DPGLabs.aiDIGITAL PROCUREMENT GARAGE
← All insights
Opinion

Most organisations don’t have a technology problem. They have a pilot-to-production problem.

WHY PILOTS FAILThe four gaps between pilot and production.PILOTPRODUCTION1234The data gapIt worked in the pilot becausethe data was clean.REALITY CHECKProduction data is messy,inconsistent, incomplete andconstantly changing.The integration gapIt worked because it satoutside the enterprisearchitecture.REALITY CHECKProduction means realsystems, dependencies,constraints and APIs.The ownership gapIt worked because a heroicproject team carried it.REALITY CHECKWhen the project team moveson, so does the momentum.The adoption gapIt worked because users wereunusually motivated.REALITY CHECKIn real life, people are busy,change is hard, and old habitsdie hard.BRIDGING THE GAP TAKES MORE THAN A GOOD IDEA.It takes design, discipline and determination.© Adam Brown, 2026
Why pilots fail: the four gaps. © Adam Brown, 2026
I once saw a pilot described as “ready to scale” because it worked beautifully with twelve suppliers, one category manager, three spreadsheets and a dataset that had clearly been cleaned within an inch of its life. That is not production, that is procurement karaoke.

1. Procurement has become very good at innovation theatre

Once you see it, it becomes very difficult to ignore. Across large organisations, procurement has become remarkably good at what can only be described as innovation theatre, where there is no shortage of innovation labs, startup showcases, AI hackathons, pilot programmes, proof-of-concept initiatives, and carefully curated “innovation days” where vendors demonstrate how their latest product will change everything. The slides are polished, the language is modern, the energy is real, and the ambition is rarely in question.

Ideas are celebrated, pilots are launched, awards are handed out, LinkedIn posts are written; and yet, when you step back six or twelve months later and look at how the organisation actually operates, very little has fundamentally changed.

This is the problem; not a shortage of ideas, or a shortage of technology, or even a shortage of effort, it is a shortage of mechanisms. Most organisations lack a structured, repeatable way to take a promising idea, validate it properly, industrialise it effectively, and deploy it into the live operating environment at scale. Most pilots are designed, consciously or otherwise, to answer a single question: “can this work?”. That is the easy question, and in most cases, the answer is yes. The harder questions, the ones that determine whether anything meaningful changes, come immediately afterwards.

  • Can it integrate into the existing landscape?
  • Can it work with real, messy data rather than curated samples?
  • Can it pass security, architecture and governance processes?
  • Can users adopt it in the context of everything else they are doing?
  • Can it scale across geographies, categories, teams and business units?
  • Can it be supported, maintained, evolved and improved over time?
  • Can it deliver measurable value once the initial enthusiasm has faded?

This is where most ideas fail, not because they are inherently flawed, but because they are never designed to survive contact with reality. They succeed in controlled environments, where data is clean, stakeholders are engaged, and complexity is deliberately reduced, but they struggle the moment they are exposed to the full weight of the organisation. The result is now a familiar pattern.

Pilots but not progress. Activity without outcomes. Momentum without movement.

2. Why pilots fail; the four gaps

One of the more subtle issues in this space is the false signal of successful pilots. Because they are designed to succeed, they generate positive signals that are easily misinterpreted as readiness for scale. The technology appears to work, stakeholders are enthusiastic, and the organisation assumes that scaling is simply a matter of time and resource.

In reality, the transition from pilot to production is not a continuation, it is a fundamentally different phase with a different set of constraints. The moment teams attempt to move into production, they encounter the things pilots are very good at avoiding, fragmented data, brittle integrations, unclear ownership, competing priorities, and the simple fact that users have other things to do. At this point, momentum stalls. The pilot remains “successful” in isolation, but the organisation does not realise value at scale. Procurement accumulates tools, rather than outcomes.

The issue is not innovation. The issue is execution.

This is not just anecdotal. Firms such as McKinsey & Company1 and Boston Consulting Group2 have consistently highlighted this dynamic, where organisations become highly effective at experimentation but struggle to translate those experiments into operational capability. Deloitte3 has made a similar point in its work on AI and digital transformation, highlighting that fragmented data, legacy architecture, and weak operating model alignment are the primary reasons initiatives fail to scale.

Organisations are not failing to innovate, they are failing to operationalise. The language differs slightly, but the pattern is the same across four gaps:

  1. The data gap, it worked in the pilot because the data was clean.
  2. The integration gap, it worked because it sat outside the enterprise architecture.
  3. The ownership gap, it worked because a heroic project team carried it.
  4. The adoption gap, it worked because users were unusually motivated.

Organisations that consistently move from pilot to production approach this very differently. They do not treat pilots as isolated innovation exercises; they treat them as early-stage production design.

That distinction matters more than it appears. It means designing pilots within real workflows rather than parallel ones, using production-grade data instead of curated datasets, and defining success in terms of adoption and usage rather than technical capability. It also means planning the path to scale before the pilot even begins, rather than treating scale as something to be addressed later.

Perhaps most importantly, it requires clear ownership from day one. Someone must be accountable not just for running the pilot, but for taking the outcome all the way through to production. Without that ownership, pilots rarely fail dramatically; they simply never become anything meaningful.

3. Changing the model at BT, the Digital Procurement Garage

This challenge was very visible at BT, at a time when the procurement technology ecosystem was expanding rapidly and the ambition was to become a genuinely digital-first procurement organisation, one that could augment its workforce with technology, attract top talent through best-in-class tools, and build a reputation as a leading place to work.

Rather than attempting to evaluate an increasingly complex market from a distance, we made a deliberate decision to change the model and this became the Digital Procurement Garage.

Instead of going to the market, we brought the market to us.

Based in London, with proximity to Shoreditch and a dense startup ecosystem, the idea was to invite early-stage companies, typically seed or Series A, into our environment and give them direct access to a large, complex enterprise setting, including around 200 procurement professionals working on real problems.

The original vision included physical co-location, with startups embedded within our offices, although this was disrupted by COVID and lockdowns, and much of the collaboration continued in a hybrid form. Despite that, the core model held.

For startups, the value was clear, access to a live enterprise environment, real-world use cases, and the opportunity to refine their products in collaboration with experienced practitioners. For BT, the benefits were equally compelling, early access to emerging solutions, the ability to shape those solutions to fit our needs, and the option to move quickly to contract if something genuinely worked, often at preferential rates and with the added benefit of giving the vendor credibility through association with the Garage. This was not vendor theatre. It was a reality test.

We focused on real business problems, particularly in areas such as risk management, where we worked with companies like Darkbeam on cybersecurity risk, Prism on macroeconomic and geopolitical exposure, Versed AI on multi-tier supply chain mapping, and Paid on supplier onboarding and validation. These were not abstract use cases, but tangible challenges the organisation was actively trying to solve.

We contracted with all of them, and several went on to significant success, including acquisition.

The Digital Garage model proved highly effective and variations have since been replicated in different forms by other large organisations, including BP and Johnson & Johnson among others, which I take as a massive compliment and confirmation of the approach. That replication is not particularly surprising, the concept itself is relatively simple. The challenge, as ever, is not in understanding the idea, but in executing it end to end, particularly the often-missing link between experimentation and production.

4. A different mechanism, the same principle, hackathons at Maersk

At Maersk, we evolved the mechanism to include in-house build, but the underlying principle remained the same, how do you move from ideas to real, deployable capability.

Most corporate hackathons produce three things: pizza boxes, exhausted teams and a prototype no one ever opens again, so we designed our hackathons deliberately.

Two full days were dedicated entirely to building, followed by a third day where teams presented their work to a panel of judges, with awards given across categories such as innovation and business value. However, the effectiveness of the model was driven by what happened before the event.

We worked closely with the business, typically two months in advance, to deeply understand the problems, and from that defined five high-level themes, each broken down into five specific challenges. Using spend analytics as an example of a theme, these challenges would cover entity resolution, automated categorisation, taxonomy development, visualisation, and insight generation. Individually, each challenge addressed a component of the problem; collectively, they formed a complete solution.

Each theme had a business sponsor, and a product owner, and in the weeks leading up to the hackathon, these sponsors ran sessions to explain the problem, why it mattered, and how it worked in practice. Dummy datasets were provided, and participants selected their preferred challenges, after which teams were carefully constructed to ensure balance.

We defined the what, the why, and the who. The hackathon solved the how.

Crucially, one of the prizes was not just recognition, but a commitment that the best ideas would be taken beyond proof of concept into minimum viable products and then into live deployment. That is the difference between a hackathon and a pipeline.

5. The reusable model: from one-off success to repeatable delivery

Why this is structurally difficult

On the surface, these models appear straightforward. In practice, both are structurally difficult because they cut across organisational boundaries and require capabilities that do not naturally sit within a single function.

Procurement is designed to run processes, not to build and deploy technology. Technology teams are not always aligned closely enough with procurement to bridge the gap. Governance slows things down and ownership is fragmented.

Ideas are generated in one place, pilots happen in another and architecture decisions sit elsewhere. Change management is treated as something separate and no one owns the end-to-end journey! That is the environment in which pilots die.

Combine this with the fact that the core problem most organisations face is not capability, it is repeatability. They can deliver a successful pilot. Occasionally, they can even deliver a successful implementation. What they cannot do is do it again, and again, and again in a consistent, scalable way. That is where the concept of a reusable model becomes critical.

The system explained

The Digital Procurement Garage is often described as an innovation construct, but that undersells it; it is, in reality, a system and a reusable model that connects the critical elements required to turn ideas into outcomes. It anchors innovation in real business problems, accelerates learning by enabling ideas to be tested and iterated rapidly, and, most importantly, creates a path to production, a practical mechanism that takes an idea from concept through to operating capability. This is not a framework on a slide, but an operational approach that allows an organisation to validate ideas quickly and then industrialise them without starting from scratch each time, standardising how problems are framed, how solutions are tested, how data is structured, how integration is approached, and how adoption is driven.

Most importantly, the system defines the time axis, because without it you don’t get innovation, you get drift. And drift leads to unconstrained proof of concepts that quietly harden into substandard products, which everyone pretends are enterprise ready until reality proves otherwise. We have all been there.

Without that path to production, everything else becomes theatre; with it, innovation becomes a pipeline.

This is what sat behind the Digital Procurement Garage. It was not just a place to experiment. It was a system for turning experimentation into execution. Without a reusable model, every initiative becomes a bespoke effort, dependent on individuals, context, and timing. With it, transformation becomes a capability, not an accident.

Garage Sprint: proving value without overengineering

The Garage Sprint is the front-end of that model. It is deliberately short, typically a matter of weeks, and focused on one thing only: proving whether something is worth doing. Not proving that it can be done, but proving that it should be done. This distinction matters.

In a Garage Sprint, the conditions are intentionally simplified: data is made usable, integration is minimised or bypassed. The focus is on the outcome, not the architecture. The objective is to demonstrate, in a tangible way, that a problem can be solved and that solving it creates real value.

This is where most organisations are actually quite good. Organisations can run pilots, show something working in a controlled environment, but that is also where many stop. The sprint is not the destination, it is the filter and tells you what deserves to be taken further, and just as importantly, what does not.

Garage Build: doing the hard work that makes it real

The Garage Build is where most transformations fail. This is the transition from something that works in a sprint to something that works in the enterprise. It is where the four gaps, data, integration, ownership, and adoption, all have to be addressed properly.

  • The data is no longer curated, it has to be real.
  • The solution can no longer sit outside the architecture; it has to integrate appropriately.
  • The heroic team disappears; ownership has to be embedded.
  • User motivation is no longer artificial; adoption has to be earned.

This is slower, less visible, and significantly less exciting work. It is also the part everyone quietly hopes someone else will take care of! It involves data engineering, architecture decisions, operating model changes, governance, and change management. It is also where the majority of the value is either realised or lost.

The Garage Build is not about scaling the pilot. It is about redesigning the solution so that it can survive in reality. Most organisations underestimate this phase. In practice, it is the phase that determines whether transformation actually happens.

6. Why this matters even more in the age of AI

The recent renewed interest in the Digital Garage model is not accidental. It reflects the increasing importance of being able to translate innovation into operation in a world where AI is rapidly changing what is possible.

AI does not remove the pilot to production gap, it amplifies it. Generative AI has made it easier than ever to demonstrate something impressive quickly with tools such as Lovable or Replit to name but two. You can layer intelligence on top of a limited dataset, connect to a few sources, and create something that looks transformative in a matter of days. But that only increases the risk because the underlying constraints have not changed. Data still needs to be structured, systems still need to connect, be industrialised and be enterprise ready. Ownership still needs to be clear and users still need to adopt it.

If anything, AI is less tolerant of weak foundations. An agent operating in a clean, well-structured environment can deliver extraordinary outcomes. The same agent operating in fragmented, inconsistent data will simply expose those weaknesses faster and at scale. As a result, the gap between pilot and production is becoming more visible, not less. AI is not the shortcut, it is the accelerator, and it accelerates both progress and failure.

Agentic AI in a bad operating model is not transformation. It is a very confident intern with system access and no supervision.

7. What leaders must do differently

From a CFO perspective, failed scaling is not an innovation issue. It is capital inefficiency dressed up as progress.

Most organisations do not lack ideas. They lack a mechanism to take an idea, test it properly, and turn it into something that changes how the business operates. Without that mechanism, innovation becomes theatre; with it, innovation becomes a source of real competitive advantage.

The answer is not to slow down, it is to be more deliberate. Start by shifting the question; stop asking what technology to deploy and start asking what outcome you are trying to achieve and anchor everything to that.

Build a reusable model, not a concept, but a way of working that allows you to move consistently from idea to production. Separate sprint from build, use sprints to prove value quickly, but recognise that the real work sits in the build phase, and plan for it accordingly.

Invest in the foundations: data, integration, ownership, and adoption are not secondary concerns, they are the work. Design for reality: If something only works with clean data, no integration, a heroic team, and highly motivated users, it does not work. Treat AI for what it is. Not a solution in itself, but a powerful layer that depends entirely on the system it sits within.

And finally, for leaders, this is not something you can delegate entirely. The most successful deployments I have seen all share a common trait: senior leaders, whether CPO, CFO, or beyond, take a personal interest, they get hands-on, they understand what the technology can actually do, and in doing so, they set the tone for the entire organisation. Curiosity at the top becomes momentum everywhere else. In the end, transformation is not defined by what you pilot, it is defined by what you run.


  1. McKinsey: The state of AI in 2023
  2. BCG: Strategies to tackle the AI skills gap
  3. Deloitte: 2025 Global Chief Procurement Officer Survey

First published on LinkedIn, 3 May 2026.