Niche markets

Eight markets where an org chart beats a chatbot.

Energy, oil and gas, distribution, security, financial, healthcare, retail and marketing. What these have in common is operational depth and regulatory weight — the two conditions under which a general-purpose assistant stops being useful and a structured system starts paying for itself. We build into them with the same four-step method, and we say plainly which ones we have built and which we have not.

The method

Four things a vertical system needs before it ships

A system that has read your industry's documents is not a vertical system. It needs to be measured in your domain, constrained by your regulator, and shaped like your operation. Skip any of the four and what you have is a general assistant wearing your industry's vocabulary.

Eight niche markets against four requirements — domain corpus, evaluation set, regulatory envelope and reference org chart — with security marked built and the rest method-ready
Built means measured and published. Method-ready means the steps are proven, not that the vertical is delivered.
01

Domain corpus

The documents, schemas, procedures and history the system reasons over — your permits, your manifests, your policies, your tickets. Assembled, versioned and access-controlled, so an agent's answer can be traced to the source it came from rather than to a model's memory.

02

Evaluation set

Questions from your domain with known-good answers. Without it, "the strongest model for the task" is a vendor's claim. With it, model selection is a measurement you can repeat — and re-run every time a new model ships, which is roughly monthly.

03

Regulatory envelope

What your regulator, your auditor and your insurer require. This is not paperwork added at the end — it is the structure. A rule that says a licensed human must approve becomes an approval gate no agent can pass. A retention rule becomes a logging control.

04

Reference org chart

The unit, role, decision-right and escalation design for that industry's operation. It is the deliverable clients keep even if they never run another agent: a written account of who decides what, and what must never be decided automatically.

What the envelope adds

Same method, different constraints

MarketWhere the operational depth isWhat the envelope adds to the chartStatus
SecurityAlert triage, evidence handling, control testing, incident responseFramework scoring, evidence retention, a named accountable owner per controlbuilt & published
EnergyOutage and maintenance scheduling, permit and compliance filing, asset recordsSafety cases, separation between operational technology and business systemsmethod ready
Oil & gasField reporting, vendor and lease administration, HSE documentationIncident reporting duties, contractor accountability, records that survive auditmethod ready
DistributionOrder, inventory and exception handling across systems that do not talkTraceability, customs and origin documentation, carrier liability linesmethod ready
FinancialReconciliation, reporting cycles, client servicing, document reviewSupervision requirements, records retention, an auditable decision trailmethod ready
HealthcareIntake, prior authorisation, coding, documentation backlogProtected health information handling, minimum-necessary access, a clinician gate on clinical judgementmethod ready
RetailMerchandising, pricing, supplier and returns operationsConsumer protection rules, payment data separation, advertising substantiationmethod ready
MarketingResearch, production, publication and measurement at volumeClaim substantiation, disclosure duties, brand and legal approval gatesmethod ready

Status is stated per market on purpose. "Built and published" means a system exists in that vertical, is measured, and the measurement is public — for security, that is our own weekly scorecard. Everywhere else the method is proven and the vertical is not yet delivered, and we would rather write that down than let a logo grid imply otherwise.

The one we built

Security, measured on ourselves

We did not pick security as a vertical because it sells. We picked it because we had to solve it to run an agent fleet safely, and solving it produced all four requirements: a corpus of controls and evidence, a 25-control evaluation set, a regulatory envelope in the NIST AI Risk Management Framework, and a governance unit with eight specialists and named owners.

The result is measured weekly and published in full — including where we score zero. That scorecard is both our proof and the instrument we run on clients.

See the audit and our scorecard

Twenty-five NIST AI RMF controls across Govern, Map, Measure and Manage, coloured by our own measured score
The security vertical, built on ourselves and re-measured every week.

Your market

Tell us the operation and we will scope the four steps against it.

Leave your company and the process that costs you the most. You get a written reply naming what the corpus, the evaluation set and the envelope would have to cover — and whether a structural system is worth it for you at all.