How to choose an AI consulting firm without buying a deck
Every ranking guide on this question is published by a firm that sells AI consulting, so the criteria below are credited to the page that prints them. We add a scorecard, a four-week shortlist and a reference call script. Digiton is one of the firms you could shortlist, and the last section says where it does not fit.
How to choose an AI consulting firm
- Shortlist length
- Three firms, four weeks
- Reference age
- Project ended 6+ months ago
- Outcome check
- Results 90 or more days after launch
- Scorecard
- Eight rows, same for every firm
- Checked
- 12 November 2026
The short answer
Choose an AI consulting firm by evidence, not by pitch. Name the one system or decision you need, ask each firm for proof it has shipped something similar and still runs, meet the people who will do the work, settle who owns the code and data, and call a reference whose project ended months ago. Score every firm on the same sheet.
What the ranking guides tell you to check
Geeks Ltd, a UK software and AI firm, published its guide on 15 May 2026. It tells buyers to match the firm to their stage: strategy-only firms produce documents, implementation-only firms build what they are asked for, and a partner that does both suits organisations ready to act. It also tells buyers to ask for outcomes measured 90 or more days after launch and to speak to a client whose project finished at least six months earlier.
Helium42 prints a 12-point evaluation guide. The points that matter most to a small or mid-size buyer:
- A documented methodology that starts with problem definition.
- Two to three case studies in your sector.
- A named delivery team, with confirmation that the people you interview will do the work.
- Clear written ownership of models, methodology and data.
- Data protection discussed before the engagement starts.
- A post-project support plan covering handoff.
Both publishers sell what they describe, so treat the lists as a starting point and test each item yourself.
The scorecard: eight rows to fill in for every firm
Print one copy per firm and fill it in from what they show you, not from what they say. A blank cell is information.
| Row | Evidence to ask for | Red flag |
|---|---|---|
| Fit | One shipped system close to yours, still running | Only slideware, or case studies with no named outcome |
| People | Names, titles and weekly time of whoever does the work | Vague staffing, or the pitch team vanishes after signing |
| Method | A diagnostic first, a prototype before the full build, success metrics agreed before delivery | A named technology recommended in the first meeting |
| Outcome | A measured result 90 or more days after launch | Success defined as deliverables handed over |
| Ownership | Code, prompts, data and evals in your accounts from the first commit | Reluctance to say in writing that you own them |
| Data | Where your data goes, which model providers see it, who is the processor | No answer on subprocessors or retention |
| Handover | Runbook, recorded walkthrough, one change shipped by your own person | A build only the firm can maintain, sold with a retainer |
| Candour | A project that went wrong and what changed afterwards | Cannot name a single failure, agrees with everything |
Geeks lists most of the red flags above, including recommending a technology before discovery and structuring the build so you depend on the firm to maintain it.
A four-week shortlist
Four weeks is enough to go from a long list to a signed scope. Longer usually means nobody owns the decision.
- Week one, write the brief. One system or decision, who uses it, which data it needs, and what a good result looks like in numbers. Use the AI project scope of work checklist for the sections. If the phases of an engagement are new to you, what an AI consulting firm does lays them out.
- Week two, build a long list of six and cut to three. Cut anyone who cannot show a shipped system like yours. Read each firm's own pages and a vetting guide before the first call.
- Week three, hold one call per firm using the same questions, taken from questions to ask an AI consulting firm. Ask for the scorecard evidence in writing afterwards.
- Week four, call one reference per firm, compare the scorecards, and ask the top two for a scoped proposal with named people, a first milestone and an exit point.
What to ask in a reference call
Ask for a client whose project finished at least six months ago, as Geeks suggests, and ask for one that did not go smoothly. Helium42 recommends references who can discuss difficulties. Questions that produce real answers:
- Who actually did the work, and was it the people who sold it?
- What is running today, and what has broken since go-live?
- Could your own team change it without calling the firm?
- What did you own at the end, and where does it live?
- What would you ask for in the contract that you did not?
- Would you hire them again for a different system, and why?
Listen for specifics. A reference who can only say the team was great has told you nothing you can score.
Where the contract decides the outcome
Ownership. Helium42 calls IP ownership non-negotiable and asks for models, methodology and data to be stated in writing. Ask that code, prompts, evals and configuration live in your repository and accounts from day one.
Data. Helium42 calls data governance non-negotiable. GDPR Article 28(3) requires a contract binding the processor. Ask which model providers touch your data, what is retained, and who signs the processor terms. An AI vendor due diligence checklist covers the rest.
Exit. Agree a first milestone and a point where you can stop without penalty. A firm confident in its work accepts that.
Check your own readiness and the kind of firm you need
Geeks Ltd lists four things the buyer needs: a specific problem, accessible data you own, an internal champion and willingness to change workflows. If one is missing, a firm will either sell you a discovery to find it or build something nobody uses. Fix the gap first. It is cheaper than finding it in week six.
The same guide says a good partner asks about your goals before proposing solutions, challenges your assumptions, runs a structured diagnostic, builds a prototype before the full build and agrees success metrics before delivery. If a firm agrees with everything in the first meeting, that is on its red flag list.
The type of firm matters more than the brand. Our guide to the kinds of AI consulting firm sets out the main types with an example of each. A short version: a strategy house suits board-level direction, a big consultancy suits programmes across a large organisation, and a product-engineering firm suits one system that has to reach production.
If the real question is firm or person, read AI consultant versus AI agency. If you are weighing a hire, read in-house AI team versus outside help.
How long each phase should take
Helium42 publishes typical timelines. They give you a yardstick for a proposal that looks too fast or too slow.
- Discovery takes one to two weeks.
- Education and planning take two to four weeks.
- Implementation takes four to eight weeks.
- Handoff takes two to four weeks.
- The total is roughly ten to twenty weeks, and quick-win pilots take six to eight.
A proposal with no handoff phase has left out the part that decides whether your team can run the system. A proposal that spends half the timeline on discovery for a single well-understood workflow is padding.
Where Digiton fits and where it does not
Digiton is an AI consulting firm that builds AI agents and custom AI software and takes systems built with Claude, Cursor or Lovable into production. Brandon Da Costa has built automation systems since 2020 and AI systems in production since 2023. The entry step is a 25-minute scoping call.
- For a new agent or workflow, see AI agent development.
- For a larger system, see custom AI development.
- For a build that exists and is not holding up, see production engineering, where we read the real repository and give you a scorecard across ten layers with evidence for each.
Digiton is a poor fit for a board strategy deck, for regulated advice, for a very small budget where a tool subscription solves the problem, and for buyers who want a firm with a local office. Run Digiton through the same scorecard as everyone else.
Frequently asked questions
How many AI consulting firms should I shortlist?
Three. Start from six, cut anyone who cannot show a shipped system like yours, and score the three on the same sheet. More than three calls rarely adds information and slows the decision.
What is the biggest red flag when choosing an AI consulting firm?
A named technology recommended in the first meeting, before discovery. Geeks Ltd lists it first, and it usually means the firm sells a product or a preferred stack.
Should I ask for references?
Yes, and ask for a client whose project ended at least six months ago. Ask what is still running and whether their own team can change it without the firm.
Is Digiton right for every buyer?
No. It is a poor fit for strategy decks, regulated advice and very small budgets. Score it on the same sheet as every other firm.
Sources, checked 2026-11-12
4 sources
- Geeks Ltd published on 15 May 2026 a guide advising buyers to ask for outcomes measured 90 or more days after launch, to speak to a client whose project finished at least six months earlier, and listing red flags such as recommending a technology before discovery.Source: geeks.ltd, checked 2026-11-12.
- Helium42 publishes a 12-point guide to choosing an AI consultant that includes a documented methodology, a named delivery team, written IP ownership, data governance and a post-project support plan.Source: helium42.com, checked 2026-11-12.
- GDPR Article 28(1) says the controller uses only processors providing sufficient guarantees for appropriate technical and organisational measures, and 28(3) says processing by a processor is governed by a contract or other legal act binding the processor.Source: eur-lex.europa.eu, checked 2026-11-12.
- Helium42 names ISO/IEC 42001:2023 as the international AI management system standard and says it is relevant when a sector requires one.Source: helium42.com, checked 2026-11-12.
Digiton Dynamics OÜ is registered in Tallinn, Estonia, and delivers from Lisbon.
Update log
- Published. Read on 12 November 2026: the Geeks Ltd and Helium42 selection guides.
Each source above was opened on the date beside it. How facts get chosen and checked is written down in the editorial policy.
Talk to an engineer about your system
Book a free 25-minute scoping call. We tell you what we would build, what it takes to keep it running, and whether we are the right fit.