Awards & Recognitions
AI consulting
AI consulting that finds what’s worth building, before you build it
We look at your workflows, data and systems, test the best candidates on your own data, and hand you a plan you can build with us or anyone else.
Volume
Hundreds a week
Data
PDFs attached to email
Owner
Accounts payable lead
Destination
ERP with an API
Risk
Wrong totals reach the ledger
Recommendation
Validate on a sample of your own invoices, with uncertain lines sent to your AP team.
Our advice comes from AI we’ve built and run
8h → 2h
building-plan review with computer vision
~1,000 / day
emails read by AI across 28 mailboxes
100+ divisions
on one AI order-email workflow
Since 2012
CMMI Level 3 · ISO 27001
01
When AI consulting pays off
Most teams don’t need more AI ideas. They need to know which one to start with.
Too many ideas, no priority
Every team has a use case and nobody can say which comes first.
A pilot that stalled
The demo worked, but it never reached the systems people actually use.
Unsure your data is ready
Records are spread across tools, and nobody knows if they’re good enough.
The board wants an AI plan
You need a credible, costed plan rather than a list of buzzwords.
Vendor pitches you can’t compare
Every tool claims to do everything. You need an independent view.
Worried about risk
Accuracy, privacy and who’s accountable when the AI is wrong.
02
Our AI consulting process, in five stages
Each stage ends with a decision that’s yours. You can stop after any stage. Each one leaves you with something you keep.
Stage 1 of 5
Discover
We meet the people who own the work and map where it arrives and where it ends up.
What we do
Interviews with process owners; a walk through the systems the work touches.
You receive
A long list of candidate workflows, with volumes and owners.
You decide
Which teams and systems are in scope.
Stage 2 of 5
Assess
Each candidate is checked against your data, your systems and the risk of getting it wrong.
What we do
Review data quality, access, volumes and the systems results must reach.
You receive
Readiness notes for every candidate workflow.
You decide
The criteria we’ll score against.
Stage 3 of 5
Prioritise
Candidates are scored on value and feasibility. Where rules or a plain integration beat AI, we say so.
What we do
Score, compare and shortlist, with the reasoning written down.
You receive
A prioritised opportunity map.
You decide
Which workflow to validate first.
Stage 4 of 5
Validate
The chosen workflow is tested on a sample of your own data before anyone commits to a build.
What we do
A small, contained test with the accuracy measured against an agreed target.
You receive
Measured results and a go or no-go recommendation.
You decide
Whether to proceed, adjust or stop.
Stage 5 of 5
Plan rollout
If it’s worth building, we plan how: integration, review rules, success measures and cost.
What we do
Design the rollout, the human-review step and the measures you’ll track.
You receive
A rollout plan with an estimate you can take to any builder.
You decide
Whether to build, and with whom.
03
How we decide what to build first
Every candidate is scored on business value and on how ready your data and systems are. Select a workflow to see why it landed where it did.
Pilot first
Plan for later
Quick win
Don’t build
Illustrative example. Real scores come from your assessment.
Pilot first
Supplier invoices
High volume, a clear structure and one system to post into. Most invoices can be read automatically, with low-confidence ones sent to a person.
Value 8/10
Feasibility 9/10
Pilot first
After-hours calls
Missed calls cost bookings, and a voice agent can answer from approved content and hand urgent calls to staff.
Value 7/10
Feasibility 7/10
Quick win
Ticket triage
Useful but modest: routing tickets saves time each day, and the helpdesk already has the data and an interface.
Value 4.5/10
Feasibility 8/10
Plan for later
Demand forecasting
Valuable, but the sales history is spread across systems and incomplete. Fix the data first, then revisit.
Value 9/10
Feasibility 3/10
Quick win
Proposal drafting
Easy to build and liked by the team, but it saves little time. Worth doing only alongside something bigger.
Value 3/10
Feasibility 8/10
Don’t build
Chatbot for everything
No single workflow, no owner and no measure of success. A rule-based FAQ page would do more for less.
Value 3/10
Feasibility 2/10
04
Is your team ready for an AI project?
Five questions. Your answers stay in this page.
Do you have a specific workflow in mind?
For example, invoices, order emails or inbound calls.
Does that work arrive in volume?
Daily or weekly, not a few times a year.
Is the data it needs available digitally?
Emails, PDFs, databases or call recordings.
Is there one person who owns the outcome?
Someone who can approve rules and accept results.
Can the destination system be connected?
An API, database or mailbox we can write to.
Your result
0 of 5 answered
Answer five quick questions
You’ll see whether you’re ready to pilot, need an assessment, or should start with discovery.
Keep going
5 more to see where you stand.
Ready to pilot
You have a clear workflow, the data and an owner. A short assessment can confirm it and move straight to validation on your data.
Book a pilot assessmentReady for an assessment
The foundations are partly there. An assessment will show which gaps matter and which workflow to start with.
Book an assessmentStart with discovery
The first step is finding the right workflow and checking the data behind it. That’s exactly what discovery is for.
Talk to a consultant05
Advice from a team that builds
What we’ve learned shipping AI shapes every recommendation we make.
Cook & Boardman Group
Building plans read by computer vision
Doors and windows are detected in PDF building plans, dimensions extracted, and estimators given a grid to confirm or correct each result.
What it taught us
Set the accuracy target with the people who use the output, then measure against it on real documents.
Measured results
8h → 2h
plan review time
70% → 95%
detection accuracy
Cook & Boardman Group
Order emails read, matched and posted to the ERP
Inbound emails and attachments are read, the PO, totals and vendor extracted, matched across divisions and posted into the ERP.
What it taught us
Start with templates and rules where they work, and let AI handle what they can’t.
Scale
100+ divisions
on one inbox workflow
US healthcare credentialing provider
A credentialing backlog cleared by AI
Each email is classified, its fields extracted and matched to the right ticket. Anything uncertain goes to a person.
What it taught us
Plan the human-review step from day one; it is what makes accuracy targets safe to raise.
Volume
~1,000 / day
emails across 28 mailboxes
VoAgents (Orion product) · Missional Software
Calls answered, qualified and routed
Voice agents answer and place calls, qualify leads and book into the CRM. For Missional Software, the receptionist answers only from approved content and routes urgent calls to staff.
What it taught us
Limit what an agent may say to approved content, and make the handoff to a person easy.
In service
24/7
calls answered
Human handoff
for urgent and complex calls
06
We’ll tell you when not to use AI
A recommendation is only useful if it’s honest. Here’s what to expect from ours.
01
Sometimes the answer is no
If a rule, an integration or a process change does the job, that’s what we’ll recommend.
02
Tested on your data
Recommendations are backed by results on a sample of your own work, not by a demo.
03
A plan you own
Scores, test results and the rollout plan are yours to build with anyone.
Models we work with: OpenAI · Claude
Hosting: Azure · AWS, North America
Delivery: CMMI Level 3 · ISO 27001
07
Questions buyers ask first
About scope, data and what happens after the plan.
Ask us something elseNo. The plan, scores and test results are yours, written so any team can build from them. Many clients do continue with us, because the same people who assessed the work can build it.
It depends on how many workflows and systems are in scope. We agree the scope, timeline and fee after a first call, before any work starts.
Time with the people who own each workflow, read access to sample data, and one person who can make decisions. We handle the rest.
Tests run on a sample you approve, in an environment agreed with your security team, on Azure or AWS in North America. NDA available on request.
We work with OpenAI and Claude models and choose per workflow on accuracy, cost and your data requirements. Sometimes the answer is no model at all.
Then we’ll tell you, and show what would work better: a rule, a plain integration or a process change. That’s a useful outcome too.
08
Book an AI assessment
Tell us what’s prompting the question. A consultant replies within one business day to agree scope, and will say plainly if you don’t need us.
Request the assessment
A consultant will reply within one business day.
NDA available on request.
