AI strategy for software companies
From Software as a Serviceto Service as a Software.
Models are available to everyone, your competitors included. What they do not have: twenty years of codified craft — your rules, your registers, your resolved cases.
That asset was becoming a commodity in the era of the tool; it becomes the scarcest resource in the era of the result. Software no longer merely equips the work — it does the work, and your clients validate.
This is where software history is heading, and you are best placed to take it.
The value creation plan we implement with your teams focuses on two key areas: growth and productivity. AI is a standalone initiative—and the only one that impacts both. Axis 1 targets your product offering, driving growth, while Axis 2 enhances the productivity of your organisation, spanning nine functions from customer support to senior management.
The examples on this page are drawn from the proptech market. While they illustrate key points, the principles they demonstrate apply to other industries—we have selected them for the clarity of the explanation.
The sorting test
What AI does on its own, and what stays yours.
Before choosing where to invest, you need to know what can be automated and what must not be. That boundary decides everything else — and it moves in your favour.
The sorting criterion, before the two axes
Software has always executed what a human had reasoned. What changes is that the reasoning itself is being automated on part of the work.
A task therefore breaks down into two materials: AI handles the first autonomously, the second stays human — and it is what makes your value.
The applicable rule
Finding information, cross-checking it, applying a reference framework, producing a compliant deliverable. The rules can be formidably complex — they are still rules. That is where AI is already autonomous, and where both your productivity gains and your first augmented products sit.
The judgement call
Deciding under ambiguity, taking on responsibility, accounting for what is not written down. It takes experience and taste. It is what your clients pay you for without ever having spelled it out, and what you must not claim to automate.
The boundary moves, and it moves your way: every judgement your users validate inside your product documents what good judgement looks like in your field.
What is a judgement call today becomes an applicable rule tomorrow — but only for whoever kept the trace of it. That is precisely what your twenty years of resolved cases amount to, and what no model hands your competitors.
There is a reason to act now, and it is not defensive.
Nothing stops a third party from wiring its agents onto your product and selling your client the finished work — your software then becomes the infrastructure, indispensable and invisible, while the relationship, the margin — and the responsibility for the result — move to someone who knows neither the trade nor the data.
The same mechanics run by you produce the opposite: it is your name that delivers the work, on your data, with your validation, and it is you who answers for the error. The technique is identical; what changes is who runs it.
Two axes
Your product offering, and your organisation's productivity.
Both are run in parallel, and they feed each other: the product offering delivers a new service to your installed base — and opens the revenue line that goes with it; the organisation’s productivity frees the capacity — and the teams practising AI day to day are the ones who then know how to embed it in the product.
The build — how you make it
The product — from the assisting tool to the work delivered
How a level is read
The build — a chain
- 1.1
- 1.2
- 1.3
The level is that of the weakest link. Shipping fast is worthless if going live takes six weeks.
The product — a trajectory
The better of the two routes counts — but the axis stays capped until the share of work delegated is measured.
The teams — a count
Nine independent functions: you count the ones that are tooled. An average would have said nothing.
The order matters more than the choice: the passage gives you the reference clients, the proven architecture and the revenue that fund the destination. Treating it as an end in itself is the costliest mistake on this axis.
And that passage shifts your buyer: the professional who was your champion must come out ahead before the buyer does — that is a design condition, not a sales argument.
The cases that instruct each of these projects are in the next tab.
The nine functions — and the time each can give back
None of these nine functions is addressed for its own sake: each frees time on a task you can name. That is what the diagnostic measures, function by function.
Tool no. 1
Where does your company stand on AI?
Twenty-seven statements, a level per axis and three prioritised workstreams — all in your browser, with no email address. Without JavaScript the questionnaire cannot render. We run it with you in an hour.
Tool no. 2
The use cases
Every case is worked through the same way — the problem, what we put in place, what is measured, prerequisites and traps. Without JavaScript the library cannot render: we walk you through them in a meeting, on your catalogue.
Tool no. 3
The slider journey
This module runs in the browser: step by step, it shows what a rental-management task becomes, and what it weighs in hours and euros.
Without JavaScript it cannot render — we are happy to walk you through it, on your own figures.
Creating and sharing value
You no longer sell a tool. You sell a result, priced against what it saves.
This is the deepest shift on the page, and it is not settled in a price list: it is settled in what the client compares.
Yesterday they compared your price to another piece of software. Now they compare it to what the task costs them internally or at their provider — a ratio of one to several, not a few percentage points.
Your subscription remains the base, but a seat counts users: the day part of the work is done without a user opening the product, that count stops measuring anything.
Hence these three forms, added to the subscription rather than replacing it — and all negotiated on the same basis: the cost avoided at their end.
The two rules
The first concerns the client: price is defended by the ratio, never by the amount.
"Two hundred thousand" means nothing in the abstract; "two hundred thousand for one point two million in processing cost removed" can be verified — and it is the client who must be able to redo the calculation.
And they must keep the majority share of the value created: it is born at their end, so it comes back to them first. A model built the other way round does not survive the first renewal, and it travels badly in their sector — but that is not the reason to set it up this way.
The second concerns you: the share you charge must be backed by a contractual commitment. Without one, that revenue is valued as a service engagement rather than a subscription — you would have created value for everyone except your own company.
What to measure, on both sides
- At the client first: time returned, errors avoided, deadlines met. Without a baseline at their end, there is nothing to share and nothing to defend.
- The ratio between what they gain and what they pay: the only argument that survives a change of contact on their side.
- On your side, average revenue per account and net retention: a result delivered defends itself at renewal better than one more tool.
- Your unit gross margin: inference cost belongs in cost of revenue and varies with usage. A variable component is not recognised like a subscription — settle it with your CFO before the first contract.
Compliance
Three matters to settle before the first deployment.
They do not slow an AI project down: they make it publishable. A doctrine written upfront costs a few days; retrofitted, it costs a rewrite.
We are not a law firm: these three matters shape the doctrine and the specification, they do not replace legal advice. We work alongside your counsel on the clauses.
Trajectory
What the first six months look like.
Neither a pilot going nowhere nor a three-year programme. Six months from a decision to attributable revenue — that is the pace we hold with our clients.
- Diagnostic across both axes, presented to the executive committee.
- Written doctrine: sovereignty, human review, the boundary.
- Two or three use cases selected, on value rather than feasibility.
- Baseline measurement established — without it, no gain will be demonstrable.
- Sovereign architecture and abstraction layer in place.
- First workflow in production on a narrow scope.
- Human review tooled, incidents logged.
- Internal functions start in parallel, without waiting for the product.
- Gain measured at the pilot clients, against the control group — that is what grounds the price.
- Packaging and pricing settled, standard contracts revised.
- Extension to other functions and other use cases.
- First attributable revenue, internal gains reconciled with the accounts.
This calendar assumes two conditions: a sponsor on the executive committee and accessible data. Without either, month one takes three — and it is the one delay we have never seen recovered later.
The method
An end-to-end AI strategy, not an experiment.
From diagnostic to go-to-market, we run the transformation with the same discipline as our value-creation plans: dimensions assessed, use cases prioritised, execution governed — and human review before any production release.
In the field
Real engagements, run in immersion.
Our mandates are confidential; the engagements speak for themselves.
How it is built
The architecture, layer by layer
This module details both reference architectures — augmented product and Service as a Software — with their technical components and guardrails. Without JavaScript it cannot render; the subject is better handled live, with your CTO.
Frequent questions
What we are asked about AI.
How do we share the value AI creates with our clients?
By measuring what they gain first, and charging only a share of it — the client must keep the majority of the gain, otherwise the model does not survive the first renewal. Only then comes the form.
An AI module as an add-on subscription strengthens recurrence; usage is billed beyond an included volume, on a firm annual commitment; value sharing requires an incontestable measure and a multi-year contract.
Usage revenue without commitment is valued like services revenue, not recurring revenue.
Do we own code generated with assistance?
It depends on three documents that must be read together: your vendor contract, your client contracts, and your own internal policy.
The prerequisite is traceability: without a log of what was generated and validated, none of these questions can be demonstrated to an acquirer or a client.
How long before revenue attributable to AI?
Six months is a sustainable pace: one month to decide and set the doctrine, two to install and put a first workflow into production, three for pilot clients, packaging and extension.
This assumes an executive sponsor and accessible data — without either, month one takes three.
What is "Service as a Software"?
The reversal of the SaaS promise: software no longer merely equips the work — it does the work, and your clients validate the result. Value migrates from the tool to the result, which changes the product, the pricing and the organisation.
How do we protect our proprietary data in an AI project?
Through doctrine before tools: data sovereignty, a strict boundary between AI-assisted code and client data, sovereign RAG over your corpora, and an LLM-agnostic abstraction layer so you depend on no vendor.
Where should a software company start its AI strategy?
With a maturity diagnostic, dimension by dimension, then the framing of the highest-value use cases — combining the executive view with field workshops in product, sales and support.
Experimentation without doctrine is expensive and produces no revenue.
Does AI make SaaS obsolete?
No: it downgrades interchangeable tool-software and multiplies domain software. Software companies that encapsulate their expertise and proprietary data in AI-augmented products come out stronger.
Get in touch
Let us continue the conversation.
Every situation is singular — yours deserves better than a page.
5 Avenue de l'Opéra, 75001 Paris — France