Engagements

Three shapes this usually takes.

Anonymised and explicitly illustrative. We do not publish client names or invented metrics — real case studies appear here only with the client’s written permission.

Book a call Engagement models

Patterns

Problem, approach, stack, outcome.

The same four things we would walk through on a call about your process — so you can see the shape of an engagement before you commit to one.

illustrative Illustrative engagement · finance operations

Invoice intake for a 60-person logistics company

The pain

Nine thousand supplier invoices a year arrived as PDFs across two inboxes and a carrier portal. Four people keyed them in, and the close slipped every time someone was on leave.

Stack

Durable workflow engine
Document extraction model
Postgres
Accounting API
Review console

What we built

Layout-aware extraction with a per-field confidence score
Three-way match against purchase orders and delivery notes
Exceptions routed to a review console with the mismatch highlighted
Posting to the accounting system through its API, with an audit record per invoice

What changes

The illustrative target: the majority of invoices touchless, exceptions worked in one queue, close no longer dependent on one person being at their desk.

illustrative Illustrative engagement · compliance

Contract review queue for a mid-market insurer

The pain

Underwriters reviewed every policy document by hand against a 40-point checklist. Throughput was capped by headcount and two reviewers could reach different conclusions on the same file.

Stack

Retrieval over policy corpus
Rules engine
Reviewer console
Evaluation harness

What we built

Checklist encoded as structured rules with cited evidence per clause
First-pass review produced by the model, always shown with the source paragraph
Reviewer corrections captured and fed back into the evaluation set
Consistency scoring between reviewers to spot drift

What changes

The illustrative target: reviewers spend their time on judgement calls rather than clause-hunting, with a measurable consistency baseline that did not exist before.

illustrative Illustrative engagement · customer support

Support triage for a B2B SaaS platform

The pain

A ten-person support team handled 4,000 tickets a month. Classification was manual, priority was inconsistent, and product answers lived in three wikis and one engineer’s memory.

Stack

Ticketing API
Documentation retrieval
Drafting model with citations
Quality dashboards

What we built

Ticket classification and priority from the same taxonomy the team already used
Draft replies grounded only in current documentation, with links to the source
Deflection for the top repeat questions, measured against a holdout group
Weekly evaluation runs so a model or prompt change cannot silently degrade quality

What changes

The illustrative target: faster first response, fewer escalations to engineering, and a documented answer for every repeat question.

Warehouse operator moving pallets of goods with a forklift — Logistics: nine thousand invoices a year, four people keying them

Operations manager working through a stack of paper documents at his desk — Insurance: a 40-point checklist applied by hand to every file

Close-up of a team working through documents and laptops around a table — SaaS support: four thousand tickets a month, answers scattered across wikis

Honesty note

Why there are no logos here.

Automation projects touch a company’s operations, its data and sometimes its customers. Publishing a client’s name, or a percentage we cannot show the working for, is a quick way to lose the trust we are selling.

So this page stays illustrative until a client agrees otherwise. On a call we will walk you through the specifics of comparable work — architecture, failure modes, what we would do differently — and put you in touch with a reference when one is available.

How we work Security & data handling

Next step

Bring a comparable process to the call.

Describe the workflow and the volumes. We will tell you how we would approach it, where it usually breaks, and what a pilot would look like.

Book a 45-minute call +1 (650) 442-5844

One business day — and the reply comes from an engineer, not a sales sequence