Skip to content
tessel advisory

Digital & AI

Copilots, models and the data underneath them, built with your team and deployed into the systems you already run: production is the deliverable, not the demo.

£160–240k, fixed · quoted before we start

Work that production.

A demo takes a fortnight. A system your staff rely on at nine on a Monday takes twelve weeks, because it needs evaluation thresholds, a rollback, an owner and a runbook. We build the second thing, in your environment, with your engineers in the room, and we spend the first eight weeks on the data, because that is where every model we have ever seen fail actually failed.

What it is not 

A vendor shortlist: we hold no reseller agreements, so there is nothing here for us to recommend.

A demo. A fortnight of prototype is not a system anyone can rely on.

A handover by documentation. Your engineers build it with us or it is not built.

Stage 03 of the method, 12 weeks of the 28.

01
02
03
04

The stages run in order, and each is bought on its own: the roadmap from stage 01 prices the rest, so nothing after it is bought blind.

Four parts, and what each is for.

Delivery week, Meridian
P.03.01

Data foundations

Ownership, definitions, quality gates and lineage, the eight weeks nobody puts in the press release.

P.03.02

Copilots & agents

Scoped to one job each, with the evidence for every suggestion shown to the person acting on it.

P.03.03

Model development

Chosen against your data, not against a leaderboard; thresholds agreed before launch.

P.03.04

Integration

Into the systems you already run, behind your access controls, with a rollback.

4 steps, in this order, every time.

01

The data, first

One definition per field, one owner, quality gates in the pipeline, lineage from source to decision.

02

The thin slice

One job, end to end, in your environment and behind your access controls.

03

The evaluation

Thresholds agreed before launch, measured after, published either way.

04Ends here

The rollout

Sequenced so the worst-served users go first, not the easiest ones.

What you are holding at the end.

You leave with 

A system in production, not a pilot

An evaluation harness with agreed thresholds

Runbooks your team wrote with us

−41% claims-processing time, one engagement

Buy this when 

01A pilot proved the idea and nobody can get it past IT review.

02Two reports disagree and nobody can say which is right.

03The board has asked for an AI plan and you have a shortlist of vendors instead.

If none of these is true yet, the audit is the cheaper place to start.

Where it has been bought, and what it came to.

Insurance 2025

Claims triage copilot at Meridian Mutual

−41%
claims-processing time

Read the case
Healthcare 2024

Radiology backlog triage at Pallas Health

−58%
radiology ticket backlog

Read the case

The full record 

64cases

Every engagement we are allowed to name, each with the figure agreed at the start.

All cases

The clients on the record.

Meridian Mutual
Cobalt Bank
Halvard Energy
Northline Logistics
Ostro Retail Group
Pallas Health
Fennimore Foods
Aldergate Council
Verano Media
Solent Water