Two install-base numbers landed the same week. Rogo: $160M Series D, 35,000 finance professionals, 250+ banks. Microsoft 365 Copilot: 20 million paid seats. Aidoc on the clinical-imaging side: $150M Series E, the same week. Both numbers grew this quarter. They are not telling the same story. If you ship AI in a regulated industry — fintech, healthcare, legal, defence — you need to read the contrast carefully before your next procurement cycle.
In horizontal markets, Copilot won — 20M paid seats and a feature roadmap that compounds. In regulated verticals, the picture is different. Vertical agents like Rogo (finance) and Aidoc (clinical imaging) are taking deals horizontal copilots can’t close, because audit trail, domain reasoning, data residency, and documented failure modes are weighted higher than feature breadth in regulated procurement. Most CTOs are still buying horizontal seats for workflows that need vertical agents. The four-question procurement filter below sorts that out in a 30-minute vendor call.
You are a CTO, COO, or Head of Operations in a regulated business. AI is on the operating-pack. You have a horizontal copilot rolled out across the company — email, docs, internal search — and the productivity story is real. Now a regulated workflow lands on your desk: a deal screen, a clinical letter, a credit memo, a litigation review, an export-control filing. The horizontal tool is in the room and your team would like to use it. Your CISO and your auditor are nervous. Procurement asks for the audit trail. The horizontal vendor sends a deck.
This is the moment the horizontal-vs-vertical decision actually matters. Up to this point, more seats was the right answer. From this point on, more seats is the wrong answer.
The thing horizontal copilots do brilliantly is reach. They sit on top of every productivity surface, they answer every kind of question, they integrate with the calendar, the doc, the email, the meeting. The thing they do badly is the question your auditor asks: what did the system decide, why, and on whose authority?
A horizontal copilot’s logs answer “what did the user ask?” The regulator’s question is several layers deeper: which data classes were retrieved, which tools were invoked, what was the confidence level on the output, who reviewed it, and where is the immutable record. A general-purpose tool that serves twenty million seats does not produce that record by default. A vertical agent built for a single regulated industry does, because it cannot ship to that industry without it.
The worst case I see in the field is variations of this: compliance asks for a six-month audit of every generated decision. The team produces a CSV of prompts. The regulator asks for the data lineage. The team has nothing. The relationship with the regulator is now harder than it was before AI arrived.
Three things that vertical agents do, and that horizontal copilots structurally don’t.
Domain-native reasoning. A vertical agent is trained on the actual primitives of the work — deal documents, prior credit memos, clinical letter formats, judgement opinions, supplier certificates. It produces the right shape of output without your team running a six-month prompt-engineering loop. The reduction in internal AI engineering effort is the second-order benefit nobody costs in the procurement case but everyone feels in the year that follows.
Audit trail by default. Every retrieval, every tool call, every decision is logged in a format compliance reads. Not “we have logs” — a single pane your auditor opens, filters by case, and exports. This is the layer Kleiner Perkins is pricing in the Rogo round; it is also why Microsoft shipped Agent 365 Runtime Protection on the horizontal side — because every prod AI org that didn’t buy vertical now has to bolt audit trail on after the fact.
Procurement defensibility. SOC 2, sector-specific certifications, a CISO who has answered the same questionnaire two hundred times, a DPIA template that lawyers recognise, a model card written for regulators. None of this is exciting. All of it shaves three to six months off the deal cycle. Vertical vendors charge for this; horizontal vendors don’t deliver it.
I run this on every vendor demo. Two of four answered cleanly is a productivity tool with a vertical decal on it. Four of four is a vertical agent worth procurement effort. Three of four is a conversation worth having about gaps and roadmap commitments.
1. Audit trail. Does the vendor log every retrieval, tool call and decision in a regulator-shaped audit trail? Not “we have logs.” A single pane your compliance team reads, with retention you can defend, exportable in a format your auditor accepts. The honest test: ask them to show you a twelve-month-old transaction reconstructed from the log, end-to-end.
2. Domain reasoning. Can the vendor pass five hard cases from your industry without your team prompt-engineering them? Bring real cases — redacted, but real — to the demo. Watch the vendor’s product handle them with no fine-tuning. If the answer is “send us your prompts and we’ll tune,” you’re a horizontal customer with a vertical wrapper, and the maintenance bill lands in month nine.
3. Data residency. Does the vendor’s data posture satisfy your regulator without a six-month data-room exercise? Region-pinned inference, contractual data-retention guarantees, and a clear position on training. The honest test: ask for the names of three customers in your jurisdiction and check the procurement timelines they actually achieved.
4. Failure mode. What happens when the agent is wrong? A confidence-scored output with a documented escalation path, or a “please consult a professional” disclaimer? Vertical agents own their failure surface. Horizontal tools push it back to you.
Pick by workflow, not by team. The same employee can use a horizontal copilot for email and a vertical agent for the deal screen, and that is the right shape. Decide on three branches.
Workflow under regulatory scope? If yes, the answer is a vertical agent or a self-built system with the same audit posture. If no, a horizontal copilot is fine and probably the right answer. If hybrid — a workflow that touches both regulated and unregulated data — segment the workflow, not the team. Run the regulated leg on the vertical agent, the rest on the horizontal copilot, with a clean handoff and the data residency posture intact.
The wrong shape is segmenting by team. Procurement wants one tool per department. The work doesn’t respect that boundary.
Finance — deal teams. Screening, CIM generation, buyer outreach, data-room diligence. All four steps are auditable in Rogo and similar vertical platforms; none of them are auditable in horizontal copilots in a way credit committee will sign for. The deals will move to vertical agents over the next eighteen months. Banks that haven’t evaluated by Q4 are running on borrowed time.
Healthcare — clinical operations. Imaging triage, prior authorisation packaging, clinical-letter drafting. The Aidoc Series E pattern shows where the dollars are: integrated workflows where the vertical agent reads the modality, produces the assessment, packages it for the radiologist, and files the audit trail clinically aligned. Horizontal tools cannot run inside a clinical workflow that an FDA submission has to defend.
Defence and export-controlled work. ITAR, EAR, NATO classifications, multi-tier data residency. No horizontal vendor will ship the contractual posture you need by year-end. The only options are vertical or self-built. The build path here is real and well-known to the audience reading this.
Real risk: vendor lock-in at the data layer. The vertical vendor learns your taxonomy, your case history, your operating playbook, and reasoning over that history is exactly the moat you’re paying for. Negotiate data portability up front. Insist on schema-level export, not just “the data is yours.”
Real risk: smaller vendor, less mature ops. A $160M Series D is real money but a five-year-old company is still a five-year-old company. Run the same business-continuity questions you’d run on any critical SaaS supplier. Get the SOC 2 Type II. Read the sub-processor list. Pressure-test the incident response plan. None of this is novel.
Overstated risk: cost. A vertical agent costs more per seat than a horizontal copilot. A vertical agent also costs less per closed deal, per processed claim, per approved authorisation. The unit-economics number is the only number that matters. Most vertical vendors will model it for you on demand. If they won’t, that itself is the answer.
Overstated risk: “we’ll wait for Microsoft to ship the regulated version.” Two years of waiting is two years of competitors compounding their workflow advantage on Rogo or its equivalent. The horizontal incumbents will eventually ship a regulated tier; they will not ship vertical-grade reasoning, because that is a different product built on different data with different specialists.
1. Inventory and tag. List every AI workflow in your stack. Tag each as regulator-scoped or productivity-only. Most companies discover the regulator-scoped list is twice as long as they thought.
2. Score the regulator-scoped list. For each, score the current vendor against the four-question filter. Anything below 3 of 4 starts a vertical-vendor evaluation this quarter, not next year.
3. Brief the CFO and CISO before procurement asks. The audit-trail gap is not yet visible on the operating pack. It will be the first time a regulator asks a question you cannot answer. Get ahead of it.
Twelve months in, you have a clean two-tier portfolio. Productivity workflows on the horizontal copilot — renewed at scale, instrumented, governed lightly. Regulated workflows on one or two vertical agents — deeply integrated, audit-trail in place, named owner, monthly metric. The two layers are explicit in the operating pack. The CFO understands which line is which. The CISO can answer the regulator’s question without a six-month archaeology project. Procurement has a written template for evaluating the next vertical vendor in twenty minutes, not twenty weeks.
The Rogo round isn’t a fundraising story. It is a procurement signal. Kleiner Perkins is pricing audit-trail moats, and the rest of the market will follow. The companies that read the signal early in 2026 give themselves a clean two-year run before their competitors catch up. The companies that don’t will spend that time bolting audit trail onto horizontal tools that were never built for it — and explaining to their regulator why they didn’t evaluate the alternatives.
Indica Tech’s two-week AI procurement audit inventories every AI workflow in your stack, tags each by regulatory scope, scores current vendors against the four-question filter, and produces a 90-day procurement roadmap your board can sign. Fixed price £3,500. Written report. Whether you hire us for the next step or not.
See the audit engagement →