Your differentiation is prompts on top of someone else’s model. Anthropic ships a board-advisor agent next quarter and you are a rounding error. What is durable here?

The full executive answer

If our differentiation were prompts, the question would end me — so let me name what is not prompts. First, governed multi-tenancy: row-level-security isolation enforced in the database and tested in CI, append-only audit logs, injection-neutralisation and PII redaction on every untrusted input, provenance tracking, and two-model consensus — Anthropic and OpenAI cross-checking each other — on board-critical analyses. A frontier lab will ship a brilliant assistant; it will not ship a system whose architecture assumes the model must be governed, second-guessed and audited, because that posture is awkward for a model vendor and existential for a board.

Second, accumulated structure: 68 modules encoding board-level methodology, a signals registry, benchmarks, compliance obligations across 25 countries, an evaluation harness that regression-tests analytical quality in CI, and — most durably over time — decision memory: the longitudinal record of what your board decided, why, and what happened. Model capability is the commodity layer; the switching cost lives in the accumulated organisational memory and the governance wrapper. That is Helmer’s switching-cost power, built deliberately.

Third, counter-positioning against the labs themselves: we are multi-model by design, so "the model vendor ships it" partially helps us — better engines drop into our pipeline, and a customer wary of single-vendor AI dependence finds a two-provider consensus architecture more trustworthy, not less. The honest residual risk: a lab plus a big GRC vendor partnering could compress our window. Our answer to that is speed to the beachhead — being embedded in the first hundred boardrooms before the giants notice the category — which is exactly why the 90-day Enterprise targets are aggressive.