# How to select and manage an AI consultancy as a mid-market business How to select and manage an AI consultancy as a mid-market business — a practical framework for senior leaders on evaluation, structuring engagements and avoiding common failure modes. Published: 2025-09-15 Author: Rodan Analytics Most mid-market businesses have already made at least one AI investment that did not deliver what was promised. A proof of concept that never reached production. A vendor relationship that consumed six months of internal resource and produced a dashboard nobody uses. A strategy document that sits in a shared drive. The mistake is not spending the money. The mistake is selecting a consultancy the same way you would procure a piece of software — evaluating on credentials, case studies and day rates, then handing over the brief and waiting. AI consulting engagements fail for predictable reasons: misaligned incentives, vague scope, no internal accountability and consultancies that are better at selling the work than delivering it. At your revenue level, you are also operating in an awkward market position. The largest firms will staff your engagement with juniors while billing at partner rates. Boutiques built for startups will struggle with your governance requirements, your data complexity and your stakeholder landscape. This article gives you a practical framework for selecting the right AI consultancy, structuring the engagement to protect your interests and managing the relationship to get outcomes rather than outputs. --- ## The market is not structured in your favour Understanding the supplier landscape is the first step. It divides, roughly, into four categories: **The global consultancies** — teams of hundreds, brand credibility, and an institutional habit of scoping work to maximise billable hours. At your revenue level, you are not their priority client. Expect to see the senior team at the pitch and the graduate team on delivery. **The pure-play AI vendors** — companies selling a platform or product dressed up as advisory. Their recommendations will, predictably, involve their technology. The conflict of interest is structural, not personal. **Strategy boutiques with an AI practice** — strong on frameworks, weaker on implementation. Excellent for a one-time AI readiness assessment. Less suited to building and deploying production systems. **Technical delivery consultancies** — engineering-led firms that can build, but may lack the commercial acumen to connect technical decisions to business outcomes. What the mid-market actually needs is a firm that spans strategic clarity and technical delivery, operates at appropriate scale, and has no incentive to inflate scope. That profile is rarer than the market suggests. A useful test: ask any prospective consultancy to describe a project that failed and what they changed as a result. Firms that cannot answer this have either not done enough work or have not reflected honestly on the work they have done. --- ## What to evaluate — and what not to Most procurement processes for consulting services over-index on the wrong signals. A polished deck is not evidence of delivery capability. A reference from a FTSE 50 client is not evidence of relevance to your situation. Evaluate on these five criteria instead: - **Domain fit, not generic AI capability.** A consultancy that has deployed demand forecasting models in retail distribution is more valuable to a retail business than one with broader AI credentials across financial services, healthcare and logistics. Depth of context matters. - **The ratio of strategic to technical resource.** Ask to see the proposed team composition. If the engagement is heavy on senior strategists and light on data engineers and ML practitioners, expect a strategy document. If you want deployed systems, the engineers need to be present from week one. - **Ownership of outcomes, not outputs.** Any consultancy can deliver a report. Ask directly: what does success look like at month three, month six and month twelve? If they struggle to answer in commercial terms — revenue, margin, cost, speed — treat that as a signal. - **Willingness to start small.** A consultancy that pushes back against a paid diagnostic and insists on a six-month engagement from day one is optimising for their revenue, not your risk management. A two-week diagnostic at modest cost should be available and encouraged. - **Data maturity assumptions.** Ask them what they need from you to begin. A consultancy that assumes clean, labelled, well-governed data has not worked in many mid-market environments. A consultancy that asks intelligent questions about your data infrastructure in the first meeting probably has. --- ## How to structure the engagement to protect your position Even with the right consultancy, poor engagement structure produces poor results. Start with a paid diagnostic. This serves two purposes: it gives you a real-world view of how they work before you commit significant budget, and it produces a prioritised assessment you own regardless of what you do next. Expect to pay £1,000–£2,000 for a rigorous diagnostic. If it is free, the discovery process is part of the sales process. Define the accountability structure before any statement of work is signed. You need a named internal owner — not a committee — with the authority to make decisions and the time to be genuinely involved. Engagements fail when the consultancy is treated as a fully outsourced function. They need access to data, people and decisions. Without an internal champion who can unblock those things, the engagement will stall. Insist on a phased commercial structure. A twelve-month engagement with a single fee does not create the right incentives. Structure the work in phases — typically a diagnostic, a build phase and a deployment and adoption phase — with commercial gates between each. This keeps both parties focused on delivery milestones rather than time spent. Define what production-ready means before the build starts. A common source of late-project conflict is disagreement about what "done" looks like. If the output is a model, what accuracy threshold constitutes success? If it is a reporting system, what does the acceptance criteria look like? Write this down before any code is written. --- ## Managing the ongoing relationship Selection and contracting are the beginning, not the end. Establish a monthly steering cadence with a standard agenda: progress against milestones, blockers requiring executive decision, risks and any scope changes. Keep it to forty-five minutes. The purpose is not a status update — it is to force the consultancy to articulate what they need from you and to hold both parties accountable to the plan. Watch for scope creep that benefits the consultancy more than you. Every engagement surfaces new opportunities. Some will be genuine. Others will be ways to extend the engagement without clear ROI. Evaluate each one the way you would any new project: what is the business case, what is the cost and what are we not doing if we do this? Consider the knowledge transfer question from month one. A consultancy that builds systems you cannot maintain or interrogate without their ongoing involvement has created dependency, not capability. Ask early: what will your team be able to do independently when this engagement ends? If the answer is vague, make it contractual. Take a mid-market ecommerce business as an example. They engaged a technical consultancy to build a customer lifetime value model. The model was delivered on time and performed well in testing. Eighteen months later it had not been updated, because no internal team member understood how it worked or how to retrain it as customer behaviour changed. The model became a liability. The problem was not the consultancy — it was the absence of a knowledge transfer plan. --- ## Before you make the decision The cost of a failed AI engagement at your scale is not just the fee. It is the internal resource consumed, the opportunity cost of the problem remaining unsolved and the organisational credibility lost when leadership AI initiatives produce nothing visible. The firms that get this right treat the selection process as seriously as any M&A due diligence. They interrogate the team, not just the pitch. They start small, build confidence and scale what works. They keep a named internal owner accountable throughout. The firms that get it wrong treat AI consulting like software procurement — evaluate, select, sign, then wait for delivery. If you are entering or re-entering the market for AI consulting support, the most useful first step is a structured diagnostic that tells you what your data and technology estate can realistically support, where the highest-value opportunities sit and what the right sequencing looks like. That is a two-week conversation, not a six-month commitment. Rodan runs exactly that diagnostic with mid-market and private equity-backed businesses. If you want an honest view of where to start, [book a diagnostic conversation with our team](https://rodan.io). --- **Meta description:** How to select and manage an AI consultancy as a mid-market business — a practical framework for senior leaders on evaluation, structuring engagements and avoiding common failure modes. HTML: https://rodan.io/insights/how-to-select-and-manage-an-ai-consultancy-as-a-mid-market-business