Choosing an AI consultancy: how to assess a provider

AI consulting covers engineering firms, strategy consultancies, implementation partners and resellers. The key distinction is how clearly each provider can demonstrate its experience, commercial interests and results.

Five criteria you can examine in conversation

Data comes before product recommendations

Where is the relevant data, what quality is available and who is allowed to process it? These questions determine which technical approaches are viable.

A provider recommending a product before this analysis lacks the basis for comparing requirements and alternatives properly.

Vendor neutrality is visible in the business model

Ask directly how the provider earns money from a recommendation. Partner programmes, reseller commissions and platform-specific implementation work are legitimate, but should be disclosed.

It is also useful to ask which tools the provider assessed and then rejected. Specific examples show whether alternatives were genuinely compared.

Engineering experience becomes visible in the details

Ask about interfaces, failure cases, monitoring and ongoing operation. Providers with implementation experience can state assumptions, weigh options and clearly identify unresolved technical questions.

Risks and limitations belong in the conversation from the start

A good provider raises error rates, privacy, operating costs and unsuitable use cases without being prompted. This makes it clear which assumptions need testing and where the initiative has limits.

The proposal defines a verifiable outcome

The proposal should state what will exist at the end, how its quality will be assessed and which responsibilities your team can take on afterwards.

Warning signs when selecting a provider

• Percentage productivity claims without a measurement method. Ask about the baseline, timeframe and comparison group. Without this information, the figure cannot be assessed or applied to your organisation.
• A strategy without a technical assessment. Claims about AI potential remain superficial if systems, data flows and existing software have not been examined.
• Early pressure to make a long-term commitment. A clearly bounded first engagement shows how the provider works and whether the relationship is effective before larger commitments are made.
• References without a clear account of the work. Client names may be confidential, but the provider should still be able to explain the mandate, constraints, their role and the project outcome.
• The same tool for different requirements. A sound recommendation follows from the data, systems and operating conditions. If the product is predetermined, there is little room for genuine evaluation.

Five questions for the first conversation

1. Which comparable initiatives have you delivered, and what difficulties arose?
2. How does your company earn money from the products or platforms you recommend?
3. Which conditions must we meet for the initiative to have a realistic chance of success?
4. What concrete outcome will the first engagement produce?
5. Under which conditions would you advise us against the initiative?

The fifth question is particularly revealing. A specific answer shows that the provider can identify limits and does not treat every situation as an opportunity to sell an engagement.

Conclusion

A credible provider makes its experience, commercial interests and working methods open to examination. Clear statements about data, risks and the first engagement’s outcome allow you to compare proposals on a sound basis.