Blog – Future Processing
Home Blog AI/ML How can a company begin an AI transformation? 
AI/ML

How can a company begin an AI transformation? 

In my experience, many companies struggle because they start in the wrong place: buying a platform, announcing a Centre of Excellence, or launching a pilot project before anyone has checked what problem AI should solve and whether the organisation is ready for it. BCG’s research shows how wide the gap is: only 5% of companies achieve AI value at scale.

Share on:

Table of contents

Share on:

As a process excellence consultant, I do not look at AI as an isolated tool. I look at how it fits into the flow of an organisation, where the real gaps are, where time and money are leaking, and whether the people who will use it understand the change and have a say in shaping it. This article walks through what that looks like in practice, at each stage of the AI transformation journey, from the first honest readiness check to what happens after the first pilot succeeds.

Key takeaways

  • According to BCG, only 5% of companies achieve AI value at scale, and 60% generate no material value at all despite continued investment. In my experience, the reasons are more often organisational than technical.
  • Before choosing a first use case, an organisation needs a clear picture of its own readiness across six dimensions: AI strategy, data, technology, governance, talent and culture.
  • A good first use case solves a specific, measurable problem and fits the process around it. A visible tool such as a chatbot can work well, as long as what happens to its output has been thought through.
  • A pilot only earns its place if it is measurable, bounded and has organisational support before it starts, not after.
  • Time saved by AI is not automatically value created. It has to be redirected deliberately, and the time needed to check AI output belongs in the cost.
  • A Centre of Excellence can select and prioritise use cases, but leadership stays accountable for the direction and results, and the people doing the work should help shape the change.

Why do most companies that begin an AI transformation stall before it starts?

According to BCG’s September 2025 study of more than 1,250 companies worldwide, 60% generate no material value from AI despite continued investment, and only 5% achieve value at scale. In my experience, the gap often opens up at the very first decision, when an organisation chooses where to start.

Readiness is not only about mentality, though that matters too. As a process excellence consultant, I do not look at one piece of the puzzle in isolation. I look at how a flow moves through the whole organisation, across business functions that each have their own pace. One part of a process might work well on its own, but depend on others where there are no measures in place, or on legacy systems where the documentation is missing and the knowledge sits in people’s heads.

When a good tool meets a broken workflow

A chatbot is a good example. It can be an excellent use case and collect valuable feedback from users. But if that feedback lands with a team that spends two weeks reading through it manually because nobody standardised how it should be processed, the value is lost further down the line. The tool works. The workflow around it was never designed to receive its output.

We don't want artificial intelligence for its own sake. We want to solve a specific problem. Without checking where the gap actually is, where we're losing time, there's no point starting.

What does a company need to have in place before beginning an AI transformation?

Technology matters, but much of what determines a successful AI transformation lies around it. That includes a measured baseline and a clear direction from leadership. Just as important is whether people understand why the change is happening and have a hand in shaping it.

How AI governance, management and transformation fit together

These terms often get blurred, and it helps to separate them. AI governance is about direction and accountability: why the organisation is introducing AI at all, which priorities it serves, who can make decisions about it and who answers for the result. Oversight, such as knowing which AI tools are in use, where and for which decisions, is part of governance, but not the whole of it. Management is the day-to-day side: planning, introducing and running the tools within the direction governance has set.

AI transformation draws on both. It includes the practical rollout of tools, but also strategic decisions about how the organisation wants to work. That is why leadership has to be an ally here: the example, the policies and the strategy come from the top.

The distinction becomes very practical when employees feel the organisation is moving too slowly. They often start using public AI tools on their own, sometimes with company data. Deciding what is allowed and on what terms is a governance question, and it needs an answer before the transformation itself can scale.

The six dimensions of AI readiness

Before choosing a first use case, I walk an organisation through six areas:

  • AI strategy alignment: a clear direction connected to actual business goals, not a standalone AI initiative sitting off to the side.
  • Data readiness: data quality and data governance solid enough to support the tool, not just enough for reporting.
  • Technology infrastructure: systems capable of integrating AI into existing workflows rather than running alongside them.
  • Governance: clear priorities, decision rights and accountability for AI, alongside oversight of how tools are used.
  • AI talent: people with the right skills embedded inside the teams who will use the tools, not concentrated only in a central function.
  • Culture: people who understand why the change is happening and help shape it, not just tolerate it.

Why data quality decides how far AI can go

From a data perspective, this is where most organisations are genuinely exposed. Gartner’s February 2025 research found that 63% of organisations either lack, or are unsure whether they have, the right data management practices for AI.

Automating a mess still produces a mess, only faster and more visibly. Before AI models can work reliably on a process, the underlying data has to be in order. Trust in the output builds gradually, as data quality improves, not the other way round.

Readiness is also rarely evenly distributed. In one organisation, an internal survey asked people honestly which AI tools they used, how often, and whether they saw themselves as enthusiasts or would rather not touch AI at all. One function had barely engaged, while in software development people had already started coding with AI assistants on their own initiative.

That variation is normal, and it is often the clearest signal of where the appetite for change already exists. Where people are already experimenting, the sensible move is to make it official and build on it. The assessment has to happen at the level of specific teams and flows, not as a single verdict on the whole company.

Why people need to shape the change

In most organisations I see two groups at once: enthusiasts who have already started experimenting, and sceptics, some of them worried that AI will take their job. Both reactions are understandable, especially with so many headlines about AI replacing roles. If a transformation is simply announced from the top, someone who has done a task the same way for years, and done it well, is unlikely to open the new tool. The most advanced solution will not help if the people expected to use it see no reason to.

That is why AI transformation is as much about managing people and emotions as about technology. People need to understand why the change is happening and help shape it, rather than being assessed on whether they are ready. The same logic applies to AI expertise: it works best embedded inside each business function, close to day-to-day work, not concentrated in one central group.

AI Readiness Assessment

Gain a clear view of how prepared your data is to support and scale AI initiatives in your organisation.

How does a company identify the right first use case to begin its AI transformation?

The instinct to “have AI” is different from the instinct to “fix something”, and mixing the two is where a lot of first use cases go wrong. I always start by sitting down with someone at C-level who genuinely feels that AI will help with a specific issue, rather than responding to general pressure to be doing something with AI. It is a bit like therapy: once you know where it hurts, you can examine it properly and take care of it.

What makes a process step a good candidate

In practice, I look for a step that is measurable, bounded, and has support from the people who own it. Repetitive, standardised steps are strong candidates for automation. Steps that require human judgement, such as a code review or a final decision, should stay with a person in the loop. A visible tool such as a chatbot can be an excellent first choice, as long as it solves a concrete problem and fits the process around it.

A useful exercise is to take a specific process, in marketing, accounting or customer service, and ask how you would design it from scratch, knowing what AI capabilities now exist. That thought experiment usually reveals the real candidate faster than a long workshop does.

How can a company build an internal business case for beginning an AI powered transformation?

Boards and the people championing AI often look at the same investment from different angles. Someone close to the technology tends to focus on what it can do, while a board weighing budget against other priorities focuses on whether it will pay back.

The two perspectives do not always point in the same direction. As a user, I might find the most value in a single place where I can ask one question and reach everything the organisation knows. Leadership, looking at the same budget, may decide the money should go into the development process instead, because that is where the return is clearer, and may prefer people to keep sharing knowledge with each other directly.

Neither view is wrong. What matters is that leadership sets a clear direction and that the people doing the work understand it and can see how their own needs fit into it. A strong business case then speaks to that direction: what the investment costs, what it is expected to return, and over what timeframe.

£1B+ in bookings, 90% of campaign parameters automated

ITV needed an ad management platform that removed manual work without adding complexity. We built Planet V with two cross-functional teams. It now serves 1,000+ monthly users and 200+ agencies, with ML forecasting that beats human predictions by 20%.

What boards expect from AI investment

Deloitte’s 2025 survey of more than 1,850 senior executives found that a typical AI use case reaches satisfactory ROI within two to four years, with only 13% of the most successful projects paying back within twelve months. That is a useful number to bring into a board conversation, because it resets the expectation that AI investment behaves like a typical software rollout.

Why time saved is not the same as value created

Time saved is often the currency used to build the case, and it needs to be handled carefully. You can free up a hundred hours a week in a team and still end up with nothing to show for it, because saved time on its own does not create value. It has to be redirected deliberately, towards higher value work, otherwise it simply gets absorbed.

There is a second trap, which comes from trust. When PeopleCert asked the ITIL community about the challenges of scaling AI, practitioners pointed out that if checking AI output means redoing the work, the expected savings effectively disappear. Any honest assessment of the benefit has to include the time needed to check and correct what the tool produces, not just the time it saves.

The opposite can happen too. If a team used to handle 90% of its capacity and, with AI tools, is suddenly expected to deliver 110%, there is no space left to review what the tools produce. That is when unchecked output starts slipping through. Verification needs a protected place in the workflow, with a human in the loop and agreed quality thresholds, rather than being squeezed out by higher targets.

What should a company's first AI pilot look like?

A pilot earns its place before it starts, not after. The discipline I bring to this as a process consultant is close to what is sometimes called a “Success Contract”: name the metric, measure the baseline, set the threshold for success, and decide in advance what happens if the pilot misses it. As a rough guide, a single, well-scoped workflow can often reach a production decision within a few months, for example 60 to 90 days, but the right timeframe depends on the scope, the data and the people involved.

If the starting point cannot be measured, or the team does not want the change, I would rather delay the pilot than force it through. A pilot set up that way will not produce a result you can trust, whichever way it goes.

Why moving too fast is its own risk

Pace is a risk in itself. In Gartner’s April 2026 research on AI in IT infrastructure and operations, only 28% of projects delivered their promised return, and many of the leaders who reported a failure said they had “expected too much, too fast”. The data covers one specific area, but in my experience the same lesson applies more widely: keeping the first pilot small enough makes its result trustworthy, whichever way it goes.

Nothing gets forced. If we see the organisation isn't ready, in the data or in the people, we hold. Burning the budget on a pilot that was never set up to succeed doesn't help anyone.

How does a company know whether its first AI adoption initiative has worked?

Knowing whether a pilot worked means checking the baseline against the result, understanding the real cost including everything that surrounds the tool itself, not just its licence fee, and tracking whether people are using it without being told to.

What the evaluation package should include

A proper evaluation covers six things:

  • Baseline measurement of the process before AI was introduced.
  • AI performance, measured against that same baseline, using the same units.
  • Gap analysis showing what genuinely improved, rather than what looks improved.
  • Full cost, including licences, implementation, and the time spent checking and correcting output, set against the savings.
  • User adoption, tracked honestly rather than assumed from a dashboard.
  • A clear decision: scale, extend, or stop.

The full cost line is where the verification point from the business case comes back. A pilot that ignores the time spent checking AI output can look successful on a dashboard while delivering very little in practice.

Who owns which number

McKinsey QuantumBlack’s April 2026 measurement framework gives each layer of value a named owner: financial impact sits with finance, strategic outcomes with the business unit lead, operational KPIs with the process owner, user adoption with the frontline manager, and technical performance with the delivery team. I use a lighter version of the same logic: without a named owner for each layer, nobody checks the number, and the evaluation quietly does not happen.

The AI will still make mistakes from time to time. The right level of trust depends on the use case and should rest on evidence and ongoing evaluation. Clean, well-organised data, a working human in the loop where judgement is needed, and monitoring that continues after the pilot ends all help justify that level of trust, but none of them guarantees that every output is correct.

What comes after the first pilot and how does a company move from starting to sustaining an AI implementation?

A successful pilot buys credibility, and that credibility is worth spending deliberately, not letting it sit. Gartner research published in June 2025 found that 45% of organisations with high AI maturity keep AI projects running in production for three years or more, compared with just 20% of low-maturity organisations. Regular measurement of value helps sustain results after a pilot, alongside how use cases are chosen, governance and engineering practices.

Where most organisations lose momentum scaling AI

In my experience, the pilot itself is rarely the hard part. What slows organisations down is everything between a working proof of concept and full deployment: who owns the result, whether the data holds up at a larger scale, and whether the people who will use the tool were involved in shaping it. Those are organisational questions rather than technical ones, which is why they are so easy to underestimate.

Building a Centre of Excellence that helps

The structural question that comes up at this stage is what to do with the Centre of Excellence idea. It is an appealing model, and it can work well. The risk is that it becomes a convenient place to pass every decision to, while the people doing the work are left out. A Centre of Excellence can and should have real authority, including the right to select and prioritise use cases. What it should not do is take over ownership: leadership remains the owner and sponsor of the transformation, responsible for its direction and results.

What a Centre of Excellence should be, in my view, is a group built from enthusiasts already inside each function: someone in marketing who already uses AI daily and surfaced that in an internal survey becomes the ambassador for marketing. That person becomes a co-creator of the change, already knows how the team works, and can speak its language when it comes to adoption.

The same discipline that selected the first use case has to keep operating here: keep leadership and the people doing the work aligned on one direction, and resist the pull to scale AI for its own sake. The transformation that sustains itself is the one that keeps solving specific problems, one measured step at a time, rather than one that declared victory after the first pilot and stopped checking.

FAQ

How long should a first AI pilot take?

There is no fixed rule. As a rough guide, a single, well-scoped workflow can often reach a production decision in around 60 to 90 days, but the timeframe depends on scope, data and the people involved. What matters more is defining the baseline and the success threshold before the pilot starts.

More often organisational readiness than technology: data that is not in order, teams that were not involved in shaping the change, or a use case chosen for visibility rather than for the gap it closes.

It can be. A chatbot is a strong first use case when it solves a specific problem and fits the wider process, including what happens to the information it collects. If its output lands with a team that then has to process everything manually, the value is lost further down the line.

It can select and prioritise use cases, and often should. Ownership of the transformation itself, its direction and its results, stays with leadership, and the Centre of Excellence works best when it is built with the people who do the work.

Not on its own. Operational efficiency, such as a shorter cycle time or a lower cost per case, is one signal, but it has to be checked against the full cost, including verification time, and against real user adoption. A pilot that only shows efficiency gains, without those two checks, is not yet ready to scale.

Clear accountability for decisions and results, defined boundaries for what AI is and is not allowed to do, a working human in the loop where judgement is needed, well-governed data, and monitoring of what happens after deployment, not just during the pilot.

Value we delivered

72

cost reduction after a seamless migration (within a 20-day timescale)

Let's talk

Contact us and transform your business with our comprehensive services.