AI adoption that lands across the whole team.
Your people already use ChatGPT. Does your organization get value from it while keeping control of your data and IP?
Platforms we build on
Three ways AI adoption stalls
Bought and bypassed
Licences get bought for everyone, a launch email goes out, and usage flatlines in three weeks. Nobody mapped the work the tool was supposed to change, so it sits beside the real job instead of inside it.
Shadow AI
When the sanctioned tool is slow to arrive or awkward to use, people paste company material into personal accounts. The work still gets done. The data leaves with no record of what left, and you find out during an audit.
A pilot that never leaves the pilot team
Ten enthusiastic users get a good result. Nothing in the pilot was designed to hold up with the other 190 people, so the rollout restarts from zero and the momentum is gone.
AI readiness: what we look at first
Four questions we answer before a tool is chosen.
Which work is worth changing?
We look for the tasks a team repeats every week and can describe out loud: month-end entries, quotes, intake, the weekly report. Those are where a new habit pays for itself. Something done twice a year makes a good demo and a bad pilot.
Can the data be reached, and is it clean?
The work needs data people can actually get to, with rules on who sees what. Structure still matters, whatever you hear: a model will happily read a mess and answer differently every time you ask. Consistent answers come from data organized once and defined once.
What do legal and security allow?
Your constraints on personal information, residency and retention, settled before a tool is chosen instead of at the security review. In Québec that means Loi 25, and knowing where the line sits is an advantage, not a delay.
Will the culture carry it?
Culture is the first of our four Cs and the check that fails most often. We look at whether the team has room to change how it works, and whether the habit survives the week the launch stops being news.
Which AI tool is best?
They overlap far more than the marketing suggests. Here are the differences that actually separate them, as of this month.
No change of habit. It is already open inside the document, the mailbox and the meeting, so nobody has to learn a new tool to get the benefit.
It is only as organized as your SharePoint. If the files are a mess, Copilot shows you that rather than fixing it, and the fix is a file-estate project.
Claude Design. Set your brand up once as a design system and documents, decks and one-pagers arrive already looking like your company instead of being reformatted by hand.
No image generation, so anything visual still has to come from another tool.
Advanced voice mode. A real spoken conversation, which is the one interface people pick up without being trained on it first.
It knows nothing about your files or your systems until you connect them, and that connection is a project of its own.
Image generation. For a team that produces visuals, it is the one that actually gets used every week.
No top-of-range model, the way GPT-6 Astra and Fable 5.1 sit above the rest when a problem is genuinely hard.
Trust and verifiability. Every answer comes back with the query it ran, so a controller can check the number before it goes into a report.
It expects the governance to already exist. Unity Catalog and agreed metric definitions are the price of that trust, and both are work you do before you turn it on.
How we roll it out
Four stages, each one earning the next.
The Adoption & Absorption Diagnostic, 30 days
Data, culture and stack. One deliverable: a map of what to fix first, in what order, with what it takes.
One team, one pilot
We pick a workflow that already matters to somebody, and agree how we will know it worked before anyone starts.
Earned autonomy
An agent earns the right to act, one stage at a time. To start, it proposes but a human reviews everything. With trust built comes stage two, assisted, where it acts where its work has cleared your benchmark consistently. Stage three, bounded auto-action: it becomes autonomous on a field by field basis, every action traceable, auditable and revertable.
Enablement and handover
Your administrators, your champions and your usage rules, documented in the language your team reads.
The same work, cheaper and smarter every quarter
Models keep getting better and cheaper, so the same work can cost you less each quarter, or do more at the price you already pay. That only happens if something is measuring it, and that is what we leave running.
Define what good is, and measure it
We agree the threshold a good answer has to clear, then hold every candidate model against your own examples instead of a public leaderboard.
Optimize for capability
You run the cheapest model that clears the threshold, so when a better one qualifies the same budget buys more capability instead of the same output.
Auditable and traceable
Every request is tagged to its case, its cost, its quality and its business outcome, so any answer can be traced back to the model and the data behind it.
Keep runaway costs in check
Budget breaches, exception spikes and drift raise an alert while they are still small. An agent nobody is watching is how a pilot becomes a surprise invoice.
In their words
The people who run the work every day, on what changed.
“I used to have a desk buried in paper. Now every task starts and ends in the same place: our SharePoint portal. It changed the way I work. And when we work with Claude, all the context it needs is already there.”
“JP gave us agility. We update our site ourselves and test a promo or a new idea the same day. No developer, no waiting: we're independent.”
Running an SMB but don't know where to start?
Learn AI on a real use case of yours, in your own environment, in French or English.
You talk to it.
The right subscription and privacy settings, your documents in one workspace, and one task done on your own files.
It works for you.
Skills, routines and connectors run tasks in your files while you do something else.
It works for the team.
What one person built is shared, reviewed and logged. Auditable, traceable, repeatable.
Questions
On raw capability they are close enough that either will do the work. The familiarity argument for ChatGPT is also fading, since plenty of people are now trying Claude at home. For a business the consideration worth weighing is Claude Design: set your brand up once as a design system and documents, decks and one-pagers come out already looking like your company. That is the difference you feel every week.
Because they do different jobs. Copilot is strongest where your day already is, inside Outlook, Teams and a SharePoint file. Claude and ChatGPT bring a platform around the model: instructions and skills you write once, memory, connectors into your own systems, and features like Claude Design that turn a draft into a document that looks like yours. Teams frustrated with Copilot are usually not hitting a limit of the model. They are hitting the limit of what one assistant inside one suite can be asked to do.
No, and this is the assumption we correct most often. Give a workbook or a deck to Claude or ChatGPT and you will usually get better analysis and a cleaner rewrite than the in-suite assistant produces, then bring the result back or connect the file store directly. Copilot’s real advantage is that it is already sitting in the document, not that the others cannot read it.
That depends on which AI subscription you are on, and it is the part most organizations have never looked at. On a business or enterprise subscription your content stays out of model training, retention is set by your administrators, access is logged and exportable, and residency can be specified. We put the subscription and its settings in writing per tool before rollout, and the usage rules then tell staff what may be pasted in.
Free and personal subscriptions (ChatGPT Plus, Claude Pro, a personal Microsoft account) are sold to an individual, so the settings belong to that individual and your company has no administrator, no audit log and no agreement. Business and enterprise subscriptions (ChatGPT Business and Enterprise, Claude Team and Enterprise, Microsoft 365 Copilot) put your organization on the contract: content excluded from training, retention and access controlled centrally, usage visible, and a document to point at when a client or an auditor asks. The personal one is what your staff are already using today.
Six to twelve weeks per team from pilot to daily habit, when the work is mapped and someone owns the change. Adoption stalls when the tool is announced but no existing workflow is retired. We measure weekly active use per team rather than licences assigned.
