The Transformation Programme Finished. The Business Didn’t. 

The Transformation Programme Finished. The Business Didn’t. 

Why the next phase of digital transformation has to embrace constant change 

I thought we had stopped talking about digital transformation. Not stopped doing it, of course, but stopped using the phrase. For a while the language moved on. We talked about cloud, SaaS, automation, modernisation and data platforms, and then AI arrived and managed to dominate almost every technology conversation going. Recently, though, I have noticed digital transformation making a comeback. I am hearing it again in customer conversations, seeing it in strategy documents and board discussions, and reading it in the research coming from some of the major consultancies.

EY’s 2026 CEO Outlook, for example, found that 52% of CEOs were already undertaking significant transformation initiatives, with another 45% planning to begin one during the year. 

That is quite a resurgence for a phrase that, not so long ago, felt slightly tired. 

What interests me more than the terminology, though, is what has changed underneath it. The last major wave of digital transformation was largely about becoming digital: moving processes online, replacing legacy applications, adopting SaaS, shifting infrastructure to the cloud and automating work that had previously been manual.

This next phase feels different. 

It is less about becoming digital and more about making all of those digital investments work together intelligently. And I think that requires us to revisit one of the assumptions we carried through the first wave: the idea that transformation has a finished state. 

The programme finishes. The business carries on. 

Businesses do not reach a steady state. A system implementation can finish, a migration absolutely should finish, and a transformation programme needs an end date if anybody ever wants to escape the steering committee. But the business itself carries on changing around all of those things. 

New products launch. Businesses are acquired. Customer expectations change. Pricing models evolve. Regulation changes. Teams restructure. Someone introduces another SaaS application because it solves a problem that did not exist when the original architecture was designed. A system that seemed strategic three years ago becomes legacy.

AI introduces a completely new set of requirements that nobody put into the transformation roadmap because, at the time, they did not exist. That is not disruption to the plan. It is simply how businesses work and every one of those changes affects the data landscape too. 

I think this is where some of our traditional transformation thinking starts to break down. We have tended to work from a current state towards a target state: define where we are, define where we want to be, build the roadmap between the two, complete the work and declare the transformation delivered. 

That model works well for projects. I am less convinced it works for data. Because by the time you reach the target state, the target has usually moved. 

The problem with the idea of ‘data ready’ 

You hear it all the time: we need to get the data ready first. 

It sounds completely sensible. Before we start doing anything ambitious with analytics, reporting or AI, surely we should clean the data, finish the warehouse, integrate the systems, agree every definition and establish our single version of the truth. Then we can start. But start when? 

Imagine an organisation genuinely achieved the perfect data estate.

  • Every relevant system integrated.
  • Every duplicate removed.
  • Every field completed.
  • Every definition agreed.
  • Every owner identified.
  • Everything documented and governed. 

Come back six months later. 

A new product has launched. A business has been acquired. There is a new CRM field that nobody has populated consistently. Two teams are now using the same term to mean slightly different things. A customer has started asking for something the original model never anticipated. The organisation has introduced a new platform and suddenly a dataset that barely mattered a year ago has become strategically important. 
Perfection lasted about five minutes. 

This is why I think the concept of being data ready can sometimes be unhelpful. It implies a binary state: not ready, then ready. Real organisations do not work like that. They sit somewhere on a constantly shifting spectrum. Some data is excellent, some is questionable, some is incomplete and some is beautifully governed but entirely irrelevant to the question anybody is currently trying to answer. 

There is also an important distinction between data being perfect and data being useful. 

A dataset can be incomplete and still be perfectly adequate for a particular decision. Equally, a dataset can look immaculate, with every field populated and every record neatly structured, and still be wrong. 

So rather than asking, is our data ready?, I think the more useful question is: 

Ready for what?

  • Ready to produce a statutory financial return? The bar should be extremely high. 
  • Ready to let an AI agent make a decision with significant financial or customer consequences?  Again, you need very strong controls, governance and traceability
  • Ready to understand whether customers are using a new product feature? You may already have more than enough information to start learning. 
  • Ready to automate a management report that currently takes somebody two days every month? Perhaps the relevant data needs some work, but that does not mean every source across the business needs to be perfected first. 

Different outcomes need different levels of confidence, completeness and control. 
That is not lowering the standard. It is applying the right standard to the job. 

Waiting for perfect data can become its own form of inertia 

This is where I think organisations can unintentionally get stuck. 

A useful analytics project is held back because the enterprise warehouse is not finished. A team continues producing reports manually because the wider migration programme still has eighteen months to run. A customer-facing analytics capability stays on the roadmap because not every possible source has been consolidated. 

Meanwhile, the business continues to change and the original requirements become less relevant by the month. 
This is one of the reasons I think transformation needs to create value while it is happening, rather than storing all of that value up for some future completion date. 
If a transformation programme lasts three years, something should get better during those three years. 

  • Maybe one reporting process becomes automated. 
  • Maybe one customer journey becomes properly visible. 
  • Maybe product and finance finally agree the handful of definitions they genuinely need to share. 
  • Maybe operational data and commercial data are brought together for one important use case. 
  • Maybe a predictive model becomes viable because the information required for that model has been brought to the standard it needs. 

Then you move to the next thing. 

That incremental approach is not a compromise. In many cases it is how you discover what genuinely matters, because using data has a habit of exposing problems that architecture diagrams and data dictionaries do not. A dashboard reveals that two teams have different definitions of an active customer. A report highlights a field that is missing much more often than anyone realised. A model exposes a historical categorisation problem. A customer asks why their number differs from the number on an invoice. An analyst tries to join two datasets and discovers that the identifiers do not behave in quite the way everyone assumed. 

Those discoveries can feel like evidence that the data was not ready. 

I see them differently. The problem existed already. Using the data exposed it. That gives you a chance to fix the thing that actually matters. 

Use. Learn. Improve. Use again. 

To me, that is not failed data transformation. That is data transformation. 

Data transformation is better thought of as a capability 

This is probably the biggest shift in thinking I would advocate. 

Perhaps data transformation should not primarily be thought of as a programme at all. Perhaps it is a capability an organisation needs to develop. 

A programme assumes a current state, a target state and a defined body of work that moves you between the two. A capability assumes that the environment will keep changing and that the organisation therefore needs to become good at responding to that change. 

That means being able to integrate a new source when it appears, change a definition when the business model evolves, improve data quality where poor quality is genuinely blocking value, retire something that is no longer needed, and leave data exactly where it is when moving it would achieve nothing. It also means recognising that the architecture itself will evolve. 

We often talk about target architecture as though the target is stationary. It is not. The architecture you designed three years ago reflected the business, technology, skills and economics of three years ago. Since then, AI has changed the importance of unstructured information, customer expectations have changed, cloud capabilities have moved on, new applications have appeared and the organisation itself may have changed shape. 

McKinsey’s 2026 Global Tech Agenda talks about leading organisations moving away from annual technology planning towards continuous strategy development, with business and technology teams increasingly shaping priorities together throughout the year. 

That makes complete sense to me. 

If business strategy is becoming more continuous, the data landscape supporting it has to be capable of continuous change as well. 

AI has made it harder to pretend the problem is solved 

AI did not create fragmented data. Businesses have been living with fragmented, duplicated and inconsistently defined information for years, and humans became remarkably good at compensating for it. We know who to ask. We know which spreadsheet is the real spreadsheet. We know that the figure in one system needs adjusting because it has always behaved that way. 

It might not be elegant, but organisations learn. AI does not have that organisational scar tissue. 

When we start asking technology to interpret information, recommend actions, predict outcomes or take action on our behalf, the weaknesses in the underlying data become much more visible. 

McKinsey’s 2026 research on AI data readiness found data to be one of the biggest constraints organisations face as they try to move from AI pilots into scaled deployment. That is hardly surprising. If AI is expected to make sense of a business, the information describing that business has to be accessible, understandable, appropriately governed and reliable enough for the task. 

But even here, I think we need to resist another version of the perfection trap. The answer is not to stop AI activity until every piece of enterprise data is pristine. It is to understand which data a particular use case depends on, what level of confidence that use case requires, and what needs to change to get there. 

The same principle applies. Solve the data problem that matters to the outcome. Then learn from what happens. 

We also need to stop making every data problem enormous 

Technology has an unfortunate habit of discovering a problem and immediately designing the largest possible solution for it. 

  • Need better reporting? Build a warehouse. 
  • Need more data? Build a lake. 
  • Need AI? Apparently now we need a lakehouse. 


I am being slightly flippant, but only slightly. 

There are organisations where those investments are absolutely the right answer. Complex data estates can require sophisticated engineering, enterprise architecture and significant governance. Nobody should pretend otherwise. But architecture should be proportionate to the problem. If data is already secure, structured, accessible and reliable enough in an operational system, perhaps the most sensible thing is simply to use it there. 

Moving it because the architecture diagram says everything eventually belongs in one central box does not automatically create value. It does, however, create another copy to secure, govern, reconcile and maintain.

On the other hand, if the information needed for a business outcome sits across four systems, those systems use different structures and the data needs to be combined, reshaped and standardised before it becomes useful, then some form of transformation is clearly required. Neither approach is inherently better. They solve different problems. This distinction has influenced our own thinking at Panintelligence. 

Historically, one of our core architectural principles has been simple: if customer data is accessible and analytics-ready, we query it where it lives. We do not believe data should be moved simply for the sake of moving it. 

I still believe that. 

But the conversations we have with customers have evolved. We increasingly work with organisations whose ambition for analytics, reporting or AI has moved faster than their data infrastructure. They know what they want to achieve, but the required information may be spread across several operational systems, may need restructuring or joining, or may not yet exist in an analytics-ready form. 

The unhelpful answer would be: come back when the data transformation is finished. Except I do not think it ever really will be. That is why our approach has evolved too. Where the data is usable in place, Panintelligence can continue to query it there. Where specific information needs extracting, combining or transforming before it can support the outcome, PiConnect provides that preparation layer. 

And if the requirement is much bigger than that, perhaps what the organisation genuinely needs is a broader enterprise data programme or specialist partner. That is fine too. The principle matters more than the product: solve the data problem you actually have, not the one an architecture trend tells you that you ought to have. 

Start with what needs to change 

For all of the technology involved, I think the best transformation conversations still begin with a remarkably simple question: 

What needs to be different? 

  • Perhaps customers need to understand their own performance without raising a support ticket. 
  • Perhaps the board needs to know what is happening now rather than receiving an answer three weeks after asking the question. 
  • Perhaps an operational team needs to identify a problem before a customer does. Perhaps product needs to understand whether a feature is actually being adopted. 
  • Perhaps finance needs to remove a manual reporting process that consumes days every month. 
  • Perhaps an AI application needs controlled access to a particular set of business information. 

Start there, then work backwards: What information does that outcome depend on? Where is it? Can we access it? Do we understand it? Which bits genuinely need improving? How accurate does it need to be for this particular use case? What controls are appropriate to the risk? 

Those questions create focus. 

Instead of telling a data team to “fix the data”, you can tell them which data matters, why it matters and what good needs to look like. That is a much more useful brief. 

Perhaps completion is the wrong thing to measure 

There is one final implication to all of this. If transformation is continuous, perhaps percentage complete is not the most useful way to measure it. 

We love those numbers:

  • 85% migrated. 
  • 70% integrated. 
  • 95% of fields mapped. 
  • Programme green. 


But if the business still cannot make the decision that justified the investment, what exactly is 95% complete? 

I am increasingly more interested in measures such as:

  • How quickly can we incorporate a new source?
  • How long does it take to answer an important new business question?
  • How many manual reporting processes have disappeared?
  • How quickly can the organisation adapt when a commercial definition changes?
  • Can customers get to the information they need?
  • Can teams trust the measures they use?
  • How much of the work completed for one use case can be reused for the next? 

Those are not really project metrics. They are measures of organisational capability. And perhaps that is the point. 

There is nothing wrong with a transformation programme finishing. It needs milestones, accountability, budget and ultimately an end. The mistake is believing that programme completion means the organisation has arrived at some permanent transformed state. 

It has not. 

Hopefully it has moved a long way forward. But the environment will move again, and so will the business. 
The organisations that handle that best will not necessarily be the ones with the grandest target architecture or the longest transformation roadmap. I suspect they will be the ones that become very good at changing: understanding what data matters now, improving it where it needs improvement, using what they already have where they can, and transforming what genuinely needs transforming. And doing all of that while continuing to create value. 

So yes, digital transformation appears to be back. Perhaps that is useful. But if we are entering another phase of it, I think there is one idea we should leave behind from the last one: the idea that eventually we will be finished. 
We won’t. 

The transformation programme may finish. The business won’t. 

Neither will the evolution of the data underneath it. 

The goal is not perfect data and it is not a perfect architecture. It is an organisation capable of continuously turning the data it has into something useful, trustworthy and valuable, while adapting as that data inevitably changes. 

That is not unfinished transformation. ..That is what transformation looks like.