Amigdala Partners
EN

Language

EnglishDeutschDEItalianoIT
Contact
AI-generated image, labelled AI Generated: at a wooden table, a woman with grey hair in a navy jacket points at one of three printed proposals laid side by side, while a younger colleague in a light blue shirt listens.AI
AMIGDALA / INSIGHTS
← All Insights
INSIGHTS / Perspective

How to choose an AI consultancy

Explore the perspective
THE PERSPECTIVE

Choose on the work to be done, not on the name of the firm. First establish which activity you want to improve, who is accountable for it and which data it depends on. Then ask each candidate who will actually work on the project, what you will see working and when, how the result will be measured and what your team will own at the end.

← InsightsBefore you look for a firm01 / 07
In this article +

How should a company choose an AI consultancy?

Before you look for a firm

Many selections start from a request that is too broad: "we want to use AI". With a request like that, every proposal looks plausible and none can be compared with the others. Three things are worth settling internally first.

The work. Which activity you want to improve, who does it today, how long it takes and where it stalls. A specific activity can be measured. An ambition cannot.

The owner. Who in your organisation is accountable for the result, and who will use the solution every day. If these people are not involved from the start, the project stops at the demo.

The data. Which information the activity depends on, where it sits and who can authorise its use. Scattered data and unclear permissions slow a project down more than any limit of the technology.

Three kinds of partner

Firms offering AI consulting fall into three groups. None is better in absolute terms: it depends on what you need.

Large consultancies and system integrators. They have scale, a presence in many countries and proven processes for long programmes. They are the natural choice when the project involves thousands of users, legacy systems to migrate or a service that must be supported around the clock. The trade-off: longer timelines, and teams in which the senior people you met during the sale are not always the ones doing the work.

Specialist boutiques. Small, senior teams that hold strategy, design, development and training together. They make sense when you need to reach something that works quickly: a first use case in production, a new digital product, a change to be adopted by a defined group of people. The limit is scale: a boutique does not run a multi-country programme with continuous support on its own. Some work with a partner network to extend their capacity.

Product vendors and vertical agencies. They sell a ready-made solution for a specific problem, for example a customer service assistant. They are a good fit if your problem matches their product. If the work you want to improve is specific to your organisation, you will end up adapting the process to the tool.

Often the right answer is a combination: a boutique to decide, build the first use case and train people, and an integrator to extend it across the organisation.

Eight questions to ask during selection

1. Who will work on the project? Ask for names and how much time they will commit. Ask also who will be in the room when the difficult decisions are taken. Finally, ask to speak to a client who has done a similar project.

2. What will we see working, and when? A good answer names a first usable version within weeks, on a narrow scope. A twelve-month roadmap with nothing to try along the way is a risk.

3. How do we measure the result? Criteria are agreed before work starts, from a measure of today's situation. The count must include the time people spend checking what the AI produces.

4. What happens to our data? Who has access, where it is processed, which actions the system can take on its own and which require a person's approval.

5. Are you tied to one technology? A partnership with a model provider is evidence of competence. It becomes a problem if every answer leads to the same product.

6. How are our people trained? Classroom training after the project has finished achieves little. Ask whether it happens during development, on the real tasks of each role.

7. What do we keep at the end? Documentation, access and the skills to run and change the solution without calling the supplier back for every change.

8. When do we stop? A serious partner sets out in advance the conditions under which a project is extended, corrected or closed. If this question causes difficulty, that is already an answer.

If you operate in a regulated sector, add a ninth: how legal obligations are handled, starting with the EU AI Act, and who in the supplier's team is responsible for them.

Warning signs

The proposal talks about technology and not about your process.

You are shown a demo, but nobody asks about your data and permissions.

The promised result is a document, not something people can use.

There is no planned point at which to decide whether to continue.

Benefit figures arrive before the starting situation has been measured.

How to start without over-committing

A small engagement tells you more about a firm than any presentation. Choose a single use case, agree how you will measure it and ask for it to be taken into production within a few weeks. You will see how the team works, how it handles your data and how far it involves your people. The decision on the next step is then based on facts.

Where we stand

This guide is written by an AI consultancy, and it is right to say so. Amigdala Partners is a Swiss AI-native boutique, but it does not stop at advice: we design, build and run the solutions. A platform built inside the firm passed 100,000 users in its first twelve months. Another has managed the content of a multinational's customer innovation centre since 2015, without interruption.

A senior partner works on every project, from the first workshop to handover. Around the team is a network of partners, specialists and technology providers, which lets us extend the team as the project grows. Work usually starts with a four-to-six-week sprint on a single use case, taken through to production. We are an OpenAI Select Partner, but we are not tied to a single technology.

Where we do not go alone: large-scale migration of legacy systems and round-the-clock support across several countries. That calls for a large integrator.

Further reading

Data & AI: how we build a use case together with its controls ↗

Culture: training by role and adoption in everyday work ↗

Before you trust an AI pilot, ask how much time the expert had ↗

What will it cost to check the agent's work? ↗

Who can change your AI system when the business changes? ↗

Talk to us about your use case ↗

FURTHER PERSPECTIVES

Keep exploring.

AI-generated image, labelled AI Generated: in the stock room of a spare parts supplier, a warehouse lead hands a small sealed parcel to an older courier driver standing by a hand truck, with tall shelves of boxes and an open loading door behind them.AI
Perspective

Before an AI assistant answers your buyers, decide which answers you can keep

When an AI assistant answers a buyer, the answer becomes a promise someone else in the company has to keep. Before launch, sort the questions buyers really ask by whether you can stand behind the answer, and give every firm answer an owner.

Nov 18, 2026 · Amigdala Team
Read more
AI-generated image, labelled AI Generated: before a meeting, a man in a light blue shirt places a printed page on a wooden table next to a printed dashboard, while a woman in a navy jacket, glasses in hand, stops to read it.AI
Perspective

When an AI number improves, check what changed underneath it

An AI workflow runs on services that suppliers update on their own schedule, sometimes including how use is counted. We propose a company change log covering your changes and theirs, and fixed test requests rerun after every change, before anyone reads the trend.

Amigdala Team
Read more
AMIGDALA / CHALLENGE EXPLORER