Practitioner story · Bellrock

Confidence, workflows, proof: how Bellrock made AI transformation stick.

How Bellrock’s Head of AI turned tools that were switched on into work that has actually changed, by treating AI as 80% a people challenge.

From the Capability Loop podcast, with Rich Merrett. Lessons any leadership team can apply, whatever AI tools you use.

Read the 5 lessons
At a glance
of the AI challenge is people, not technology
80%
policy tiers that say what people can do
4
volunteer AI champions across the business
71
to deliver a feasibility sprint
10 days
The starting point

Licences are live. Has the work changed?

Visit most organisations a year after their AI launch and the picture is the same. Licences are live, there is an acceptable use policy and the launch session was well attended. Ask how a typical team’s work has changed, though, and it goes very quiet.

Rich puts the split at roughly 80% people and 20% technology. Organisations may get to grips with the technology, but if they misjudge the people side the return simply is not there.

At Bellrock he runs a four pillar strategy covering governance, people, solutions and monitoring. Those pillars line up with three conditions we see in every AI programme that has impact: confidence, workflows and proof. Most organisations have one of these, sometimes two. Bellrock covers all three.

80%of the AI challenge is people: confidence, habits and the way work gets done.Rich Merrett · Bellrock
20%is the technology itself, the part most programmes spend their energy on.Rich Merrett · Bellrock
The lessons

Five lessons in AI transformation.

Confidence gets people started, workflows keep them going, and proof shows leadership the effort was worth it.

An AI policy should say what people can do

In Rich’s experience, a lot of AI policies are really security documents with a new title. His governance starts from the opposite position: a clear policy spells out what people are allowed to do, so they feel safe trying.

Bellrock uses four tiers. The lowest covers personal productivity, such as brainstorming or drafting something with no company information or personal data. Higher tiers need a risk assessment, which Rich has made take 15 to 20 minutes on purpose, so anyone deploying something serious has had to think it through.

Blocking tools does not stop people using AI, because everyone has a phone and a personal laptop. Sanctioned access, even if it is limited to brainstorming on free tools, gives the business a foundation it can add controls to later. For teams starting out, he suggests:

  1. 01Agree your risk appetiteBefore anything else.
  2. 02Decide the tiersHow many you need and what belongs in each.
  3. 03Form a small governance groupBring together the DPO, CISO, CTO and legal lead, because views of risk can differ sharply.

Build confidence in the open

Every nine months Rich runs a week of optional half hour lunchtime sessions, open to everyone. Each day builds on the last.

  1. 01What AI isThe basics, in plain language.
  2. 02AI at homeThree school apps in one family calendar.
  3. 03Tools and rulesWhat is available and what the policy allows.
  4. 04Hands onA day of trying it for real.
  5. 05Show and tellColleagues show what they built, without being experts.

A learning week, repeated every nine months.

The home session matters more than it sounds: most people relate to an assistant that pulls three school apps into one family calendar. Confidence also means dealing with doubts openly. Rich raises objections himself before anyone else can, with a slide on hallucinations and a section on sustainability.

Happy for AI to handle the school admin. Never happy for it to read an output for you.

Rich Merrett, Head of AI, Bellrock

Open several routes into the work at once

Training gets people curious; workflows are where that curiosity becomes a different way of working. Bellrock runs several routes at the same time, so change comes from the top, from within teams and from somewhere in between.

Leadership workshops
Managers learn to spot use cases, estimate the return and reassure teams that AI will change their jobs without replacing them.
Build days
Renamed Art of the Possible-athons after finance grew wary of the word hacking. Small teams leave with something usable the next morning.
71 champions
Volunteers with an immersive day away, monthly check ins, their own channel, vendor sessions and a spotlight in a fortnightly newsletter.
Access matched to role
Tools follow role and skill level. A declined request still comes back with what the person could already do with the tools they have.

This is AI moving from single player to multiplayer. One enthusiastic person with a clever prompt helps one person; shared build days, a visible pipeline and a champions network let the whole organisation learn from each other.

Make every AI idea visible, then sort it

Requests, along with a chat assistant for ideas and questions, feed a pipeline of hundreds of ideas the whole business can see. Each is scored and weighed against effort.

  • Impact
  • Strategic fit
  • Time to value
  • Reach
  • Effort
Quick wins
Problems people can now solve themselves, with the tools and skills they already have.
Feasibility sprints
Builds the AI team can deliver in 10 days.
Transformation projects
Larger pieces of work tied to a measurable effect on EBITDA.

Proof is ongoing, for tools and for people

Rich learned the value of monitoring the hard way. When a model provider retired a model his solutions depended on, he did not have enough time to migrate and retest everything. Now every AI solution is checked against the return predicted at the start, every quarter. He tracks model drift, and every six months subject matter experts retest.

A deterministic system keeps doing the same thing. A probabilistic one needs ongoing attention from the people who understand the work.

Rich Merrett, Head of AI, Bellrock

Proof applies to people too. Rich is introducing 12 levels of AI capability, each with a badge, so anyone can see who has trained others and who to ask for help. The skill he values most is careful reading: if someone’s name goes on an AI-drafted report, they have to read it properly. Skip that step and managers become the bottleneck, and the time saved earlier disappears.

What changed

What changed at Bellrock.

All three conditions, confidence, workflows and proof, are now in place and working together.

  • Before: A policy that reads like a security document
    After: Four tiers that say what people can do, with risk assessments where it matters
  • Before: One launch session
    After: A recurring learning week, open to everyone, that starts with life at home
  • Before: Doubts left unspoken
    After: Objections on hallucinations and sustainability raised up front
  • Before: Enthusiasts working alone
    After: Leadership workshops, build days and 71 champions running at once
  • Before: Ideas lost in inboxes
    After: A visible pipeline, scored and sorted into quick wins, sprints and projects
  • Before: Tools shipped and forgotten
    After: Quarterly return checks and six monthly retests by subject matter experts
Questions for your organisation

Take these into your next leadership conversation.

On permission

If you asked five colleagues what they are allowed to use AI for, would you get five different answers?

On confidence

Does your training start from where people already are, and does it name the doubts they have?

On routes in

How many AI ideas from the last quarter were picked up by a team other than the one that had them?

On prioritisation

Can everyone see where their idea sits, and what it would take to move it forward?

On proof

Who is responsible for retesting each AI tool your business already relies on?

Your next step

See where your own workforce stands. Take the free diagnostic.

How AcademyAI helps

Confidence, workflows and proof, in one platform.

Most organisations have one of the three conditions, sometimes two. AcademyAI helps you build all three, with evidence at every step.

01 · Confidence

AI policy training people can act on.

Safe & Responsible Use is one of six capabilities everyone is assessed and trained on. Your own policy is loaded as context, so training reflects what your people are allowed to do.

From this story: A policy should say what people can do.

02 · Workflows

Shared workflows, not lone experiments.

When one person finds a workflow that works, it becomes a team standard, with governance. Role-based courses turn curiosity into a different way of working.

From this story: Open several routes into the work at once.

03 · Proof

Capability levels you can see.

Scored assessments, re-assessment and certification show who can do the work, who has improved and who to ask for help, by team and role.

From this story: Proof is ongoing, for tools and for people.

Capabilities
AcademyAI capability radar scoring a team on the six capabilities

Capability, measured in six.

Every role is scored on Literacy, Safe & Responsible Use, Framing, Specification, Application and Evaluation & Reflection. Works with ChatGPT, Copilot, Claude and Gemini.

  • Baseline every role with a short, scored assessment, not a usage report.
  • See gaps by team and role on one capability radar.
  • Re-assess and certify, so leadership sees what changed.
Questions

AI transformation: common questions.

Something else? Talk to the team.

Why do so many AI transformations stall after launch?

A year on, the licences are live, there is an acceptable use policy and the launch session was well attended. Ask how a typical team’s work has changed and it goes quiet. Rich Merrett puts the split at roughly 80% people and 20% technology. Programmes that stick cover three conditions: confidence, workflows and proof. Most organisations have one, sometimes two.

What should an AI policy for employees include?

A clear statement of what people are allowed to do, so they feel safe trying. Bellrock uses four tiers. The lowest covers personal productivity, such as brainstorming or drafting with no company information or personal data. Higher tiers need a risk assessment, which takes 15 to 20 minutes on purpose, so anyone deploying something serious has thought it through.

How do you govern shadow AI?

Blocking tools does not stop people using them, because everyone has a phone and a personal laptop. Sanctioned access, even if it is limited to brainstorming on free tools, gives the business a foundation it can add controls to later. Start by agreeing a risk appetite, deciding the tiers, and bringing together the DPO, CISO, CTO and legal lead, because views of risk can differ sharply.

How do you run an AI champions network that works?

Bellrock has 71 volunteers. They get an immersive day away, monthly check ins, their own channel, vendor sessions and a spotlight in a fortnightly newsletter. They run alongside leadership workshops and build days, so change comes from the top, from within teams and from somewhere in between.

How do you measure AI ROI after deployment?

Every AI solution is checked against the return predicted at the start, every quarter. Rich tracks model drift, and every six months subject matter experts retest. Ideas are sorted into quick wins, feasibility sprints the team can deliver in 10 days, and larger transformation projects tied to a measurable effect on EBITDA.

Which AI tools does AcademyAI work with?

AcademyAI works with ChatGPT, Copilot, Claude and Gemini. Safe & Responsible Use is one of six capabilities everyone is assessed and trained on, and your own policy is loaded as context.

More from AI practitioners

Practitioner stories from people doing the work.

Sources

  1. Capability Loop Podcast, AI Is 80% People: The Four Pillars Behind Bellrock’s AI Strategy, with Rich Merrett, Head of AI at Bellrock
  2. AcademyAI, Confidence, workflows, proof: three steps to make AI transformation stick
Your next step

Make your AI transformation stick.

Get a capability radar for your organisation and a plan for confidence, workflows and proof.