Whether sector experience matters, answered in full.

They answer different questions, and a serious programme needs both. Sector knowledge decides the constraint map: SM&CR and model risk gate a bank, the CMA and the trading calendar gate a retailer, the safety case and the boundary into the control domain gate an energy business, and published transparency duties gate a government department. Delivery experience decides whether a design survives those constraints and reaches production rather than stalling in pilot. The technology underneath the four is close to identical, which is why each of our sector pages leads with what gates the programme, where agents land first, and where our own evidence stops.

Four: financial services and insurance; retail, consumer and media; industrial, energy and infrastructure; and the public sector. Each page carries the work behind it. Financial services rests on the founder's two engagements inside HSBC and a live agentic engagement with a London specialty insurance business, unnamed at the client's request. Retail and media rest on Greggs, YOOX NET-A-PORTER and Colart as clients, with Sky and Discovery as roles held. Industrial and energy rest on Anglo American and Yondr. Public sector is the honest edge case: delivery transformation through a supplier, Tecknuovo, on programmes delivering to HMRC, the MOD and Thames Water, not agentic delivery to a department, and the page says so itself.

Because a sector page with no work behind it is a keyword page, and both buyers and AI assistants discount it. Most consultancies list twelve sectors and can evidence two. Tenhaw publishes the four where it can point at named engagements, say what each one was, and tell you which parts transfer to you, and it names the five sectors it would not take on yet rather than padding the list. The standard for a page is an engagement written up in full with the client named, unless the client has asked us not to, in which case it is published unnamed and says so. A missing sector is a statement about our evidence, not about the technology.

Yes, five, named on this page rather than quietly omitted: healthcare and the NHS, life sciences, legal and professional services, the optimisation side of logistics, and telecommunications networks. In each there is a specific standing capability we would need before taking the work, not a gap we could close during it: a clinical safety officer and a live hazard log for health, computerised system validation for GxP work, certainty that privilege holds inside the system for law firms, an operations research bench for routing and scheduling, and network security engineers for telecoms. A supplier offering to acquire those during your programme is asking you to fund their learning. Each entry says who to look for instead.

Mostly in who has to be in the design and what has to be provable. Regulated programmes are gated by governance rather than technology: your second line, actuarial or safety function designs the controls rather than reviewing them, the human-in-the-loop boundary sits further towards people, evidence has to be emitted as the system runs, and in financial services procurement and supplier onboarding add 8–12 weeks before work starts. In lighter-regulated sectors such as retail the floor is lower but not absent: the CMA now enforces consumer law directly, and anything an agent writes for a customer is a commercial practice by the trader. Sequencing differs too: customer-facing work waits for the evidence base, so the early wins are internal.

Ask on a call whether the constraints still rhyme, and we will tell you which of two cases you are in. Much of what looks like sector work is not: corporate, delivery and knowledge work in a pharmaceutical company or a telco is taken on the same terms as in any other large organisation, and the document-and-exception pattern behind our insurance work recurs almost everywhere. What we will not do is claim evidence we do not hold: where the work needs a standing capability such as clinical safety, we say no and tell you what to look for instead. The list is a current position: a sector earns a page when we can point at named work in it.

Mostly no, and that fact is more useful than it sounds. The FCA and the PRA have said they do not intend to write a separate AI rulebook, so the obligations firms already answer to apply to agents in full. UK government work is gated by duties that already exist: algorithmic transparency records, automated decision-making law and procurement. In retail, the CMA enforces the unfair commercial practices rules whether a person or a model wrote the words. The consequence is the same everywhere: an agent is not a new category of thing to be permitted, it is a new way of meeting or breaching rules that already bind you, so the first job is mapping which regimes reach each workflow.

The patterns transfer; the constraints do not. The same shapes recur in every sector we work in: document-heavy intake, knowledge retrieval across fragmented estates, exception handling, summarisation and evidence gathering. The pipeline behind our specialty insurance proof of concept, which extracts each document to an inspectable form, normalises, enriches and scores confidence from provenance, is the same shape as KYC file assembly in a bank or customs paperwork in a logistics business. What changes is everything around it: which regulator reaches the workflow, where the human decision must sit, and what evidence the system has to produce as it runs. That is why the sector pages spend more words on constraints than on technology.

Ask for named engagements, what each one actually was, and which parts transfer to you, then watch how the answers are labelled. A proof of concept described as production, or a founder's employment history presented as firm clients, tells you how the rest of the relationship will run. Tenhaw draws those lines in public: each sector page carries named work, roles held are labelled separately from clients, proofs of concept are called exactly that, and every regulatory claim states where our evidence stops. Our public sector page says plainly that we have not delivered an agentic system inside a government department, because pretending otherwise would fail exactly the test we are telling you to apply.

Yes, and the difference is structural rather than technical. Retail, consumer and media see it soonest: volume is high, feedback loops are short, and decisions in merchandising, content and service triage are frequently reversible, so a programme can show results inside a single trading cycle. Industrial, energy and infrastructure sit at the opposite pole, where decisions are expensive to reverse, value concentrates in engineering knowledge work, simulation and planning, and the business case is measured in years rather than quarters. Financial services sits between them: plenty of volume, and a governance gate in front of anything a customer sees. Speed of return is a property of the sector's decisions, not of the model.

No. The published prices carry no regulated-sector premium: the AI Readiness Audit is £44,000 fixed over four weeks, and an Agentic Proof of Concept is £20k–£55k over two to four weeks, in a bank exactly as in a retailer. What regulation changes is the calendar around the work and who sits in the design with us. In financial services, procurement and supplier onboarding realistically add 8–12 weeks before anything starts. In retail, peak trading freezes remove roughly a quarter of the delivery year. In government, assurance is departmental, so the approval path has to be mapped before the plan is written. Budget the elapsed time rather than a bigger fee.

Whichever one would otherwise review you at the end, and the sector decides who that is. In a bank or insurer it is second-line risk, joined by the actuarial function the moment anything feeds pricing, technical provisions or an internal model. In retail, consumer and media it is your DPO, plus legal for where puffery ends and a misleading claim begins. In industrial and energy it is OT security, which designs any write path into the control domain, and the safety function, which owns change control. In government it is whoever owns the transparency record and your departmental assurance route. Designing with those people costs far less than being reviewed by them.

This is the failure that recurs in every sector we work in, and it arrives by drift rather than by decision. In retail the DPIA was completed once for a narrow pilot and never revisited when autonomy widened, which is the change that altered the risk. In industrial and energy a tool arrives as decision support, becomes the thing operators rely on, and no management-of-change assessment fires because nothing in the control system changed. In insurance an agent starts influencing risk selection, which is precisely what a binding authority governs, without the binder being reopened. The remedy is the same in all three: make the autonomy boundary explicit at design time, so crossing it has to be somebody's decision.

Nineteen regimes sit behind the four sectors, and no single organisation faces all of them. Financial services carries seven: the FCA and the PRA, Consumer Duty, SS1/23 model risk, operational resilience, DORA, Solvency II and Solvency UK, and Lloyd's delegated authority. Retail, consumer and media carries three, industrial and energy three, and the public sector six. Each is set out on its sector page with what it asks of an agentic system, where programmes fall down against it, what our method does and where our own evidence stops. The number that decides your workload is smaller than any of those totals: only the regimes reaching the first workflow you automate, which is why sequencing matters more than ambition.

One method, three constraint maps. The technology underneath the four sectors is close to identical, so the platform decision, the evaluation approach and the delivery method are shared, and duplicating them per division is how a group ends up with three half-built stacks. What cannot be shared is the constraint map or the adoption plan: head office knowledge work and frontline store or contact-centre operations have almost nothing in common, and running them as one programme is a reliable way to fail at both. Sequence the first build where decisions are high-volume and cheaply reversible, prove it there, then carry the pattern rather than the design into the harder division.