Services
Your first agentic decision is too important for a self-serve trial. We build it with you.
A forward-deployed engineering team, not a sales engineer with a demo script. Two practices: Decision Verification, where we prove one real decision end to end; and AI Adoption, where the blocker isn't one decision but getting anything live at all.
Verification engagements
Four steps. Weeks, not quarters.
Pick one decision
01 · Scope
Not a platform rollout — one real, currently-manual, expensive-to-get-wrong decision your team already makes.
Seed the graph
02 · Instrument
Your domain experts encode the failure patterns they already know. This is configuration, not a model to train.
Run it in shadow
03 · Prove
The verification layer runs alongside the existing process until its confidence labels earn trust against real outcomes.
Convert to platform
04 · Handoff
Once proven on one decision, the same graph and gates extend to the next — on the platform, without us in the room.
Engagements · verification
Three sizes. Pick the one that matches how much you've already committed to.
Decision Diagnostic
$15K · 2 weeks
We scope one decision, map its failure patterns with your domain experts, and run the Calibration Benchmark against your own historical data.
Deliverable: a written constraint framework and a go/no-build recommendation. Credited toward a Build.
Single-Decision Build
$60K–$150K · 6–10 wks
Full build: causal graph, gates, confidence labels, audit log — shadow-run against real outcomes before it touches production.
Deliverable: one decision type live, verified, handed off to your team on the platform.
Multi-Decision Rollout
from $40K/mo
Ongoing FDE staffing to extend the graph to additional decision types across the org as each one proves out.
Deliverable: a growing portfolio of verified decision types, and a dedicated liaison.
Cloud co-sell
marketplace rate
Any of the above, purchased through AWS, GCP or Azure marketplace and routed through existing cloud spend commitments.
No new vendor relationship for procurement to open.
Platform SKUs are priced separately — Verification Core, Vertical Verification Packs, the Governance & Audit Layer and Enterprise are all on the Pricing page.
A typical first engagement · weeks 1–10
Week 1: we sit with your credit ops team and write down the six failure patterns they already recognise by feel — the ones nobody had encoded because nobody thought a rule tree needed writing down.
Week 3: those patterns are traversals in the graph, running in shadow against real applications, next to the analyst who already does this by hand.
Week 7: the first live decision — a same-day underwriting exception — clears CONFIRMED, and the risk committee sees a rendered audit trail instead of a Slack thread.
Week 10: we hand off. Your team owns the graph, and the next decision type is a platform license, not a new engagement.
Shape of the engagement
Ten weeks, and we leave.
The shadow run is the part that matters: the verification layer decides alongside the person who already does this by hand, and its labels earn trust against real outcomes before anything executes.
Weeks 1–3
Failure patterns your experts recognise by feel become traversals in the graph.
Weeks 3–7
Shadow run. Labels are scored against the analyst's own conclusions, not against a benchmark.
Weeks 7–10
First live decision, then handoff. The next decision type is a platform licence, not a new engagement.
The second practice
When the blocker isn't one decision, it's getting anything live at all.
Forty agent pilots and nothing in production isn't always a verification problem. Sometimes it's a legacy system nobody fully understands, a workflow that was never going to survive contact with compliance, or nobody in-house who has shipped an agentic system before. That's a different job — general AI adoption and modernisation — and we staff it with embedded FDEs.
Generic AI delivery firms ship the system and move on. Every adoption engagement we run also asks one more question: was this hard adoption problem actually a decision-verification problem in disguise? When it is, it becomes a candidate for a new Vertical Verification Pack. This practice doubles as the discovery engine for where the product line expands next.
Engagements · adoption
Scoped as a sprint first. We don't sell the multi-quarter version cold.
AI Adoption Sprint
$25K · 3 weeks
An FDE pair spends three weeks inside your stack — legacy systems, existing pilots, data estate — and returns a prioritised map of what's actually blocking production.
Deliverable: a scoped backlog of adoption blockers, ranked by effort and by whether they're verification problems in disguise.
Embedded FDE Pod
from $50K/mo
A standing embedded team of 2 to 4 FDEs building, changing and running whatever's on the backlog: legacy modernisation, a bespoke agentic workflow, an internal tool nobody else will own.
Deliverable: shipped systems, on a cadence, plus a running log of which turned out to be verification work.
Modernisation Program
custom
Multi-quarter, multi-team engagements — the SAP-migration, mainframe-rewrite scale of problem — usually co-sold through a systems integrator.
Scoped like a Sprint first.
The two practices are different jobs, sold to different buyers, staffed by FDEs from two different Academy tracks. Don't confuse “we'll verify this one decision” with “we'll rebuild your claims system,” because your buyer won't either.
Where the FDEs come from
Both practices are staffed out of Academy.
Decision Verification
Track A · 10 weeks
Causal reasoning, audit design, calibration, and a vertical elective. Staffs verification engagements.
See the Academy syllabus
AI Adoption & Modernisation
Track B · 10 weeks
Legacy-system reverse engineering, agentic workflow architecture, regulated-industry change management. Staffs adoption engagements.
See the Academy syllabus
Engagements
Bring the one decision you'd never let a self-serve trial near. Or bring the mess.
We start with a scoping call: what the decision is, who owns it today, and what wrong costs. For adoption work we start with a Sprint — what's blocking you, what it costs to fix, and which of it should really be a verification engagement instead.