Most middle market AI projects are aimed at the wrong layer
The pattern is consistent across the engagements we see. A middle market operator commits to AI, hires a vendor or builds an internal team, and within six months has deployed something that demos well, generates a press release, and produces no measurable change in unit economics.
The reason is that almost every initial AI deployment targets the visible layer of work when the actual leverage sits in the invisible layer underneath. The visible layer is where the work shows up. The invisible layer is where the cost lives.
Below are the patterns that determine whether an AI deployment produces returns.
Pattern #1: Visible Work Gets the Attention
Customer-facing chatbots, sales email assistants, and marketing copy generators are visible. They demo well. They produce stories. They also represent perhaps 15 percent of operating cost in a typical middle market business, and the gains from automating them are usually offset by integration overhead and human review cycles.
Pattern #2: Invisible Work Is Not Decomposed
Order classification, exception handling, vendor reconciliation, freight claim processing, AP coding. None of this is glamorous. All of it is structured, repetitive, and rule-governed. In a typical $200 million distribution business, this work represents 40 percent or more of operating cost, and almost none of it is being measured at the task level.
Pattern #3: Tasks Are Not Inventoried
The diagnostic question is what does the back office actually do, broken down to the task level, with time per task and error rate per task. Most middle market businesses cannot answer this question. The first finding in any operational AI engagement is the inventory itself, and the inventory is usually surprising even to the people who manage the function.
Pattern #4: Automatable Work Is Not Separated from Judgment Work
In most back-office functions, 30 to 50 percent of task time is consumed by work that is fully rule-governed and fully automatable today, with current technology, at modest cost. The remaining work requires judgment and should remain with people. The two have not been separated, which is why people are spending most of their time on work that does not need them.
The Diagnostic Question
What are the ten most time-consuming tasks in the back office, who performs them, how long do they take, and what percentage of each task involves a judgment call versus a rule that could be applied by a machine.
If the business cannot answer this with specificity, the AI investment cannot be sized. If it can, the opportunity is usually concrete and immediate.
The Implication
The mistake is treating AI as a strategic initiative requiring a roadmap, a center of excellence, and a transformation narrative. In the middle market, it is a back-office automation problem with better tools than were available three years ago.
The opportunity is not abstract. It is a list of specific tasks, with specific volumes, performed by specific people, that can be moved off human time within 90 days. The work that remains is the judgment work, which is what those people should have been doing in the first place. Treat AI accordingly and the returns appear within a quarter. Treat it as transformation and you will spend two years building a capability that produces decks.
- AI DEPLOYMENT
The integration tax will eat your AI business case unless you price it correctly
Every AI vendor pitch in the middle market understates the same line item. The model works. Then the project hits integration, and timeline and cost double.
3 MIN READ