The cheapest way to start an AI transformation is to spend almost nothing for the first 90 days. Pick one workflow that genuinely hurts, use the AI tools your company already licenses, measure the result honestly, and only then decide what to buy. Money gets wasted when organisations reverse that order: they buy a platform, then go looking for a problem.
Where AI budgets actually leak
Before thinking about how to start, it helps to know where the waste usually happens. In our experience working with managers on AI rollouts, the same four leaks show up again and again:
- Buying before using. Enterprise licences bought for a whole department, then opened by a handful of people.
- Pilots with no finish line. A "test" that runs for nine months because nobody defined what success looked like.
- Consultants writing strategy nobody owns. A polished deck, no internal person accountable for delivering it.
- Training that is too generic. A one-hour webinar on "what is AI" that changes nothing on Monday morning.
Every one of these is a sequencing problem, not a technology problem.
Step 1: Find one workflow worth fixing
Ask each team lead a simple question: which recurring task wastes the most skilled time each week? You are looking for work that is frequent, text-heavy or repetitive, and currently done by people who should be doing something more valuable. Typical first candidates are meeting notes and follow-ups, first drafts of reports or proposals, customer email triage, and summarising long documents.
Choose one. Not five. A single well-chosen workflow teaches you more than a scattered programme.
Step 2: Use what you already pay for
Many organisations already have AI built into software they license: Microsoft Copilot features in Office, Gemini in Google Workspace, AI in their CRM or helpdesk. Check what is switched on before you buy anything new. A few paid seats of a general assistant such as ChatGPT or Claude for the pilot group is usually all the extra spend you need at this stage.
Step 3: Set a baseline and a finish line
Write down, before you start:
- How long the task takes today, and how often it happens.
- What quality looks like (error rate, rework, customer response).
- What result after 60 to 90 days would justify expanding.
This is the step most teams skip, and it is the reason so many pilots drift. Without a baseline you cannot prove value, and without a finish line you cannot stop.
Step 4: Train the pilot group properly
Give the five to ten people running the pilot real, hands-on practice on their own tasks: how to brief AI with context, how to check its output, what data they must never paste in. Two or three focused sessions beat a day of theory. This is where most of the return comes from, because the same tool produces very different results in trained and untrained hands.
Step 5: Review, then scale or stop
At the finish line, compare results to the baseline. If the time saved is real and quality held, write the new way of working down as a standard process, then roll it to the next team. If it did not work, you have lost very little, and you know why. Either outcome is a good use of money.
What to spend on, and when
| Stage | Sensible spend | Avoid |
|---|---|---|
| First 90 days | A few licences, focused training for the pilot group | Enterprise-wide contracts |
| After a proven pilot | Licences for the next teams, internal champions' time | Custom-built AI before off-the-shelf is exhausted |
| Year one onwards | Integration work, governance, wider training | Tools without an owner or a measured use case |
Who should lead it
An AI transformation needs one named person who owns the outcome, usually a manager close to the work rather than IT alone. That person does not need to be technical. They need to understand what today's AI tools can and cannot do, how to run a small experiment, and how to bring sceptical colleagues along. Our Leading AI Innovation & Organisational Transformation course is built around exactly that role, and finishes with a 90-day rollout plan for your own team.
For a deeper look at the costs that appear later, read our guide to the hidden costs of AI implementation.
Frequently asked questions
How much should a first AI pilot cost?
For most organisations, very little: a handful of licences for the pilot group plus training time. If a first pilot needs a large contract, the scope is probably too big.
How long should an AI pilot run?
Sixty to ninety days is usually enough to see whether a workflow genuinely improves. Set the end date before you start so the pilot cannot drift.
Do we need an AI strategy before we start?
You need a clear problem and a named owner. A full strategy is easier to write once one or two pilots have shown you what actually works in your organisation.