If you don’t know where you’re going, any road will get you there.
That old line keeps coming to mind in conversations with leadership teams about AI, because the journey has usually started long before anyone has settled on a direction. By the time the subject reaches a board agenda, it is already in the workflow.
AI is no longer waiting for a board-approved transformation programme. It is already shaping how work gets done. Someone in sales drafts the first pass of a proposal with it. A marketer uses it to get past a blank page. A finance analyst leans on it to phrase a variance more clearly for the management pack. A support agent runs an awkward client reply past it before hitting send.
None of that is a problem in itself. In most cases, it is people doing exactly what we would want them to do: working faster and spending less time on the repetitive parts of the job.
The harder issue is what leadership can actually see. Most executives know, in a general way, that their people use AI. Far fewer could tell you where, for what, with which data, and to what effect. That distance between activity on the ground and visibility at the top is where both risk and wasted spend tend to gather - quietly, until something forces the question.
The scale of it is not really in dispute. Microsoft and LinkedIn’s 2024 Work Trend Index put adoption at 75% of knowledge workers, with 78% of them bringing their own tools to work rather than waiting for an approved one. Tellingly, 59% of leaders in the same study said they worry about quantifying the productivity gains of AI. Read together, those figures describe a familiar gap: a great deal of usage, and far less certainty about its value.
So the question for leadership has shifted. It is no longer whether AI will turn up in the business - that argument is over. It is whether the business can see clearly enough what is already happening to guide it.
Informal use isn’t the problem to solve
Almost nowhere does AI adoption begin with a launch plan. It begins with workarounds: a manager tidying up meeting notes, a consultant turning a messy brain-dump into something a client can read, a developer with an AI assistant open beside the editor, an analyst asking for a dense paragraph in plainer English.
This is sensible behaviour, not rebellion. People are carrying more email, more meetings, more documents and more status updates than the day was built to hold, and AI happens to be good at exactly that kind of load.
The trouble only starts when these helpful habits quietly settle into the way work gets done, without anyone having decided that they should.
Knowing that staff “use AI” is not the same as knowing the things that actually matter:
- which teams rely on it and for what
- where the work is flowing
- what data is being pasted in
- whether anyone reviews the output before trusting it
- whether the tool the company pays for is the one people actually open
- whether a free browser tab is still doing the real work on the side
Left unmanaged, all of that adds up to a mix of enthusiasm, inconsistency and exposure.
The instinct to crack down rarely helps. Heavy-handed bans mostly push the same behaviour somewhere harder to see. The more durable fix is duller and more effective: make the safe path the path of least resistance.
A pattern we see often
Strip the names off enough engagements and a recurring picture emerges, especially in services firms.
Leadership signs off on an enterprise AI tool. The board ticks “adoption” off the list. A short note goes round reminding everyone not to put confidential material into public tools. On paper, the bases are covered.
On the ground, it is looser. Sales does its account research in the approved tool but still rewrites proposals in a free one because it feels quicker. Finance polishes management commentary with AI, with no agreed line on which figures or client names can go in. Operations summarises process notes with it, and nobody checks the summary before it is reused three documents later.
None of this is scandalous, and none of it means people are being careless. It simply describes a company with plenty of AI activity but very little visibility into it - one that has bought access without landing real adoption, and issued guidance without ever connecting it to the actual work.
That, far more than any dramatic failure, is where many leadership teams find themselves today.
Owning the tool isn’t the same as adopting it
Plenty of organisations have already bought, or are weighing up, the obvious names: Copilot, ChatGPT Enterprise, Gemini, Claude. Often that is the right call. But a licence settles only one question: can people use AI?
It says nothing about whether they use it well, use it safely, use it where it genuinely creates value, or whether anyone is tracking the result.
This is the point where adoption tends to stall. The tool arrives, the launch generates a flurry of interest, a handful of confident users sprint ahead, most people poke at it once and drift back to old habits, a few quietly keep using the free version they already know, and middle managers are left with no clear sense of what “good” looks like.
Access is everywhere; adoption is patchy. That is not really a software problem - it is an operating-model one.
For a CFO, the discomfort is twofold. There is the uncontrolled-usage risk, and there is the plainer question of value. A tool that is paid for and barely used is simply cost with no return. If people stay on informal tools because they are easier, the business ends up carrying a governance gap and a wasted licence at the same time.
Resist the urge to do everything at once
The moment leadership grasps how far AI has already spread, the temptation is to answer in kind: every department, every use case, every risk, every tool, all at once.
It looks impressive on a steering-committee slide. In practice, it usually turns slow, costly and vague.
Far better to pick one function and go deep. Choose somewhere AI is clearly already in use, where the work is repetitive or document-heavy, or where you suspect both real upside and real risk - sales, operations, finance, support, HR, legal operations, delivery - wherever the pressure is most visible.
Then ask the unglamorous questions. Where is AI actually entering the day, and through which tools? What data is involved? Where is the work slow or inconsistent? Where could it genuinely speed things up, and where could it quietly create a problem? What needs governing first, and what could realistically move in the next 30, 60 and 90 days?
Working one function at a time keeps things honest. It gives you a scope small enough to act on and evidence solid enough to justify scaling. And the aim is not only to reduce risk; it is to find that first workflow where AI sharpens speed, quality or consistency without introducing new problems to manage.
That is the kind of adoption a leadership team can actually stand behind: visible, useful, and measured.
Governance that helps the work, not governance that watches it
The standard reflex to AI risk is to write a policy. Useful, but rarely sufficient on its own.
A policy in a shared folder changes nothing. One pitched too broadly is no help to someone making a call in the middle of a task. And one drafted before anyone has looked at how AI is really being used tends to govern a company that does not exist.
The weakest policies are almost always the ones written first and observed never.
Governance earns its keep when it is wired into the workflow: when it is clear what data can go into which approved tools and what must never be entered; when people know which output needs a second pair of eyes; when ownership is named rather than assumed; and when there is an obvious route to get a new use case approved instead of smuggled in.
Good governance is not there to stop useful work. Done well, it simply makes the right thing the easy thing.
Five questions worth answering before you spend more
Before committing further budget, a leadership team should be able to answer five questions without hand-waving.
- Where is AI actually being used? Not where the policy imagines it is - where it really is.
- Which tools are in play? Approved ones, personal ones, free ones, browser extensions, and the AI quietly baked into software you already own.
- What data is going in? This reaches well beyond personal data: commercial terms, client details, contract language, financial commentary, strategy and operational know-how all count.
- Is anyone checking the output? Fast is not the same as right, complete, or fit to put in front of a client.
- Which function should you fix first? Pick the one with clear workflows, realistic use cases and an obvious next step.
If you cannot answer these yet, that is not a reason to panic. It is a reason to go and get visibility.
Where WorkLex AI fits
WorkLex AI is RSVR’s governed AI-adoption service for everyday business work. It exists to move leadership teams away from informal, patchy, hard-to-measure usage toward something safer and easier to stand behind.
The entry point is deliberately small: the AI Workday Teardown.
It is not a transformation programme, a training course, or a compliance audit. It is a short, senior conversation aimed at finding where AI is already entering one part of the business, and whether a closer look would actually be worth your time.
It helps surface which function has the clearest issue or opportunity, whether informal use is already widespread, whether the approved tools are genuinely being adopted, whether there are data or output-quality risks worth attention, and whether there is a real case for a focused diagnostic.
Sometimes the honest conclusion is: “Not yet - you do not need a diagnostic.”
We are fine with that. The point is not to manufacture a project; it is to tell you whether there is a real gap worth closing.
Where there is, the usual next step is a WorkLex AI QuickScan: a focused, single-function review of current usage, workflow opportunity, data and governance risk, and adoption readiness - ending in a practical 30/60/90-day plan you can act on.
The bottom line
AI is already part of the working day, whether or not the organisation has formally decided how it should be used. The only question left for leadership is whether you can see how it is being used, steer it safely, connect it to the work that matters, and tell whether it is actually paying off.
Most companies are not standing at the start of this. They are somewhere in the middle of it, often without having consciously chosen to be.
The next move is not to do everything everywhere. It is to pick one function, look hard at the workday, and find where AI is already turning up, what it is risking, what it is worth, and what to do about it first.
Start with one function.
That is exactly what the AI Workday Teardown is built to surface. If your leadership team wants a clear read on where AI is already entering one function - and whether it is creating risk, value, or both - start with an AI Workday Teardown from RSVR.
Source
Microsoft and LinkedIn, 2024 Work Trend Index: AI at Work Is Here. Now Comes the Hard Part, May 2024.
The report states that 75% of knowledge workers use AI at work, 78% of AI users bring their own AI tools into the workplace, and 59% of leaders worry about quantifying AI productivity gains.
Reference: Microsoft & LinkedIn 2024 Work Trend Index
