Most mid-market companies in non-technology industries are, by now, on their second or third attempt at an AI program. The first program was usually the enthusiasm program: a pilot, a vendor, a board update, and a quiet retirement six months later. The second program was the rationalization program: a reset, a more sober vendor, a more careful set of use cases, and a slower retirement eighteen months later. The third program is, often, the one that begins to hold.
The pattern is not principally a technical one. The technology has, mostly, become accessible enough that the technical work is straightforward for a competent team. The pattern is organizational, and it is the same organizational pattern that has accompanied every previous wave of operational technology in the mid-market: from ERP, to e-commerce, to data warehousing, to advanced analytics. The technology arrives faster than the organization is equipped to absorb it, and the programs that succeed are the programs whose design accepts that constraint rather than pretending it does not exist.
The disciplined version
The disciplined version of mid-market AI adoption shares a small number of attributes. The use cases are narrow, specific, and tied to operational outcomes the executive team can articulate without reference to the technology. The deployments are sequenced — three or four use cases in production before any organization-wide initiative is contemplated, and each one with a clear operating owner who is responsible for the outcomes. The technical architecture is built for what the company actually requires, not for what the vendor's reference architecture proposes. The investment in change management is at least equal to the investment in the technology, and is conducted with the seriousness that operational technology investments warrant.
The disciplined version is also, importantly, conducted without theatre. The board is not subjected to a quarterly recitation of the program's transformative ambition. The all-hands meetings do not feature the vocabulary of revolution. The communications are sober, specific, and concentrated on what is actually being done and what is actually being learned. This restraint matters because the alternative — the program conducted with the rhetorical apparatus of transformation — sets expectations that the early use cases cannot meet, and the disappointment that follows the early use cases destroys the political capital required to sequence the work that will eventually produce the value.
What the vendors are selling, and what to buy
Most AI vendors selling into the mid-market are selling a version of the program that is well-designed for the vendor's revenue model and poorly-designed for the mid-market company's actual requirements. This is not because the vendors are dishonest. It is because the vendors' product, sales, and pricing models are calibrated to enterprise customers, and the mid-market customer is being sold the enterprise program at a discount. The mid-market customer who buys the enterprise program at a discount usually receives the enterprise program — with its scope, its complexity, and its assumptions about the customer's organizational maturity — at a cost that is too high for the value the program will produce in a mid-market environment.
The mid-market customer who buys well buys narrower than the vendor proposes, sequences more cautiously than the vendor recommends, and invests more in the organizational work than the vendor's services proposal contemplates. This is not the program the vendor wants to sell. It is, in our experience, the program that produces results in the mid-market.
What we recommend, in two sentences
Pick three use cases the executive team can describe in operational terms; deploy them in sequence; staff each one with an operating owner who is accountable for the outcomes; resist every invitation to conduct the program at the rhetorical scale the vendor would prefer. Then pick three more.