For your CFO

Stop asking what AI costs. Start asking what a unit of it costs.

Total spend is the wrong number — it only ever goes up, and it tells you nothing. The question a finance team can actually act on is what one unit of AI work costs, and whether the thing it replaced cost more.

Where are you today?

Most companies are further behind than they think.

Each level has a different cost profile. Levels 1 and 2 cost you in ungoverned risk and stalled pilots. Level 3 is where the spend starts turning into a number you can defend.

Level 1

Exploring

A few people trying AI tools on their own, informally — and no way to know what it’s costing or saving.

Level 2

Standardizing

The company has picked some approved tools and written basic rules — spend is visible, but return still isn’t.

Level 3

Scaling

AI is used safely across multiple teams, with usage tied to a cost baseline — this is where DouJou operates.

Level 4

Compounding

AI is tied directly to business goals, and every new use makes the next one cheaper to add and easier to justify.

Most companies are stuck between Level 1 and Level 2 — not for lack of ambition, but because nobody has built the layer that turns usage into a number leadership can rally around. That’s what DouJou is.If your finance team already runs a FinOps practice, this maps onto the crawl / walk / run maturity model they’ll recognise.

How the ROI gets tracked

From individual requests to one board slide.

This follows the same shape a FinOps team already works in — inform, then optimize, then operate. The difference is that the unit being measured is a piece of AI work, not a server.

1

Inform — every AI-assisted task is logged.

The same audit trail that makes DouJou safe to use also makes it possible to see exactly what AI is being used for, team by team. You cannot allocate what you cannot see.

2

Baseline — it’s tied to a real cost.

Each task is measured against what the manual alternative actually costs your company — your salary bands, your volumes. Not a vendor’s assumed benchmark, which is the part that usually makes these numbers unusable.

3

Optimize — cost per unit of work comes down.

Caching, routing, deduplication and document compression cut what each request costs to serve. Because everything is attributed, you can see the unit cost falling rather than just the total rising.

4

Operate — it rolls up to one number.

The figure leadership sees isn’t “we’re using AI” — it’s the cost it actually saved, per team and in total, tracked rather than guessed at.

Total cost of AI ownership

The licence fee is rarely the expensive part.

When AI tooling is priced per seat, the bill grows with headcount whether or not the tool gets used. Self-hosted changes the shape of that: the compute sits on the cloud bill you already negotiate, the data never has to be duplicated into someone else’s platform, and the cost of adding the next team is close to nothing.

1

Compute on your own bill

Runs in the cloud account you already have a commercial relationship with — at your rates, not a vendor’s markup.

2

No second copy of your data

Nothing has to be migrated into a vendor platform, which is usually where the hidden integration cost lives.

3

Token spend actively reduced

Document compression and response caching cut what each request costs — the savings are measured, not assumed.

In numbers

What a defensible number looks like.

0

AI spend that isn’t tied to a tracked outcome

Every request runs through DouJou, so every request is attributable to a team and a use case.

100%

of usage tied to a real cost baseline

Measured against what the manual alternative actually costs your company.

1

number your whole company can rally around

One rolled-up figure for the board, not a slide of anecdotes.

We don’t publish a headline ROI percentage, because yours would be made up. What we can show you is a real engagement, with figures verifiable in the customer’s own git history.

The cost of not doing this

Both directions cost you — just differently.

Ungoverned AI spend

Every team picks its own tools, on its own budget line, with no way to see the total — or to tell which of it is actually paying off. The first real number arrives when procurement adds it up at renewal.

Locking it down instead

You avoid the visible risk, but you give up the upside entirely — and your competitors who found the third option pull ahead while you wait.

Next step

Build the number for your own board.

We’ll work out what the baseline actually looks like for your business before you commit to anything.