Two real deployments, with the figures.
Our founder, Brent Phillips, led both programmes himself, as the senior lead inside each organisation. One is where the model came from. The other put it to work at scale, with more than a thousand AI agents. Both organisations have been anonymised.
Case studies
Choose a case.
A large UK consumer-facing regulated organisation
1,224 AI agents found, every live one given a named owner, and the controls handed to the organisation’s own team. Within three months.
Read this caseA large UK regulated infrastructure organisation
Where the model came from. Six weeks from design to live governance across more than ten business units.
Read this caseCase study one
1,224 AI agents, brought under control within three months.
A large UK organisation in a regulated, consumer-facing sector. AI was already in daily use across the workforce, and colleagues were building their own agents in four separate tools, with no AI governance in place. Our founder diagnosed the position and deployed the Ambient Agent® governance model, tailored to the organisation’s own requirements, within three months. He designed and installed it himself, without a delivery team.
What was found.
The number kept growing while the work was under way. Finding and checking agents by hand does not scale, which is why a control system was needed.
1,224 agents at the first count
705 of them live. Nobody had counted them before.
109 live agents with no owner
No recorded owner meant no accountable person if one went wrong.
No sharing data on about 9 in 10
Nobody could show who could see what.
No policy and no forum
Nobody had the authority to approve or block an agent.
What was installed.
The Ambient Agent® governance model, with the organisation’s own requirements built in.
One register
Every agent recorded in one place, built from the reports in the platforms already in use, so every figure put to the board traced to a source.
A named owner for every agent
Every live agent was given an accountable owner. When an owner moves on, the agent is found and reassigned. That rule is now part of how the organisation governs AI.
Rules, run every week
A governance rulebook written for the organisation, run as a weekly review that rates each agent and sends it to the right level for a decision.
Learning for everyone
Mandatory AI awareness learning for the whole workforce, with role-based pathways and AI champions.
Value measured, not assumed
A method approved by the finance team. Time released for people and cost avoided by agents are reported separately and never blended.
Reporting for the board
Weekly reporting for the executive, for change leads, and for governance and risk.
What was measured.
Every figure comes from the programme’s own reports to its steering committee.
From first diagnosis to the governance model running, tailored to the organisation.
At the end of August 2026, across four build platforms. Every one reviewed.
109 at the first count. By early October 2026, every live agent had a named, accountable owner.
One agent helped the same team handle 18.6% more complaints, with no extra headcount.
By that one agent, under a value method the finance team approved.
Of the assigned workforce, by 21 September 2026. The evidence Article 4 of the EU AI Act asks for.
One agent, measured end to end.
The yearly cost avoided by a single agent, under the value method the organisation’s finance team approved. Figures are rounded down.
The first agent built for a business process helped staff resolve complaints. With it, the same team handled 18.6 per cent more complaints than in the same months of the year before, with no extra headcount. That is 18.6 per cent more customers helped.
To value it, real complaints were handled twice, once by hand and once with the agent, and both were timed. The time saved was priced at a fully loaded hourly rate, then halved to stay cautious. The finance team approved the method.
Diagnosed, installed and handed over.
The model was running within three months. The organisation’s own team then took the weekly review over. In one of its own reviews it assessed 33 agents and sent 14 to its governance forum for a decision.
- Every agent counted and checked: 1,224 found, and 109 live agents with no owner.
- The governance model deployed and tailored to the organisation: register, owners, rules, weekly review and learning.
- The organisation’s own team runs the weekly review. The controls keep running without us.
Case study two
Where the model came from.
A large UK regulated infrastructure organisation with many business units. Our founder led the work in a senior role. The control approach proven here is the origin of the six-engine model, written down afterwards from what worked.
What happened, in order.
Three steps, taken from the project records. The line is a sequence. It is not drawn to scale.
- Eight weeks. Every business unit and group function. 80 or more people engaged and 250 or more documents reviewed.
- Six weeks from design to live governance, reporting, capability and enablement, running at the same time.
- 300 or more people enrolled in externally accredited AI and data learning pathways.
“Data maturity across the organisation is present in pockets of excellence, but this often struggles to get scaled within the home business unit, let alone across the group.”
The findings were clear. Data and AI maturity existed in pockets but could not be scaled across the group. Each part of the group ran things differently. Teams were adopting AI on their own, without governance, access controls or any measure of value. Leadership could not see who could use AI, at what level, or to what effect.
What was measured.
Every figure is drawn directly from the engagement record.
An annual saving identified and documented from a single intervention, noted within the independent maturity assessment.
Assessed across business units, with personalised development pathways.
In externally accredited AI and data learning pathways within six months of go-live.
In the independent eight-week maturity assessment, across all business units.
Brought under a single operating model and governance framework.
From design to live governance, reporting, capability and enablement, running at the same time.
The facts of the case.
Sector
Regulated infrastructure, United Kingdom.
Scope
Multiple operating business units and group corporate functions.
Basis of the work
Led by our founder as a senior leader inside the organisation, which has been anonymised.
Where the model came from
The control approach proven here is the origin of the six-engine model. The engines were written down afterwards, from what worked.
Also put in place
A structured capability assessment, repeated every six months. Access set by capability through the Agent Passport® framework. A set process for proposing and approving uses of AI, with named decision makers and a control gate at every stage. Nothing advanced without evidence.
Designed around
The EU AI Act, the UK Government’s principles for AI regulation, the NIST AI Risk Management Framework, ISO/IEC 42001, and what the FCA and the ICO expect.
Build your own evidence base.
The Diagnostic produces your own map of AI use, your risk position and a first view of workforce capability. That is the evidence your board and your regulator will ask for.