01 / TRANSFORMATION
Make sense of competing priorities.
Examine initiatives, dependencies, evidence gaps and trade-offs to support clearer organisational choices.
INDEPENDENT AI GOVERNANCE & STRATEGIC CONSULTANCYFOR LEADERS WHO NEED ANSWERS THEY CAN QUESTION
AI can sound convincing.
That doesn't make it right.
AI can lose the thread, overlook crucial evidence and turn assumptions into confident recommendations. When that happens in a strategy paper, transformation plan or important decision, the consequences are yours.
ChatAdjust helps organisations get AI work back on track. We combine independent consultancy with governed AI methods to surface drift, test the boundaries of a claim and make the work easier to inspect and improve.
Start with one consequential challenge. We bring your context, rules and people into the process — rather than asking you to trust the model on its own.
Discuss your AI challenge See what we can do ↘
Independent, defined consultancy projects from £4,950 + VAT. Start with a conversation.
CHECK THE CLAIMS. KEEP THE CONTEXT. STAY IN CONTROL.SEE WHAT WE CAN DO ↓
01 / WHAT WE CAN DO
Not another AI demo.
Work worth doing.
AI can draft an answer in seconds. But a convincing answer is not the same as a sound one. We help leaders examine the evidence, spot missing assumptions and make progress on the difficult organisational questions that ordinary AI output cannot settle on its own.
02 / STRATEGY
Know what the evidence supports.
Separate established findings from assumptions and unanswered questions before they become decisions.
03 / OPERATIONS
Do the work that never gets done.
Surface overlooked issues, investigate options and carry complex thinking further than available time usually allows.
But what would it take to trust AI with work that really matters?
02 / THE QUESTIONS THAT MATTER
The opportunity is real.
So are the questions.
Before AI shapes a board paper, a major recommendation or a change programme, you need to know what the evidence supports, where the model has drifted and who has authority to decide. These are practical questions, not just technical ones.
Can we trust the answers?
ChatAdjust checks AI reasoning against defined boundaries, blocks conclusions that break those rules, and makes the problems it identifies visible so people can understand what was prevented and why.
Will it understand our world?
We work with your organisation's context, evidence, priorities and requirements.
Who makes the decisions?
Your people retain authority. AI supports the work; it does not quietly set your policy.
How do we get started?
Bring one consequential problem. We agree the scope, undertake the work and deliver something useful.
To answer that, we began by looking more closely at how AI produces knowledge.
03 / THE THINKING BEHIND CHATADJUST
A DIFFERENT STARTING POINT
We looked at AI differently.
Then built something different.
Those questions led us to look beyond the technology itself. And that’s when we realised something extraordinary. Beneath the engineering, AI was actually a system of knowledge, language, representation and creativity. Suddenly, we saw a completely different way to approach AI. Three breakthroughs followed.
Where can AI reasoning fail?
We identified 26 points of potential failure in how AI handles knowledge and reaches conclusions. Then we built a system designed to prevent them.
KNOWLEDGE & JUDGEMENTHow can a small shift change everything?
Our research revealed how tiny changes in meaning can accumulate across an AI conversation. We developed methods to spot and address this micro-drift before small mistakes become accepted as facts.
LANGUAGE & INTERPRETATIONThen we saw what we could build.
Once we could govern AI, a whole new world of possibilities opened up. We could build AI environments shaped around an organisation's knowledge, priorities and ambitions — and put them to work on challenges that had previously seemed beyond reach.
REPRESENTATION & CREATIVITYAnd ChatAdjust was born.
These ideas became software for ambitious organisational work — shaped by your knowledge, your rules and your judgement.
The discoveries began with a question most people never think to ask: how many different ways can AI reasoning go wrong?
REPORTS FROM THE LAB01 / KNOWLEDGE & REASONING
26POINTS OF
POTENTIAL FAILURE
POTENTIAL FAILURE
DID YOU KNOW?
AI can go wrong in more ways than you think.
We spent months examining how AI handles knowledge and builds conclusions. Our research identified 26 distinct points of potential failure — places where an AI answer can become unreliable even while sounding entirely convincing.
We didn't just catalogue the risks. We developed SGC: a framework designed to recognise and address each of those failure points, setting boundaries before AI carries its reasoning further.
WHY IT MATTERSTrustworthy AI starts by knowing where reasoning can fail — and putting safeguards in place before those failures shape the work.
Identifying where AI reasoning can fail was the beginning. The next question was what we could build once those risks were governed.
04 / HOW IT WORKS
Your reality.
Your rules.
Get to work.
Identifying the problems was only the beginning. We turned those ideas into a working environment designed around your organisation — so its knowledge, rules and priorities can shape what AI does next.
01
Your reality.
Build a structured picture of the context, evidence and relationships that make your organisation distinct.
WORLD / CONTEXT02
Your rules.
Keep priorities, requirements, boundaries and human decision rights visible as work develops.
GOVERNANCE / CONTROL03
Get to work.
Explore complex questions, trace gaps and dependencies, and develop useful, inspectable outputs.
TRAVERSAL / EXECUTIONBut how do we keep that reasoning on track as the work continues? Our second report from the lab shows what we found.
05 / REPORTS FROM THE LAB · 02 / MEANING OVER TIME
A SECOND DISCOVERY WE PUT TO THE TEST
Did you know
AI can drift?
A small shift in meaning can change the course of an AI conversation. We call this micro-drift: an assumption quietly hardens into a fact, or a possibility becomes a recommendation.
We put our micro-drift methods to the test.
Here is what we observed.
104combined governance and drift findings
83distinct shadow reasoning-drift issues
11maximum reasoning-drift propagation depth
WHAT ORDINARY AI DID
It kept going.
In the Northstar test, it recommended rejecting a £25 million transformation and selecting an alternative without an established board decision rule or the required comparison.
WHAT CHATADJUST DID
It knew when to stop.
ChatAdjust upheld the missing decision and reasoning boundaries. Across 25 questions, it recorded 21 distinct governance intervention episodes while its shadow analysis identified 83 reasoning-drift issues.
WHAT THIS MEANS FOR YOUR ORGANISATION
Confident answers aren't always sound answers.
ChatAdjust helps keep AI within the facts, rules and decisions you've actually established — so your team can see when more evidence or a human decision is needed.
About the test: The completed Northstar 25-question test on 19 October 2026 recorded 104 combined results: 21 governance intervention episodes and 83 distinct shadow reasoning-drift issues. The detector tracked reasoning lineage across the conversation to a maximum propagation depth of 11.
Ask for the test report ↗WHAT THE EVIDENCE DOES — AND DOESN’T — SHOW
Our completed 25-question Northstar test demonstrates ChatAdjust enforcing reasoning boundaries in 21 intervention episodes and identifying 83 distinct shadow reasoning-drift issues, including unsupported strategic rankings and a £25 million board recommendation. The combined result is 104, with drift lineage reaching depth 11.
Discuss how we would test this on your work ↗Two different risks. Two ways to govern them. Now the bigger question: what becomes possible when AI works within those safeguards?
Detecting problems is only useful if people can understand the intervention and remain in control of what happens next.
06 / YOU REMAIN IN CONTROL
You stay in
the driving seat.
Your judgement matters.
ChatAdjust supports human-defined boundaries and decision rights. It does not replace organisational accountability.
The work can be inspected.
Audit outputs can show interventions, contested assumptions and important limitations.
Designed to travel.
Structured state is intended to be portable across supported AI environments. Compatibility is confirmed for each engagement.
Your information, agreed upfront.
We establish data access, handling, return and deletion arrangements before sensitive material is shared. Deletion scope is documented for each project.
07 / A SMALL FIRST LOOK
NOT READY FOR A PROJECT?
See what we could find.
Before you commit.
What might this reveal about your organisation? Tell us about one specific challenge. For a limited number of suitable enquiries, we can prepare a free, personalised ChatAdjust Preview using publicly available information.
A short, illustrative analysis to help you see the possibilities — not a free pilot, a verified assessment of your organisation, or a substitute for a commissioned project.
Request a free previewNo obligation. Requests are reviewed individually; availability is limited. Please don't send confidential or sensitive information.
What you'll receive.
- A brief picture of the challenge, based on public sources
- Two or three observations, open questions or evidence gaps
- An example of how ChatAdjust approaches the work
- A suggested next step if deeper investigation would be useful
Typically one or two pages. We identify assumptions and limitations rather than presenting preliminary findings as established facts. A paid project allows us to work with agreed inputs, validate the picture and produce substantive deliverables.
And if you are wondering who is behind this unusual approach, it began with a different kind of AI founder.
08 / THE PERSON BEHIND CHATADJUST
FOUNDER · CHATADJUST
Meet William Watkin.
Professor William Watkin has spent more than 25 years studying how knowledge is constructed, how meaning changes, and how ideas shape what people accept as true.
“As I began studying generative AI, something struck me. Everyone was talking about the technology. But I saw something else.
AI is fundamentally a knowledge system. It works through language, creates representations of the world rather than accessing truth directly, and is, at heart, profoundly creative.
Knowledge, language, representation and creativity. These are the subjects I've spent my academic career studying.
So I began asking: what would happen if we brought that understanding into the way AI actually works?
I set about turning those ideas into software. And ChatAdjust was born.”
William is Professor of Contemporary Philosophy and Literature at Brunel University London. ChatAdjust is an independent business; the university affiliation is not an endorsement.
Discuss a project with WilliamThat research became a practical way of working with organisations, one defined problem at a time.
09 / THE PROCESS
A BETTER WAY TO BEGIN
One difficult problem.
One useful result.
If the preview opens up a useful question — or you already know the problem you want to tackle — the next step is a defined project. We agree the scope, deliverables, boundaries and price before work begins. First Projects start at £4,950 + VAT; larger Transformation Projects start at £15,000 + VAT.
Explore the two project options01
Bring the challenge. Tell us what needs doing and why it matters.
02
Agree the boundaries. Scope, inputs, data handling, price and success criteria.
03
Do the work. Combine your people's expertise with a governed AI environment.
04
Review the result. Inspect the output, limitations and next steps.
So what does it cost to begin?
10 / TWO WAYS TO BEGIN
WORK WITH CHATADJUST
Start with one problem.
See what becomes possible.
This is independent, hands-on consultancy, not a self-service chat subscription. We begin with a problem your organisation needs to solve, agree the evidence and boundaries, and deliver work your team can review. Choose a defined first project or a broader transformation engagement.
01 / FIRST PROJECT
One important problem.
One useful result.
£4,950 + VAT
Starting price · tightly defined engagement
For a leader who wants to test ChatAdjust on a real piece of organisational work, without committing to a larger programme.
- One agreed challenge and limited evidence set
- ChatAdjust configured for the task
- One substantial output with assumptions and limitations
- Review meeting and recommended next steps
Best for a contained first engagement.
Discuss a First Project02 / TRANSFORMATION PROJECT
A bigger challenge.
Deeper work.
£15,000 + VAT
Indicative starting price · expanded scope
For organisations ready to tackle a substantial strategic or transformation challenge with more connected evidence and decisions.
- Deeper discovery and broader evidence review
- Multiple connected questions or workstreams
- Iterative analysis and substantial deliverables
- Executive review and next-stage opportunities
Best for a larger, connected organisational challenge.
Discuss a Transformation ProjectBoth are proposed starting prices, excluding VAT. Scope, timetable, data handling, deliverables and acceptance criteria are agreed in writing before purchase. The same governance approach applies to both offers. These are services, not self-service software subscriptions.
Before you decide, here are the questions we are most often asked.
11 / QUESTIONS
Good questions
deserve clear answers.
Why not just use ChatGPT or Copilot?+
ChatGPT and Copilot generate useful material, but they can also miss context, drift or sound confident about unsupported claims. ChatAdjust brings your organisational evidence, rules and decision rights into the work, with inspectable checks and clear limitations. We demonstrate the difference on a specific task.
Do our staff need AI expertise?+
No specialist AI expertise is required to begin a scoped engagement. Your team brings knowledge of the organisation and its work; we support configuration and delivery.
What happens to our data?+
We agree what information is needed, where it is processed, who can access it, how deliverables are returned and what deletion can be verified. These arrangements must be approved before sensitive data is shared.
Can you guarantee every AI answer?+
No. ChatAdjust is designed to make complex AI work more dependable, inspectable and governed. Human review and task-specific acceptance criteria remain important.
How do we begin?+
Start with a short discussion of one problem. If ChatAdjust is a fit, we define a paid engagement with agreed scope, price, deliverables and evaluation criteria.
If the next step feels useful, tell us what is on your desk.
THE NEXT STEP IS SIMPLEBring us the work
Bring us the work
you need to get right.
You do not need to know how the software works. If AI is producing work you cannot fully rely on — or your organisation has a difficult problem AI could help solve — tell us about it. We can discuss a defined project or one of our limited free ChatAdjust Previews. No technical specification needed.
A direct contact email must be configured before this preview goes live.Not ready for a project? See the limited free preview offer ↗