Electrical Tape, Wind, and the Reality of AI Governance

Greg Coleshill - Chief Commercial Officer

Electrical Tape, Wind, and the Reality of AI Governance

I don't mind an early start but I definitely do mind traffic. With that in mind I planned an early start for the drive up to North Yorkshire. I thought, "get there early, choose somewhere that looks good, and treat yourself to a Full English with some locally farmed bacon and sausage".

Well, that didn't work out.

Turns out that not many places are open at 7.30 am in small Yorkshire villages. After a brief panic I found a lovely looking hotel. I got in to find my Full English plan completely foiled and that I had walked into the UK's only completely plant-based hotel.

For those of you who know me this wouldn't be my first choice but I can honestly say it was amazing. Great setting (right by a stream), lovely food and amazing service from the owners. Beck Hall in Malham, it was a beautiful spot and I highly recommend it.

After that I headed to meet Paul, Helena and Alan and it is fair to say things did not go exactly to plan.

We started by discovering we couldn't access the place Paul had planned for filming. No problem, there's a spot on the top explained Paul.

We all jumped in our cars and headed up, but best laid plans continued to fail.

It was incredibly windy on the top, the camper awning broke, and the show is usually a two-person format, so we found ourselves one chair short for a three-person panel.

With some electrical tape (I went full dad mode) and strategic camper van positioning Paul somehow worked his production magic to keep the audio clean and the shoot moving.

Morning chaos over, we sat down to unpack the FCA Mills Review and what artificial intelligence actually means for financial services leaders operating without a dedicated AI team and it was a really enjoyable conversation.

The Mills Review and the Delegation Gap

It is clear that firms want to capture the benefits of this technology. Recent industry data shows 98% of firms are active and 63% are already piloting agentic AI.

However, progress is being held up. Almost half of those firms are stalled by concerns over accuracy, governance, and compliance.

The FCA Mills Review explains exactly where this is heading. The report maps out how the sector will transition by 2030, moving from basic tools to highly autonomous systems.

The review focuses on four major systemic shifts. The most critical is the transformation of firms. AI will move from a peripheral productivity tool to the absolute core of operations and customer service.

The human role inside the firm will fundamentally shift from executing daily tasks to observing outcomes and setting operational parameters.

This trajectory exposes a significant operational risk. We call it the delegation gap.

You can automate the execution of a process. Under the Senior Managers and Certification Regime, you can never delegate the accountability to a machine. If an autonomous system makes a non-compliant decision, the named senior manager is personally on the hook.

The response that the AI made a mistake simply will not cut it. The regulator expects human accountability. This is exactly why treating this transition as a traditional technology rollout fails in regulated businesses.

Stop Installing Software, Start Onboarding Intelligence

Firms need to fundamentally change their mental model. You do not install software. You onboard intelligence.

If you bolt these systems onto existing IT delivery methods, you bottleneck progress. You also distance the accountability from the people who actually understand the regulatory nuance of the work.

Organisational structure dictates success. You must put the digital teammate inside the same reporting line as the human doing the job today. The operational team must own the tool.

AI teammates need a manager. They need someone who is responsible for their performance, training, and improvement. Those managers need the ability to monitor their new digital workforce without having to become deeply technical prompt engineers.

We use a specific framework to guide this transition. We tell clients to onboard, collaborate, trust, and scale. You introduce the digital worker to your policy documents and operational boundaries. Human experts work alongside them to verify their performance. You only scale the capacity once that trust is mathematically proven.

Our unique learning loops enable exactly this. They enable the teammate to learn from the whole team, not just one person updating a prompt in the hope it improves outputs.

The Human in the Loop Reality

Alan brought a brilliant perspective to the panel as a CISO. He noted that trust must be earned in stages rather than assumed on day one.

This requires an active, pre-execution enforcement framework. Retrospective logging is not enough to protect a regulated business. You must implement a strict human in the loop workflow where the AI prepares the work and the human retains absolute authority over the final action.

This approach allows firms to target high-friction, cognitively heavy workflows like complex complaints or underwriting. Efficiency and compliance can move together rather than trading off against each other.

Proportionate Governance

We also tackled the dual mandate facing every Chief Risk Officer today. They must manage the risk while making sure they do not block innovation.

Paul made a crucial point about proportionality. Good governance does not require an enterprise budget or a forty page policy document that nobody reads.

It is about proportionate, practical control. The regulator expects you to know where these systems touch your operations and who is accountable. You must be able to explain a decision to a customer who asks. That is a highly reachable bar.

The future of work is a hybrid one. Those who build the infrastructure to support human and AI collaboration now will be materially harder to compete with in three years.

Greg Coleshill - Chief Commercial Officer

Electrical Tape, Wind, and the Reality of AI Governance

I don't mind an early start but I definitely do mind traffic. With that in mind I planned an early start for the drive up to North Yorkshire. I thought, "get there early, choose somewhere that looks good, and treat yourself to a Full English with some locally farmed bacon and sausage".

Well, that didn't work out.

Turns out that not many places are open at 7.30 am in small Yorkshire villages. After a brief panic I found a lovely looking hotel. I got in to find my Full English plan completely foiled and that I had walked into the UK's only completely plant-based hotel.

For those of you who know me this wouldn't be my first choice but I can honestly say it was amazing. Great setting (right by a stream), lovely food and amazing service from the owners. Beck Hall in Malham, it was a beautiful spot and I highly recommend it.

After that I headed to meet Paul, Helena and Alan and it is fair to say things did not go exactly to plan.

We started by discovering we couldn't access the place Paul had planned for filming. No problem, there's a spot on the top explained Paul.

We all jumped in our cars and headed up, but best laid plans continued to fail.

It was incredibly windy on the top, the camper awning broke, and the show is usually a two-person format, so we found ourselves one chair short for a three-person panel.

With some electrical tape (I went full dad mode) and strategic camper van positioning Paul somehow worked his production magic to keep the audio clean and the shoot moving.

Morning chaos over, we sat down to unpack the FCA Mills Review and what artificial intelligence actually means for financial services leaders operating without a dedicated AI team and it was a really enjoyable conversation.

The Mills Review and the Delegation Gap

It is clear that firms want to capture the benefits of this technology. Recent industry data shows 98% of firms are active and 63% are already piloting agentic AI.

However, progress is being held up. Almost half of those firms are stalled by concerns over accuracy, governance, and compliance.

The FCA Mills Review explains exactly where this is heading. The report maps out how the sector will transition by 2030, moving from basic tools to highly autonomous systems.

The review focuses on four major systemic shifts. The most critical is the transformation of firms. AI will move from a peripheral productivity tool to the absolute core of operations and customer service.

The human role inside the firm will fundamentally shift from executing daily tasks to observing outcomes and setting operational parameters.

This trajectory exposes a significant operational risk. We call it the delegation gap.

You can automate the execution of a process. Under the Senior Managers and Certification Regime, you can never delegate the accountability to a machine. If an autonomous system makes a non-compliant decision, the named senior manager is personally on the hook.

The response that the AI made a mistake simply will not cut it. The regulator expects human accountability. This is exactly why treating this transition as a traditional technology rollout fails in regulated businesses.

Stop Installing Software, Start Onboarding Intelligence

Firms need to fundamentally change their mental model. You do not install software. You onboard intelligence.

If you bolt these systems onto existing IT delivery methods, you bottleneck progress. You also distance the accountability from the people who actually understand the regulatory nuance of the work.

Organisational structure dictates success. You must put the digital teammate inside the same reporting line as the human doing the job today. The operational team must own the tool.

AI teammates need a manager. They need someone who is responsible for their performance, training, and improvement. Those managers need the ability to monitor their new digital workforce without having to become deeply technical prompt engineers.

We use a specific framework to guide this transition. We tell clients to onboard, collaborate, trust, and scale. You introduce the digital worker to your policy documents and operational boundaries. Human experts work alongside them to verify their performance. You only scale the capacity once that trust is mathematically proven.

Our unique learning loops enable exactly this. They enable the teammate to learn from the whole team, not just one person updating a prompt in the hope it improves outputs.

The Human in the Loop Reality

Alan brought a brilliant perspective to the panel as a CISO. He noted that trust must be earned in stages rather than assumed on day one.

This requires an active, pre-execution enforcement framework. Retrospective logging is not enough to protect a regulated business. You must implement a strict human in the loop workflow where the AI prepares the work and the human retains absolute authority over the final action.

This approach allows firms to target high-friction, cognitively heavy workflows like complex complaints or underwriting. Efficiency and compliance can move together rather than trading off against each other.

Proportionate Governance

We also tackled the dual mandate facing every Chief Risk Officer today. They must manage the risk while making sure they do not block innovation.

Paul made a crucial point about proportionality. Good governance does not require an enterprise budget or a forty page policy document that nobody reads.

It is about proportionate, practical control. The regulator expects you to know where these systems touch your operations and who is accountable. You must be able to explain a decision to a customer who asks. That is a highly reachable bar.

The future of work is a hybrid one. Those who build the infrastructure to support human and AI collaboration now will be materially harder to compete with in three years.