AI adoption
Make your company AI-native
Becoming AI-native is four pieces of ordinary work: identifying the tasks where AI would make a useful difference, establishing what that is worth and what it costs, configuring tools that suit how your company actually runs, and helping people use them well enough that the change survives the first busy month.
Softblues is a London-based AI practice. We work with knowledge-work businesses across Microsoft 365 Copilot, Claude, ChatGPT, Google and custom builds, and the first conversation is about your objective rather than a platform.
- Anthropic Partner Network member
- OpenAI Select Partner
- Google Cloud Services partner
- Member of the Microsoft AI Cloud Partner Program
Where the value is
What could improve?
Useful AI work starts from a constraint somebody already complains about. Every benefit falls into one of five categories, and naming the category early stops the same benefit being counted twice.
- Capacity released
- Proposal preparation that takes a senior person most of two days, much of it assembling things the company has already written.
- Cost avoided
- An outsourced service or an overtime line kept on to absorb a volume peak that a prepared workflow would flatten.
- Contribution gained
- A team that turns down work because it cannot prepare enough proposals, where the demand it is turning down would carry margin after the cost of serving it.
- Risk and control improved
- Customer-service answers that vary by whoever picks up, with no record of which policy version was applied.
- New service enabled
- A monitoring or review service clients keep asking for that the firm has never been able to price, because doing it by hand often enough to be useful would cost more than anyone would pay.
Those are hypothetical situations, written to show what each category covers. None is a client, and there are no measured results in them. What any of them would be worth in your company is the question the business case calculator answers, from your own baseline.
The operating model
What does AI-native mean?
The phrase gets used loosely, so here is the version we work to. How much depth each test needs varies: a 30-person firm with one expensive process clears the fourth with a single workflow, and a 250-person firm with four of them has more to do.
Everyone has a governed assistant
People reach it inside the tools they already work in, and it is configured so it surfaces only what each person is already permitted to open. Getting the permissions to that state is a piece of work in its own right rather than something a licence confers. The aim is a governed tool good enough that nobody has a reason to paste client data into a consumer app.
The company method is written down, not remembered
The procedures, the checks and the house language exist as something a new joiner can run on day one, rather than as four long-serving people and their habits.
The systems of record are reachable
The case system, the CRM or the back office can be read and written by the assistant rather than re-keyed by hand.
The heaviest process runs itself, with a name on it
The workflow that eats the most skilled hours runs on a trigger, with a named person approving anything that leaves the building.
Management sees one page a month
Use, quality, cost and value in one place, reviewed by whoever owns the objective. This is what separates running on AI from piloting it.
This is the operating model we work towards, and each test is a piece of work rather than a setting. It is not a description of any one client, and no company we work with is presented here as having reached all five.
The five stages
Where do we start?
Five stages, each with one decision and something you keep at the end of it. Read them in order from a standing start. If the platform is already chosen, go straight to rollout; if use is uneven, start at improvement.
Evidence
What does this look like when it has been done?
Two cases, and it is worth being precise about what each one proves.
- Key Capital
An Irish corporate finance and investment management firm. Rolled out in April 2026 with the client's IT provider and data-protection team: sign-in, the Microsoft 365 connector, permissions and governance boundaries. Softblues then led nine weekly training sessions built around the firm's own work.
Evidence that a governed rollout and a training programme were delivered. The case describes that work. It does not report measured time savings, cost savings or financial return.
- Softblues
We run our own departments on AI, with our methods written down as something the team runs rather than remembers. It is why we can describe the last two tests from practice rather than from a vendor deck.
Proves we use what we sell. It is our own company, so read it as practice rather than as an independent result.
Which platform should we look at?
It follows from the work rather than the other way round, and choosing by brand is the expensive mistake. The selection page carries the criteria and the five routes worth investigating.
Where should your company start?
Tell us what you want to improve and how the work runs today. We can discuss whether a scoped discovery is the right next step.
Implementation is agreed against the scope discovery defines. Ongoing support runs with a named owner inside your company and a monthly review.
How it runs
- An initial conversationFree. What you want to improve, what you already run, and who would own the result.
- A scoped discovery, if it is the right next stepPaid, because it measures your work rather than discussing it. Sometimes the answer is that something cheaper would settle your question.
- A pilot decision on the evidenceDiscovery returns a defined pilot scope with criteria for stopping as well as continuing. Starting it is your call.
Common questions about AI adoption
What does it mean for a company to be AI-native?
A company is AI-native when five things hold at once. Everyone has a governed assistant inside the tools they already use. The company method is written down as something a new joiner can run, rather than remembered by a handful of people. The systems of record can be read and written rather than re-keyed. The heaviest process runs on a trigger with a named person approving the output. And management gets one page a month showing use, quality, cost and value. Depth varies by company: a 30-person firm with one costly process needs far less of the third and fourth tests than a 250-person firm with four.
Do we need to choose a platform before we start?
No, and choosing first is the more expensive mistake. The tool follows from the work, the data it has to reach, the controls you need and who is going to use it. Plenty of companies already hold Microsoft 365 licences, which narrows the question without settling it. If you want to read before you talk to anyone, the platform selection page sets out the criteria and the routes worth investigating.
How do we know whether AI is worth the investment?
Measure the workflow before anyone touches a tool, price the whole twelve or twenty-four months rather than the licence line, and discount the modelled benefit by the share of people who will genuinely use it every week. The calculator on the business case page does the arithmetic from your own numbers and returns a payback month, or tells you there is not one inside 24 months. It is free and it will sometimes tell you not to build.
We already bought licences and hardly anyone uses them. What now?
Start at rollout and training rather than at the beginning. Three things worth checking before buying anything else: whether each group was given a task worth doing with it, whether the sources it can reach are current and correctly permissioned, and whether a named person owns the result. The support and improvement page covers how monthly measurement then keeps it from sliding back.
Can we start with one process rather than the whole company?
Yes, and if you can name the process in one sentence it is usually the better start. A company that says "every file gets forty checks a month" should go to process discovery under AI Automation and treat wider adoption as the second move. A company whose work is reading, drafting and analysis across the board cannot name one process, and for them broad adoption comes first and the process work follows.