Four ways to make a regulated firm AI-native
Claude, ChatGPT, Microsoft Copilot or your own agents: what it takes for a UK or Irish mid-market firm in financial services, healthcare or legal to run on AI rather than experiment with it.
A regulated firm is AI-native when everyone has a governed assistant, its method is encoded rather than remembered, its systems of record are reachable, and its heaviest process runs itself with a named person signing. Copilot is the floor at £24.60 per user per month, Claude the ceiling for depth work, and custom agents finish what neither seat can.
Report Overview
Most writing about AI platform choice compares features. This compares outcomes against five tests of AI-nativeness, across four paths a firm can actually buy in September 2026: Claude, ChatGPT, Microsoft 365 Copilot, and custom agents in your own Azure tenant. Google Gemini Enterprise is treated as a fifth, because you will be asked about it.
Eighteen concrete tasks are compared side by side, along with governance and residency across five options, what each path really costs for 120 seats, and how long each takes to implement. Every price and regulatory date was checked against vendor documentation between 1 and 3 September 2026.
Key facts
- £24.60
- Per user per month for Microsoft 365 Business Premium with Copilot at 50 to 300 seats, which is the correct SKU rather than the £23.10 enterprise add-on. List price as verified 1 to 3 September 2026.
- 18
- Concrete tasks compared across Claude, ChatGPT, Copilot, Gemini and a custom build, with the verdict on each one stated rather than implied.
- $22
- A month in tokens to run a 500-file compliance review on Claude Sonnet 5 through the Batch API. The run cost is rarely what stops these projects.
- 62
- Pages of research behind the deck, checked against vendor documentation and regulator publications between 1 and 3 September 2026.
Every price and regulatory date on this page is as verified between 1 and 3 September 2026.
01 · WHAT IT MEANS
What does AI-native actually mean?
Four things true at once, and a fifth a board recognises. Everyone has a governed assistant inside the tools they already use. The firm's method is encoded rather than remembered. The systems of record are reachable through one connector you own. The heaviest process runs itself, with a named person signing it off. Then the board gets one page a month with numbers on it.
02 · THE ANSWER
Which platform should a regulated firm choose?
Copilot is the floor at £24.60 per user per month, the cheapest way to give every seat a governed assistant. Claude is the ceiling for the twenty to forty people whose work is long documents, spreadsheet models or code. ChatGPT is a strong second surface with the widest brand recognition. Custom agents finish the work no seat product starts on its own. Prices as verified between 1 and 3 September 2026.
03 · THE WORK
What does the work actually look like?
Eleven pieces of work regulated firms ask us for, on one complexity scale, from an afternoon's setup to a four-month build. Tier 1 a champion does unaided. Tier 4 needs compliance in the room from week one. They are worked examples of the shape rather than a product list, so your list will differ even though the ladder will not.
04 · THE HONEST PART
Why do these programmes stall?
Five constraints, and not one of them is the model. Permission sprawl that Copilot makes legible. The consumption line under the licence line. A residency question nobody minuted. A rubric nobody wrote down. Adoption stalling once the licences are on. All five are decided before the build starts, which is why the deck spends a slide on each.
Why Download This Report?
This comprehensive guide is more than just statistics; it is a strategic roadmap. Inside, you will uncover:
The comparison as a table, not adjectives
Eighteen tasks a firm like yours does every week, with what each path can finish and where it stops. Governance and residency side by side across five options.
What each path costs for 120 seats
The annual licence and the consumption line underneath it, for Copilot, Claude, ChatGPT, Gemini and one custom agent, in the currency each vendor publishes.
Eleven worked examples on one scale
From an afternoon's setup to a four-month build, each scored on budget, ease, scale and security. They are examples of the shape, not a product list.
The twelve-month sequence
Permission and label work first, seats next, the connector after that, and one build once the firm has learned how it wants to review agent output.
Get the deck
Twenty-one slides carrying the whole argument: the five tests, the comparison, the eleven worked examples, the fee bands and the twelve-month sequence. Free, with no obligatory follow-up. Behind it sits the 62-page research document, which we bring to the conversation.
Continue exploring
Claude implementation guide
The same argument with Claude as the chosen platform.
Microsoft 365 Copilot guide
The correct SKU, the permission work, and the credit meter.
AI automation and custom agents
What a build looks like when no seat product finishes the job.
Compliance file review, automated
A four-stage review agent for a financial advice firm.
Frequently asked questions
What does AI-native actually mean for a mid-market firm?
Four things being true at once. Everyone has a governed assistant inside the tools they already use, seeing only what they are already permitted to see, with the activity evidenced. The firm's procedures and house language exist as something a new joiner runs on day one, rather than as four long-serving people's habits. The case system, the CRM or the back-office platform can be read and written through one connector the firm owns. And the workflow that consumed the most skilled hours runs as an agent in the firm's own tenant, with a named person signing and an evidence trail a regulator can read. There is a fifth test a board recognises, which is that the board gets one page a month with numbers on it.
Which AI platform should a regulated firm choose?
For a Microsoft-resident firm the honest answer is a combination. Copilot is the cheapest way to give every seat a governed assistant inside Word, Outlook and Teams, at £24.60 per user per month on Business Premium with Copilot for 50 to 300 seats. Claude is measurably better on long documents, spreadsheet work, encoded procedures and code, so it goes to the twenty to forty people whose work is that. ChatGPT does much the same job with wider brand recognition and a UK storage region on Enterprise. The mistake that costs the most money is choosing the combination by brand loyalty rather than by the shape of the work. Prices as verified between 1 and 3 September 2026.
Is it lawful for a UK or Irish firm to use an AI platform that stores data in the United States?
Yes, on the route most firms already use for other United States software. Claude's first-party platform stores data in the United States, which is lawful under Anthropic's data processing agreement with Standard Contractual Clauses and the UK Addendum, recorded in a DPIA and a transfer risk assessment. Neither the FCA nor the Central Bank of Ireland prohibits it. It bites where a client mandate or a public-sector contract demands UK or EU-only processing, or where health data is involved. For most firms of this size it is a documentation task rather than a barrier. Position as verified between 1 and 3 September 2026.
What does AI actually cost for a 120-person firm?
On list prices verified between 1 and 3 September 2026: Microsoft 365 Business Premium with Copilot at £24.60 a seat is about £35,400 a year for 120 people. Claude Enterprise at $20 a seat is $28,800 a year if you buy all 120, although most firms buy 20 to 40 of those seats. One custom agent on a high-volume process is a £15,000 to £30,000 build with a run cost under $150 a month. The line that surprises finance directors is not the licence, it is the consumption meter underneath it. Grounding one Copilot message in your own tenant costs 10 Copilot Credits.
Can Microsoft 365 Copilot do everything, so we only buy one thing?
It reaches everything in your tenant and it is governed by controls you already run, which nothing else matches. It is thinner on open-ended reasoning, and Excel is its weakest surface. Its agents are also chat-shaped: they start when a human types. Work that starts when an email lands, a file is dropped, or the clock strikes 6am on the first of the month is a build on any platform, Copilot included.
How long does this take?
Copilot Chat is days. A paid Copilot rollout is two to six weeks once the permission and labelling work is done, and that work is the actual first project. Claude or ChatGPT seats are days, with one Entra consent for the Microsoft 365 connector. A Copilot Studio agent is two to six weeks. A custom agent is three to sixteen weeks depending on the band. The twelve-month plan in the deck puts them in the order that works.
Is the deck really free?
Yes. A name and a work email, and the 21 slides open in a new tab. No paywall and no obligatory follow-up. The 62-page research document behind it is not a download. We bring it to the conversation if we have one.
The other guides
Pick the one that matches your stack
Each guide covers one platform in the same shape: what AI-native means, what you actually buy, the eleven worked examples and where that platform is the wrong answer.
Claude Implementation Guide for AI Native
Making your firm AI-native with Claude
21 slides · free
Get the deck
Making your firm AI-native with Microsoft 365 Copilot
The baseline to set before any agent: the correct SKU at £24.60 per user per month, the oversharing work that comes first, and the credit meter that decides what it actually costs.
21 slides · free
Get the deck
Making your firm AI-native with ChatGPT
The baseline to set before any agent: which plan clears a regulated bar, where the UK region stops, and what a product line that has changed shape four times in twelve months means for anything you build on it.
21 slides · free
Get the deck
AI-native on Microsoft: Copilot and Claude
Copilot as the floor for every seat, Claude as the ceiling for the people whose work is depth, and custom agents in your own Azure tenant for the work neither seat can finish. One Microsoft stack, three layers, one bill you can explain.
21 slides · free
Get the deck
AI-native on Microsoft: Copilot and ChatGPT
Copilot as the floor for every seat, ChatGPT as the ceiling for the people whose work is depth, and custom agents in your own Azure tenant for the work neither seat can finish. One Microsoft stack, three layers, one bill you can explain.
21 slides · free
Get the deck