Multiverse and Transport for Britain brought senior leaders from the private sector, the public sector and politics to the National Railway Museum for a closed-door roundtable on one big question: how do you move from AI ambition to AI adoption?

Rail has no shortage of AI ambition. Turning it into operational change is the hard part, and the evidence is clear: around 95% of AI pilots fail. At the National Railway Museum in York, delegates met under Chatham House Rule to work out why, and what the sector should do differently. Dr Laura Gilbert CBE gave the keynote, Peter Longman of Transport for Britain set out the sector's wider ambitions, and Alan James gave an inspirational peek at Hyperloop.

The venue suited the question. The museum is a global benchmark for heritage rail conservation and a custodian of industrial innovation, with roots in the 1968 Transport Act's consolidation of Britain's retired mainline steam assets. Delegates debated the next chapter of rail's engineering story among the exhibits that tell the earlier ones.

Start with the problem, not the technology

Most pilots fail for the same reason: teams start with "let's use AI" instead of a problem worth solving. Handing every employee a Copilot licence changes nothing on its own.

Start with a different question: should this process, project or programme exist at all? If so, ask how AI can speed it up or change it. If not, AI will not rescue it. Either way, anchor the work in a real operational challenge, define what success looks like and decide how to measure it. The rule that stuck: "Don't start with AI being the answer."

Six questions to ask before building

1. Do you know what the problem is?
2. Can you build a solution for it?
3. Do you have the right people?
4. How will you monitor it and prove the return?
5. What are the benefits, and who gets them?
6. Can you deliver it?

One test overrides the rest: was it built with the people who own the problem? A strong business case will not save a project that lacks influence, or one designed without its users in mind.

People decide whether AI is adopted

Adoption depends on practical skills, and on employees who feel empowered to use AI in their day-to-day roles. It also depends on focus. On the diffusion of innovation curve, innovators and early adopters carry the change, and the rest follow, or never will. Win those first, then use AI to help people decide and act.

One technique worth borrowing: AI agents for each group affected by a decision, then test every scenario against all of them. Teams surface objections and pressure points before the idea reaches real people. AI becomes a rehearsal space for difficult decisions, not just a faster source of answers.

Collaboration scales what works

Rail and transport organisations face many of the same challenges, so there is little reason to tackle them alone. Sharing knowledge and working across organisations speeds up innovation and helps the sector scale what works. Radical transparency is the aim: the closer organisations get to it, the more freely they can share what works and what does not. Partnerships between organisations are opening up a new market, and the prize is a public sector that uses everything AI can offer.

Transport for Britain's vision gives that collaboration a destination: a fully integrated and sustainable network of roads, railways, cycleways, pathways and waterways that meets the current and future transport needs of the UK population. That is why the conversation reached beyond rail. A joined-up network needs joined-up skills.

Beyond productivity gains

The biggest opportunity goes beyond productivity. Used well, AI can improve the reliability, efficiency and quality of services for the millions of people who depend on Britain's railways.

What the room said

The ambition in the room matched the stakes. Transport is critical to everyone in the UK, and the leaders at the table, with others who will join them, are shaping its future vision, not just discussing it. Their candour made the session: delegates traded real experience of what works, what stalls and why. One theme ran through every contribution. AI is a people challenge and a cultural challenge, and across the sector the technology is often ready before the organisation around it is.

Ultra high speed ground transport

Alan James gave the room an inspirational peek at Hyperloop and Ultra High Speed Ground Transport (UGT). A former Vice President for Worldwide Business Development at Virgin Hyperloop One and CEO of the UK Ultraspeed maglev project, he now advises governments on multi-decade infrastructure investment and is widely acknowledged as one of the world's leading experts in UGT. His talk put the day's question in a longer frame: the decisions made now will shape UK transport for decades.

‍

What happens next

The technology is working. Adoption depends on people. The group will reconvene in the new year to focus on the AI use cases already making a measurable difference in transport, sharing both the wins and the lessons learnt. Multiverse is bringing apprentices from right across the sector together to upskill in Data and AI as part of a united rail cohort, the first of its kind in UK rail, with content customised for the sector. Co-sponsored by Transport for Britain, the cohort will turn that shared ambition into skills.

‍

Earlier this week, we published new data showing that British workers lose 28 working days a year to inefficient work with data. 97% of the workers in our study want to improve their AI skills, while three in four UK adults have never had any AI training.

Knowing where the skills gaps are is one challenge. Closing them is another. Today, at our Workforce 2030 conference in London, we launched Mastry to bring everything together into one platform.

Why AI adoption stalls between investment and results

Buying the tools is the easy part. The difficult part is getting people to use them well in their day-to-day work. A plan on paper doesn’t change how someone works the next day, and neither does training that sits apart from the work. That leaves leaders assembling the answer themselves, from a consultancy, a training provider… and a lot of hope.

What Mastry does

Mastry is an adaptive learning platform that runs a company’s AI workforce change from start to finish. It finds where the skills gaps are, teaches people in the course of their actual jobs, and shows leaders the impact the learning has delivered.

• Diagnose: Each person has a short voice or chat conversation with our AI agent about their role and how they work, and leaders get a view of capability across the organisation, mapped to business goals.
• Guide: Mastry builds a learning plan for each person from what the business needs and what they can already do, using their own work projects as the material. Mae, our AI guide, works alongside people inside Claude, Microsoft Copilot and other workplace tools. She offers guidance on live tasks and records evidence of new skills as people apply them.
• Prove: leaders see progress across teams, so they can report what their AI investment is producing.

‍

Learning that sits inside the work

Jay Richman, our Chief Product and Technology Officer, said: “If learning competes with work, work will always win. So we built Mastry to sit inside the work.”

Mastry can be used on its own or alongside Multiverse apprenticeships, which continue to be delivered with expert human coaches. Both will be run from the same platform.

Built on ten years of training people at work

Multiverse was founded ten years ago as an alternative to university. With Mastry, we are taking a modern approach to ground long held by management consultancies: a single partner for a company’s AI workforce change.

Euan Blair, our founder and CEO, said: "Companies have bought the AI tools, but what most still lack is a workforce that can use them to make a difference. Until now they've had to rely on external consultants, with results often not materialising. We focus on installing expertise across the workforce so they can drive the change themselves, and Mastry puts that provision in one place. It's built on our deep expertise training workers on the job and insights from hundreds of thousands of real workplace projects."

Frequently asked questions

When will Mastry be available?

Mastry is available now to eligible development partners through a waitlist, with full launch planned for early 2027.

How does Mastry work?

It finds where the skills gaps and organisational blockers are, teaches people in the course of their actual jobs, and shows leaders the impact that the learning has delivered in real time.

Mastry works alongside people inside Claude, Microsoft Copilot and other workplace tools. Mae, Mastry’s AI guide, helps people with the task in front of them, so they build skills while doing real work, not in a separate course.

Does Mastry replace human coaches?

No. Mastry runs alongside Multiverse apprenticeships, which are still delivered with expert human coaches. Coaches bring accountability, pastoral care, group coaching and more, and Mae, Mastry’s AI guide, helps with the work in the moment.

Ask Multiverse learners what makes their programme work and most of them will talk about their team of coaches and instructors. Learners give our coaches a 97% satisfaction rating, and coaching quality is what we guard most closely.

That takes two things: a clear view of how every session goes, and delivering personalised, point-of-need development for every instructor. Until now, one constraint limited both.

The limits of a monthly observation

A Delivery Manager could observe each instructor about once a month. One session gives a narrow view. It can miss a problem that is building, and it gives little basis for development tailored to the person. Managers also spent much of their time gathering evidence, which left less for acting on it. Schools and training providers everywhere face the same problem.

"As a former teacher, I know how hard it is to build an observation system that helps people improve. In my first years in the classroom, I was observed once a month, with feedback based on a 20-minute snapshot of a single lesson. It always felt like putting on a performance, so the feedback didn't reflect what my lessons looked like day to day. Later, when I ran Delivery Enablement at Multiverse, I observed every member of the team once a fortnight. I loved giving that feedback, but I knew the approach couldn't scale. Compass is the result of nearly a year spent working on that problem. Every instructor now gets feedback based on how they teach in every session, and we have a much clearer picture of quality across all our programmes."

Chloe Barker, Senior Director, Quality, Multiverse

What Compass does

The Compass agent is an AI tool that reviews instructors' sessions against the Compass Framework, the skills required to deliver exceptional workshop-based training that we use to hire, train and develop every coach. Before, a manager could review one session a month. Compass reviews every session and combines what it finds with learner feedback.

It produces two things. First, a review for each session, which shows managers where quality may be at risk so human managers can review the session and check if the concern is valid - acting quickly to remediate and mitigate if needed. Second, suggested development areas, which give each instructor individual feedback, with data drawn from across their teaching practice, not just a snapshot.

The first protects the standard learners and employers expect. The second helps instructors raise it. Neither is a verdict on anyone's performance, but both enable us to spot positive trends and innovations across our delivery that can be recognised and promoted.

Where people stay in charge

Our AI principles commit us to keeping a human in the loop and to AI that is open about how it works. We designed Compass around both.

Delivery Managers review what Compass produces, check it against what they know and decide what to do, if anything. The instructor and their manager agree development goals together. Compass is one important data point among many. Performance management rests only on human observation, and the tool makes no decision about anyone's job.

Rolling it out in stages

We've brought Compass in gradually, so that instructors understand it before it plays any part in their development.

• August: we built the reports and tested them with Delivery Managers only.
• September: we introduced the tool to instructors and explained how it works. Reports stayed visible to managers only.
• October: reports move into our internal coaching platform, where instructors will see the same information as their manager.

Later, we plan to use the same approach for learner progress reviews and for Cohort Coaches, and to link what Compass finds to the training we offer our coaches.

In rolling it out, we've reviewed and learnt from similar tools being rolled out across education, including academic research, Stanford's M-Powered Teachers, and IRIS Connect.

Part of a wider approach

Compass follows the path we took with Atlas, our AI guide. Atlas now resolves 88.3% of routine support queries and has answered more than 1.5 million questions, with a 99% helpfulness rating from learners. The share of coach time spent on routine support has fallen from 41% to 18%.

That time has gone back into the work only people can do: mentoring, pastoral care and developmental coaching. We've also kept adding people, hiring 40 new coaches and instructors in a single quarter.

"AI adds real value in the flow of work, for time-constrained tasks and the urgent things you need help with right now. That work does not need a human doing it. But at the core of it, the human coaches are the ones who then have more time to work with learners on their goals, their ambitions, their values, what motivates them."

Kelly Griffith, SVP of Coaching & Delivery

Compass applies the same idea to how we develop our coaches and check their work: AI collects the data and takes a first pass through it, but people take a judgement, and then decide how to act.

What this means for learners and employers

Learners get sessions held to a consistent standard, with problems caught in days where a monthly observation might have taken months. Employers get clearer assurance about the quality of the coaching their people receive. Instructors get personalised development built on their real teaching, which is what keeps the standard rising.

Read our AI principles, or visit our Impact Hub to see how Multiverse coaching turns skills into results.