How does it think?
Agents that cite their sources, with a bill you can read line by line.
Agents, retrieval, voice and document AI over tenant-scoped data, with evaluations in CI, cost ceilings per run, and a human step before anything commits the business.
What it is
We build the thinking layer: retrieval over your own documents scoped to the tenant that owns them, agents that do one named job with one named owner, voice and document pipelines, and the routing that decides which model answers at what cost. Every answer carries its sources, every agent has an evaluation set that runs on each change, and anything that commits price, scope or a date stops for a person. Tenant isolation is proved by tests that ship with the system, not asserted in a policy document.
Who it is for
If none of these sound like your situation, this is probably the wrong arc — and we would rather tell you that than sell you a discovery.
- Twelve years of quotes, contracts and specifications sit in a shared drive nobody can search.
- Your team answers the same forty questions every week and the answers live in one person's head.
- You piloted a chatbot, it invented a price, and the pilot ended there.
- Half your inbound arrives as a phone call after six in the evening.
How the work is shaped
Repeatable shapes, not bespoke proposals. Each one has been run before and has a duration we hold to.
- 01Retrieval Pilot · 3 weeksOne document corpus indexed and tenant-scoped, with a golden question set, measured answer accuracy, and a citation trail on every response.
- 02Agent Build · 8 weeksOne agent doing one job end to end: tools, guardrails, escalation path, an evaluation suite in CI, and a cost ceiling per run.
- 03Agent Layer · 14–18 weeksSeveral agents under one registry — shared retrieval, routing by task and cost, human approval queues, and per-agent spend reporting.
What sits inside it
Agents
One job, one owner, one evaluation set — and a human step on anything that commits price, scope or a date.
RAG
Retrieval scoped to the tenant that owns the document, with the isolation proved by tests you keep.
Voice
Calls that book, confirm and escalate, with the transcript written back to the record it belongs to.
Document AI
Intake, extraction and routing for the paperwork that currently moves by hand.
Model routing
The cheap model answers the easy question; the routing rules and per-task ceilings are yours to change.
What it has produced
Every number here comes from work on this page. Follow it to the story and check it.
- fewer stock-outs across 400 SKUs
- 31%fewer stock-outs across 400 SKUsSee the story
- returned to clinic reception staff
- 11 hrs/weekreturned to clinic reception staffSee the story
- median time to clear a freight exception, from 38
- 4 minmedian time to clear a freight exception, from 38See the story
- cross-tenant retrieval leaks in the isolation suite
- 0cross-tenant retrieval leaks in the isolation suite
The work behind it
Three days of stock guesswork, gone
A two-floor clothing retailer replaced a Monday morning reorder meeting with a forecast read from its own till data.
31% target reduction in stock-outs across 400 SKUs
The waiting room emptied before the doctor arrived
Four clinics moved patient intake to the phone in the waiting room and got eleven hours a week back at reception.
11 hrs/week target reception time returned across four clinics
Thirty-eight minutes to clear an exception, down to four
A freight operator's exception desk stopped being four people reading four screens.
4 min target median time to clear an exception, against 38 observed
Proposals that stopped eating the fee-earners' evenings
A consultancy assembled proposals from its own past work instead of from last week's near-miss document.
6-9 hrs → 90 min design target for a first proposal draft
Weekend enquiries answered before Monday
A lettings agency stopped losing the enquiries that arrived when nobody was at a desk.
8 min target median first reply to a weekend enquiry
Is Intelligence what you need?
Tell us what is slow and what it is costing. If the answer is a different arc, or no arc at all, we will say so before anyone writes a proposal.
Scope an agent