How to choose an AI consultancy: a buyer's checklist

You must choose an AI consultancy. Six firms have replied to your request, each showing comparable websites and case studies that include large figures. Your selection will decide whether your organisation gets a functional system or merely a report. This guide offers a way to tell these outcomes apart before you sign a contract.

AI consulting services include data and readiness assessments, choosing an initial use case, and creating the required data foundation. They also encompass building and integrating the AI, setting up governance frameworks, and providing post-launch support. Most firms do not offer all of these services, and they rarely disclose that fact. Apply the questions in this guide to find out which tasks a firm handles itself, which it outsources, and which tasks stay your responsibility.

Shipshape Data operates as a specialised AI consultancy. Our process builds a data foundation before creating AI applications. This guide targets people who are buying these services. Each question presented here applies to our firm as well as any other company you might interview.

The components of AI consulting services

The term includes six types of work. A proposal must state which of these types it covers.

• Assessment. It involves reviewing your existing data, the systems storing it, your team's skills, and the outcomes of any prior AI projects. This process lets a company decide which tools can be built right away and which need additional preparation.
• Strategy and use-case selection. It involves choosing the first problem to solve, assessing the value of a result, and setting the order for later projects. Our guide to AI implementation strategy describes how to handle that sequencing.
• Data foundation. It includes the pipelines, definitions, and access controls needed for a model to use data. This work is often left out of proposals and is the most common cause of project failure.
• Build and integration. It includes the model, the surrounding retrieval or automation systems, and the link to the tools users employ.
• Governance covers managing access permissions, verifying outputs, and logging activity. It also ensures the system complies with the regulations that apply to your data.
• Support after go-live. This includes monitoring, retraining, and updating content, with a designated person to call if any issues arise.

A company that does two of these tasks well and finds partners for the other four is a reliable choice. Be wary of any firm that says it handles all six tasks but cannot describe its process for building a data foundation.

The four kinds of firm you will meet

AI consulting firms fall into one of four categories. Their website branding varies, so you should identify them by their business practices instead.

Strategy firms supply senior staff to develop board-level business cases and roadmaps, but they seldom manage the implementation. This service is helpful when you need a board decision and already have a team to execute the work. The risk is that the plan may be impossible to cost accurately because the authors are not the ones who will build it.

Systems integrators offer large delivery teams and platform partnerships. They are built to manage long programmes across many departments. There is a risk that these firms choose a platform before they understand the specific problem. Their commercial incentives often favour expanding the team size rather than finishing the project. These firms suit organisations that have internal teams capable of directing their work.

Agencies deliver high-quality work on short timelines. These organisations focus on design and are effective at creating visible products or pilot programmes. They are often the best option for developing features that the public will use directly. However, there is a risk that the underlying data infrastructure will be insufficient. Support often ends after launch, which can stop a pilot from becoming a permanent production system.

Specialist consultancies are small teams that handle both data infrastructure and AI development. These projects are usually defined by fixed outcomes. The main risk with these firms is their limited capacity. Because they work with few clients at the same time, project timing becomes a critical factor. You should enquire about their availability to start rather than their overall ability to perform the work.

The question is not which kind of firm is best. It is which kind of firm your problem needs, and whether the firm in front of you is the kind it says it is.

Twelve questions to ask

Send these questions in writing to every firm on your shortlist and compare their responses. The exact wording of their answers matters less than whether the firm actually provides an answer.

Data and foundation

• Will you look at our data before you recommend anything? A strategy that lacks data analysis is merely an essay. The condition of your data determines what can be built, the sequence of the work, and the total cost.
• What condition does our data need to be in for the first use case, and who does that work? The answer shows whether the proposal includes the foundational work or leaves it to you.
• How do you decide what not to build? Ask for an example of a time the firm advised a client not to build. A firm that has never recommended against a build acts as a sales team rather than a consultancy.

Scope and pricing model

• Does the first phase have a fixed price for a specific outcome, or is it billed at a day rate? A day rate means you take on the financial risk if the scope of work is not clear.
• Does the first phase pay for the second? A successful plan begins with a measurable result that provides the funding for subsequent steps. If the project does not produce value until the second year, you should evaluate the purpose of the first year.
• What do you need from us, in people and hours, for the plan to hold? Every plan relies on your personnel. If a firm claims it requires no resources from you, it has not considered the requirements properly.

The delivery team

• Who will do the work, and can we meet them before we sign? The staff members who present the initial proposal are frequently different from those assigned to the actual project.
• Would you build what you are recommending, at that price and on that timeline? Observe the reaction of the people in the room. This question distinguishes a firm that delivers results from a firm that only provides presentations.
• How much of the work is a product or a partner platform, and how much is yours? Both answers are acceptable, but you should understand exactly what you are purchasing and which components can be substituted in the future.

Governance

• How will you manage our data, our permissions, and any regulated content? Our AI governance checklist specifies the controls that a complete answer must include.
• How will we measure the result, and what happens if it is not met? Ask for the specific measurement, the current baseline, and the date of the final reading.

Aftercare

• What happens after go-live? Who manages the system, who repairs errors, and for how long is that support included? These details distinguish an AI consultancy from a company that only provides demonstrations.

An AI implementation consultant who answers all twelve questions immediately has relevant experience. You should also consider a consultant who answers nine questions well and states that the other three fall outside their expertise. In that case, you must identify who will handle the remaining three tasks.

Red flags

You should end the conversation if you notice certain early signs.

• A proposal arrives before anyone has looked at your data.
• The roadmap delivers its value in year two.
• The price cannot be explained by the scope, in either direction.
• Every problem is answered with the same platform.
• The case studies have no numbers or no named clients.
• The proposed pilot lacks a plan for production and has no designated owner for the period following its launch.
• The strategy consists of a list of use cases that have no specific order or dependencies.
• The senior staff members present at the meeting will not work on the delivery team. The firm is unable to identify the specific individuals who will be assigned to the project.

A company that does not define what it will not do has not decided what its actual business is.

A selection process that lasts three weeks

A long procurement process does not result in a better decision. It only delays the start of the data work. Three weeks is sufficient time if the steps are determined beforehand.

Week one: the brief and the shortlist

Write a one-page brief: the problem, where the relevant data lives, how success will be measured, and any constraints on data, budget, or timing. Take the readiness assessment described below so the brief reflects what your data can support. Shortlist four to six firms across the four kinds, and send each of them the brief and the twelve questions.

Week two: the conversations

Schedule a 45-minute call with each firm. Speak with the person responsible for leading delivery instead of the sales lead. Request a written scope for an initial phase with a fixed outcome. Ask for two references you can contact by phone.

Week three: the comparison and the decision

Evaluate the written scopes using four criteria: the requirements for your team, the specific exclusions, the metrics for success, and the post-launch support. Contact the provided references to ask about past problems and the firm's methods for resolving them. Select a firm and sign a contract for the first phase only. The firm must complete the first phase successfully before you commit to the second phase.

The role of a readiness assessment

You should complete a readiness assessment before you brief any firm. This process determines if your initial discussion should focus on product development or the underlying data. It prevents you from paying consultants to identify issues you could find yourself in a few hours. Our assessment includes sixteen questions covering strategy, people, data, and change. It provides an immediate score. You should take the assessment, include your results in the brief, and ask each firm how those results would affect their proposed plan.

If the score is low, your first project should be smaller than you anticipated. You should establish the data foundation for a single use case and implement the AI only after that foundation is stable. If the score is high, you can begin the build phase immediately. The previous questions will help you determine which team members should handle the work.

Where this leaves you

You should choose an AI consulting firm using four criteria. First, the firm must examine your data prior to giving recommendations. Second, the initial phase should deliver a fixed outcome that supplies the funding for the next stage. Third, the people who present the proposal must also be the ones who carry out the work. Fourth, the firm must offer support after the system goes live. Finding a firm that satisfies all four requirements is difficult in the current market.

Our AI consultancy services include assessment, strategy, the data foundation, the build, and support after go-live, and our AI strategy work starts with your data instead of a deck. Talk to us and pose the twelve questions.

Frequently asked questions