note what this assessment has to account for to be credible: the current maturity of the AI tools in question, integration readiness, and change management constraints. This is the step where do-it-yourself approaches usually break down, because manual assessments lack consistency, benchmarks and confidence weighting.

Segment the savings honestly

Not all AI-driven savings are the same, and presenting them as one number is what destroys credibility with finance. They fall into distinct buckets, each with a different confidence level and time horizon.

• Structural efficiency is permanent role or FTE reduction from sustained automation — eliminating manual reporting roles, consolidating redundant analyst positions. High confidence, typically realised over 12 to 36 months. In our example, applying a conservative 30% conversion of the 45% effort reduction to actual FTE reduction gives 24 FTEs from a base of 80 ICs, or $2.88M a year.
• Operational efficiency is reduced effort per role without immediate headcount reduction. Medium confidence, faster to realise, and often reinvested rather than banked. The remaining 56 ICs save around 500 hours each, which is 28,000 hours or 14 FTEs of unlocked capacity — $1.68M of avoided hiring rather than $1.68M of cash removed from the P&L.
• Organisational structure efficiency is the reduction of management layers that better visibility and decision support make possible. Moving from one manager per three ICs to one per six, and one director per three managers to one per five, takes managers from 30 to 18 and directors from 10 to 6 — $1.98M plus $840K, or $2.82M a year. This bucket is politically sensitive and frequently overlooked, and it delivers outsized returns when handled thoughtfully.

There is a fourth, optional bucket: skill redeployment. Shifting 20 ICs from reporting work to GTM strategy, at $250,000 of incremental revenue impact each and a 40% contribution margin, is $2.0M of operating profit. The ROI is indirect and harder to defend, but it is the bucket that enables growth without proportional hiring.

Separate theoretical, realisable and committed

Converting these buckets into dollars requires conservative assumptions about FTE equivalents, fully loaded cost per role, phased realisation timelines, and which savings are one-time versus recurring. The output should be a multi-year view of headcount reduction, cost savings, cash flow impact and margin improvement rather than a single annual figure.

• Year 1 — Structural: $1.2M · Operational: $0.8M · Org design: $0.9M · Total: $2.9M
• Year 2 — Structural: $2.4M · Operational: $1.4M · Org design: $2.0M · Total: $5.8M
• Year 3 — Structural: $2.9M · Operational: $1.7M · Org design: $2.8M · Total: $7.4M

The most important discipline here is distinguishing theoretical maximum savings from expected realisable savings, and both from committed savings. Boards forgive a conservative number that lands. They do not forgive a maximum presented as a commitment.

Why this stalls internally, and what to do next

Enterprises reasonably ask why they should not run this analysis themselves. In practice it requires large analyst teams, ongoing data ingestion and normalisation, continuous tracking of AI tool capability, skill taxonomy maintenance, and repeated task-level assessments. An analyst-led effort typically takes nine to twelve months to produce a first ROI view, at an internal cost of $1M to $2M, and confidence at the end is often still low. Draup’s Etter platform exists to compress that — standardised role and skill decomposition, confidence-weighted impact scoring, scenario modelling across roles, functions and geographies, and CFO-ready outputs — to a six-to-eight week exercise. In most evaluations, the platform cost is immaterial next to analyst labour and the opportunity cost of a delayed decision.

The method matters more than the tool. If you do one thing after reading this, pick a single function where you already believe AI is having an effect, build the baseline properly, decompose two or three representative roles into tasks, and produce a three-year phased number with the buckets separated and confidence stated. The question for leaders is no longer whether AI will change the workforce. It is whether you will be able to say how much that change is worth, in dollars, before someone on your board asks.