What these are
Seven engagements, published to show the kind of work this is rather than to stand as a client list. If one of them describes a room like yours, that is the point of it being here.
The work runs from some of the best known brands in the world to companies nobody outside their own market has heard of. A great deal of it is the second kind: small and medium-sized businesses working out what AI should change, how to give their people room to use it well, or what to tell their own clients. Some are early-stage startups and scale-ups you would only recognise if you worked in that segment. The questions turn out to be roughly the same at both ends, which is most of the argument for taking either seriously.
Clients are unnamed here. Several cannot be named, and where anyone is named anywhere on this site it is with their agreement.
Why this page is written the way it is
A speaker's case study is the least trustworthy document in this market. No control group, no baseline, the outcome described by the person who was paid or the person who paid, and every incentive pointing the same way. The rest of this site grades 367 external sources on what they prove and what they do not prove. Publishing unqualified success stories underneath that standard would say plainly that the standard is for other people.
So each of the seven below carries the same closing line the evidence base uses. Where a number appears, it is the client's own reported figure and is labelled as one. Where the outcome is that people said they found something useful, it says that rather than dressing it as impact.
Clients are unnamed. Several cannot be named. One is a company Rahim genuinely cannot now identify with enough confidence to name, and that is recorded here rather than quietly filled in.
1. Fifteen hundred people, a fortnight of tool churn, and a room in conflict
The situation. A medium-sized technology scale-up, mid-transformation, changing AI tools at a pace the organisation could not absorb. The technical teams were moving. The non-technical teams and support staff had adopted nothing and were frightened of it. The two groups were in open conflict about it.
What was done. Booked at very short notice. A forty-minute version of We Are Superheroes with extended question and answer, delivered from the head office in central London to about 1,500 people dialling in. The brief was to move a divided audience from where it was to where the transformation needed it to be, in one session.
What happened. The session turned into a longer internal discussion rather than ending with the applause. Rahim was brought back to work with the executive team on how the change was being communicated, to support the transformation lead, and to coach team leads who were at the beginning of their own journey with it.
What this does not show. That the talk caused the change. A transformation programme was already running and this was one input inside it. Nothing was measured before or after, and the account of what shifted comes from the people who commissioned it.
2. Fifty people surveyed, and judgement they had already handed over
The situation. The executive team of a company covering Asia-Pacific and India, which had built its own AI tools in-house. Judgement tasks had been cognitively offloaded to those tools without anybody deciding that they should be, and the effect on the quality of decisions had started to show.
What was done. A survey of fifty of their people first, then the findings read back to the executive team in their own words. Then a short workshop in which the team named four or five of their own processes where the offloading had gone furthest.
What happened. They took those processes away and refined them themselves, rebuilding them so the tool augmented the judgement rather than replacing it. The distinction between an augmented and an outsourced approach came out of that room and stayed in their language.
What this does not show. Whether decision quality improved. Fifty responses in one organisation is a diagnostic, not a study, and no follow-up measurement was taken. What it demonstrates is that a team shown its own offloading will usually recognise it, which is not the same as fixing it.
3. Naming what an agency was already doing, in four regions at once
The situation. A large agency business brought its people together from Asia, Asia-Pacific, EMEA and North America into a single event. The business was transforming and multiple stakeholders were describing that transformation in incompatible language, which meant nobody could tell whether they were disagreeing about the plan or about the words.
What was done. Substantial research into what was actually happening across the business, and then the harder half: putting names to the situations. Giving the recurring patterns vocabulary the organisation did not have.
What happened. The company adopted that wording into its ongoing work and its strategy formulation, and retained Rahim as an advisor for three to six months to carry it through the rest of the business.
What this does not show. Any commercial outcome. Shared vocabulary makes a disagreement legible; it does not resolve it. The evidence here is that the language was kept and paid for, which is evidence of usefulness rather than of results.
4. One permitted tool, and what happened when the constraint was examined
The situation. A small division inside a very large company, permitted to use exactly one AI tool. The constraint had been set centrally and had never been revisited, so the division's understanding of what was possible had been shaped by a single product.
What was done. A session setting out what had actually changed in the field, what was changing for their customers specifically, and what was happening in the wider market, so the division could position its own marketing against reality rather than against one vendor's roadmap.
What happened. They moved to a sandbox of several tools, and different functions settled on different ones according to the work they were doing. Rather than one copilot for everything, teams tested across several major models before consolidating on enterprise tools genuinely fitted to their tasks. It was repeated for two or three further divisions in the same company.
What this does not show. That the outcome was better. A division running several tools has more optionality and more governance surface than one running one, and no measurement was taken either way. What it does show is that a constraint nobody had examined turned out to be a decision nobody had made, which is the argument at drift versus design in its smallest form.
5. A school group deciding what to tell parents
The situation. A school group and trust, thinking about AI on two horizons at once: the students they were recruiting now, and the world those students would enter. They were also being handed a steady flow of reports, not all of which were sound.
What was done. A working through of the implications, and an explicit sorting of the material in front of them. Rahim agreed with some of it and disagreed with the rest, and said which was which. That was the useful part: a school leadership team being told which of the reports on its desk would survive scrutiny.
What happened. They arrived at a position they could take to parents. AI as a supplement, and AI as a parallel path that students walk alongside their own developing capability before they reach university and work. That framing became how the group talked about it.
What this does not show. Anything about student outcomes. It is a position adopted, not an effect measured, and the evidence on what AI does to learning is genuinely mixed. That evidence, including the parts that cut against optimism, is at how AI changes teaching.
6. A charity, three hours, and the only number on this page
The situation. A charitable body where most people were using AI personally and nobody was using it at work. Technologically immature by its own description, with no obvious first step.
What was done. A three-hour workshop to find the quickest wins available to an organisation starting from nothing, and to separate what was safe to speed up from what needed to stay in human hands. Then quarterly contact across a year with the key stakeholders and their change lead, checking what had actually been done rather than what had been agreed.
What happened. A year on, they reported a 20 per cent increase in speed on general processes from where they had been, after implementing straightforward changes.
What this does not show, and read this before quoting the number. That figure is the client's own reported estimate. It was not measured by Rahim, there is no baseline document, no control, and no definition of general processes that would let anybody else reproduce it. It is what an organisation told him a year later. That is worth something and it is not a finding, and the difference between those two things is what the rest of this site is about.
7. A property business, and the data it already had
The situation. A medium-sized real estate business working out what AI meant for its customers and where to invest, at a point when the core of what it sold was not yet something AI could provide. The question was where value could be added around that core.
What was done. Work with senior leaders across the whole country on the data they already held: what sources existed, how they might be brought together, and where the output could genuinely be augmented rather than merely automated.
What happened. It changed how the leadership thought about their own data. Rahim's own note on this one is the honest part: several of the ideas had come up inside the business before and had never been implemented. The session did not invent them. It made them actionable.
What this does not show. Whether they were implemented afterwards. No follow-up was taken. An idea a business had already had, restated in a way it could act on, is a real contribution and it is not a result.
What these seven have in common
Six of the seven produced work after the session rather than at it. An all-hands became executive coaching. A survey became a workshop became four processes rebuilt. A set of words became a strategy and a retainer. That pattern is worth naming for anybody considering booking a keynote as a one-off: the talk is usually where the conversation starts.
And a caution about the whole page. Seven engagements chosen by the person who delivered them, described from memory, with no measurement in six cases and a self-reported figure in the seventh. That is what a case study page is, everywhere, including the ones that do not say so. If you want the material that has been tested properly, it is at the evidence base, where 298 sources are graded and each one states what it fails to prove.
For what a session actually involves, formats and what is needed on the day, take the speaker pack to whoever is running it.
