Two behaviours dominate the opening of most AI strategy conversations and both of them point away from the problem that actually needs solving. The client wants to know what their industry peers and competitors are doing. The vendor or consultant asks what use cases the client is interested in pursuing. The conversation proceeds fluently. Everyone sounds purposeful. And the specific, quantifiable friction point sitting in the room, the decision being made sub-optimally, the cost that results from it, the performance gap that an AI investment could actually close, goes unexamined and unaddressed. Too often, neither party asks about it explicitly. Neither party is structurally incentivised to.
This is not incompetence. It is avoidance, and it is structural.
Benchmarking is a way of not looking at yourself
The client who opens with “what are others doing” is asking a question that feels strategic but functions as a redirect. Benchmarking has legitimate uses: assessing maturity, validating strategic direction, or building organisational confidence that an investment is reasonable. But if the competitor’s investment becomes a reference point that shapes the conversation before any examination of whether that same area represents the highest-value friction in this specific organisation, it has stopped being strategy and become mimicry.
It usually is. Organisations have different process histories, different data environments, different cost structures, and different points of failure. The place where AI creates the most value in a competitor’s operation reflects their specific context. It tells you almost nothing about where AI would create the most value in yours. A competitor who has deployed AI in fraud detection may have done so because fraud detection was their most urgent problem, or because their Chief Risk Officer had the strongest internal voice, or because a vendor offered a compelling pilot, or because their previous fraud system was particularly weak. You do not know which of those is true. Benchmarking the outcome without understanding the context produces a strategy built on another organisation’s answer to a question you have not yet asked yourself.
The benchmarking instinct persists because self-examination is uncomfortable. Identifying where your own organisation is underperforming, quantifying the cost of that underperformance, and committing to a programme that will be measured against a specific improvement requires acknowledging a specific problem. Asking what competitors are doing requires nothing of the sort. It is a way of reaching for an AI strategy without the vulnerability that honest diagnosis requires.
Asking for use cases is abdication dressed as discovery
The vendor or consultant who asks “what use cases are you interested in” is performing a needs assessment while actually performing a sales technique. The question invites the client to name a problem from their own understanding of the solution space, and then provides the capability that matches what was named. It is efficient. It closes deals. It is also a systematic failure to do the work that would identify whether what the client named is actually their highest-value problem.
Experienced clients will name use cases that are familiar and defensible, not necessarily the most valuable. They name what they have heard about at conferences, what they have seen in vendor demonstrations, what their peers have mentioned, what their teams have proposed. The use case they name is shaped by everything except a systematic examination of where their decisions are underperforming and what that costs. The vendor who accepts the client’s use case nomination and builds a proposal around it has transferred the discovery responsibility to the person least equipped to discharge it, the client, while retaining the sale and releasing the obligation to find the real problem.
The use case question is also a signal. It tells the client that the vendor’s starting point is capability, not problem. Commercial pressure naturally favours shorter sales cycles and capability-led conversations. A vendor confident enough in their own method to say “before we discuss what we offer, let me understand where your decisions are underperforming and what that costs you” is making a harder commercial bet. They may not close the deal as quickly. They will close better deals, build more defensible programmes, and produce outcomes the client can measure.
The problem in the room
What is actually in the room in most of these conversations is a combination of performance data, operational experience, and domain knowledge that, properly interrogated, would identify the specific decisions where AI investment would produce the largest measurable return. The operations leader who has run the claims function for eight years knows where the delays are, what they cost, and why. The risk director who has managed the credit portfolio through multiple cycles knows which decision points produce the most expensive errors. The fraud team knows which transaction types escape their current detection.
These are not secrets. They are the specific texture of the organisation’s decision quality gaps, and the people who understand them are usually present. They are not asked about them because the conversation has been redirected toward what competitors are doing and which use cases are relevant. The questions that would surface the real problem, where are our decisions most economically inefficient, what constraint produces that inefficiency, and what would improvement be worth, are not being asked because they require both parties to commit to something specific and measurable, which is harder than a conversation about market trends and capability demonstrations.
The AI strategy that emerges from a benchmarking and use case discussion is almost always a diluted version of what is possible. It reflects the intersection of what competitors have publicised, what vendors are selling, and what clients feel safe naming. That intersection exists. It rarely corresponds to the highest-value opportunity available to this specific organisation in this specific moment. The highest-value opportunity is sitting in the room. Someone has to be willing to ask about it.
The organisations I have seen build AI programmes that deliver consistent, compounding value share one characteristic. At some point early in the process, someone asked the uncomfortable question. Not what are competitors doing. Not what use cases are you interested in. Where are your decisions underperforming, what does that cost, and what would it be worth to close the gap. The answer to that question is the AI strategy. Everything else is preamble.
