A0
40 days · 8 weeks
Prove what your AI investment returned, in figures your finance team can rebuild.
Vendor cases price theoretical time savings as cash. A board can only spend money that moves in the ledger.
AI licensing has moved from the experimental budget to the operating budget, so the spend recurs every year. The evidence behind it is still a vendor benchmark.
There are two ways to get this wrong, and they sit at opposite ends of the same licence term: one at purchase and one at renewal.
Buy without recording how long the work took beforehand, and you can never prove what the licences did for you. Renew with no evidence of your own, and you commit a second time to the vendor’s arithmetic.
Both failures come from one gap. Nobody measured the return on your own data in the first place.
Your finance director either approves spend they cannot prove, or rejects one that may well be working, and nobody can tell which.
Two modules produce a return figure for your AI licences, worked out on your own data: four to six weeks, or six to eight. You see the first figure in week two, before either module finishes.
In both, I report the figures that fail to support the spend alongside the ones that do. You take one, depending on whether the money is already committed.
Your CFO is asking what this did to revenue, what it took out of cost, and what it did to risk. A dashboard of tokens and daily active users answers none of those.
| What you get shown | What it leaves out |
|---|---|
| Tokens used, or daily active users | Which line of your accounts either one moves |
| Hours saved per person per month | Whether those were hours anybody was short of |
| Hours saved, turned into money | Whether the money leaves your budget, or the hours go back to people you still pay |
| Adoption percentage | Whether adoption held past month two |
| Quality, as a satisfaction score | Whether the output was usable without editing |
| The gross saving | The three costs that come off it |
| Return per seat | Which teams returned nothing |
If a study reports a single number, make it the share of output usable without editing.
It is the only figure telling a director how much of the work the tool finished. It usually sits well below the volume figure printed beside it.
At EY, I ran the study behind a fifteen million dollar commitment to twenty thousand Microsoft 365 Copilot licences.
Ninety-two per cent of users intended to keep it, and thirty-six per cent of its output was good enough to use unedited. Both readings were true, and EY adopted the framework I built to carry both.
| Module | For | Who buys it | Duration |
|---|---|---|---|
| A1 | A spend you have not committed to yet | A CIO or head of AI holding live pilots and a scaling budget | 4 to 6 weeks |
| A2 | An estate you already own and have to defend | A CFO, a CIO, or whoever owns the renewal | 6 to 8 weeks |
A1 is the cheaper way in, and it is the one to buy first if you are choosing. Four to six weeks, a scored list in week two, and a gate you can act on before the assessment closes.
A1 also leaves behind a record of how long the work takes now. That record is the baseline A2 measures against once the term has run, so buying A1 today makes a measured A2 possible next year.
Two terms carry both modules, so I define them here.
A cash saving takes money out of your budget, which means a contractor stops, a vacancy closes, overtime ends or an invoice is cut. A time saving gives hours back to people who remain on your payroll, real but not money this year.
Where the A1 gate returns a decision to fund nothing, I don’t write the business case, and I don’t charge you for that half of the work.
I set the fee after the first conversation. It depends on the number of teams in scope and how much usage and timing data your administrators can already export.
| You get | From | What it changes |
|---|---|---|
| A return figure worked out on your own data | Both | Your board stops deciding on the vendor’s benchmark |
| Cash savings and time savings reported separately | Both | Finance knows which portion it can budget against this year |
| Three costs deducted and itemised | Both | The figure holds when your audit committee checks the arithmetic |
| The share of output usable without editing | Both | Your board learns how much of the work the tool finished |
| A record of how long the work takes now | Both | You can measure the same estate against it next year |
| A scored list, one row per pilot, with a gate against each | A1 | Somebody can stop a pilot and survive having done it |
| A stated headcount position | A2 | The reduction somebody promised your board gets tested |
| A return figure for each team | A2 | You move seats towards the teams that use them, and reclaim the rest |
Each page has its own phases, limits, and what it settles.
The three costs I deduct are training time, the hours spent correcting output that came back wrong, and the work of connecting the tool to anything.
If your problem is a spend nobody has committed to yet, then AI Scale Assessment is for you. Start there if both apply, since its baseline is what makes the second one measurable.
If your problem is an estate you already own and have to defend, then Workforce Productivity Assessment is for you.
A1 - AI Scale Assessment is the scored list that decides which of your AI pilots get the scaling budget. You are past the pilot stage with more live pilots than scoring criteria, so the decision falls to whoever argues loudest, on figures that belong to the vendor. With the AI Scale Assessment module, we score every live pilot against four criteria, based on the seat usage and timings your own administrators can export.
20 daysA2 - Workforce Productivity Assessment is a return figure for your AI licences, measured within your own estate rather than modelled from a benchmark. Your dashboard reports assisted hours, which the documentation reveals to be action counts times a fixed six minutes at a US labour rate, so your finance director discounts it and is right to. With the Workforce Productivity Assessment module, we measure what the tool did to the work: your own people record the baseline, then observe the period, and cash and time come back on separate lines.
6 to 8 weeks