Ask a revenue team what they are doing with AI and you will usually get one of two answers. Either everyone has ChatGPT open in a tab, or there is a chatbot on the website that deflects some support questions.
Both are real. Neither is what the money is for.
The adoption number is not the interesting number
MIT's Project NANDA looked at this directly in 2025, across more than 300 publicly disclosed AI initiatives, interviews with 52 organizations, and surveys of 153 senior leaders. Its finding on general-purpose assistants is worth sitting with:
Tools like ChatGPT and Copilot are widely adopted. Over 80 percent of organizations have explored or piloted them, and nearly 40 percent report deployment. But these tools primarily enhance individual productivity, not P&L performance.
Eighty percent adoption. Almost no measurable effect on the business.
That is not a criticism of the tools. Individual productivity is a real thing and a rep who drafts follow-ups faster is genuinely faster. It is a statement about where the leverage is not. If your AI strategy is that everyone has access to a good assistant, you have bought a better pen for each person and left the process they are writing into exactly as it was.
What actually sits between an inquiry and a booked meeting
Walk the path an inbound inquiry takes through a revenue operation and count the steps where a human is doing something a human does not need to do.
Someone notices it arrived. Someone works out whether it is worth pursuing. Someone finds the company, reads a bit about them, decides who should own it. Someone writes the first reply. Someone remembers to chase it on Thursday. Someone updates the record so the pipeline report is not fiction by Friday. Someone reconciles the calendar invite with the CRM with the notes.
Not one of those steps is the chatbot's job. They are the steps an appointment-led operation recognises immediately, and they are the same in a software business — only with longer names and more systems between them. Every one of them is a step where the work stops moving unless a specific person picks it up. And every one of them is where an inquiry dies quietly, because nothing in that list generates an alert when it does not happen.
This is the difference between AI as a feature and AI as plumbing. The feature is something a person opens. The plumbing is something that runs whether or not anyone remembers it exists.
The distinction matters more than it sounds, because it changes what failure looks like. When a feature is not used, nothing visibly breaks — the tool sits there, the licence renews, and the only evidence is a usage report nobody reads. When plumbing fails, an inquiry sits unassigned and somebody notices by Thursday. One of those failure modes gets fixed. The other funds itself indefinitely.
Why the plumbing version is harder to buy
There is an obvious reason the chatbot wins the budget: you can see it.
A conversational interface demos beautifully. It has a face. You can put it on a slide, show it to a board, and everyone immediately understands what it is. A routing rule that assigns an inquiry to a named owner within ninety seconds and attaches a task with a due time demos like a spreadsheet. It is the more valuable thing and the worse presentation.
MIT's report describes the consequence in the same section: enterprise-grade tools stall because of "brittle workflows, lack of contextual learning, and misalignment with day-to-day operations." The systems that fail are not failing at conversation. They are failing to fit the shape of the work.
That gap — between a good demo and a system that survives an actual Tuesday — is where most of the spend in this category disappears. It is the same failure described in more detail in most automation fails because nobody designed the workflow first: the tool is asked to run a process that was never written down, meets the exceptions, and quietly stops being used.
There is also a budgeting consequence. Because the visible thing is easier to justify, the plumbing work tends to get funded as a follow-on phase that never arrives. The demo passes the board; the routing rule waits for next year.
What to ask instead
The useful question is not where could we use AI. Asked that way, the answer is always the most visible surface, which is usually customer-facing chat.
The question worth asking is narrower and duller: which step in this process only works because a specific person remembers to do it?
That question finds different things. It finds the handoff between the form and the CRM that someone does by copy-paste. It finds the follow-up that happens when the rep is not travelling. It finds the weekly report that is rebuilt by hand every Monday and is therefore always describing last week. None of those are chat problems. All of them are places where the work depends on a person's memory, and memory is the least reliable component in any operation.
Fix those and the numbers move, because the numbers were never limited by how fast anyone could type. This is the same argument the firm makes about designing systems rather than adopting tools, applied to the one process where the cost of getting it wrong is easiest to count.
The revenue team that has genuinely changed its numbers with AI rarely has an impressive thing to show you. It has a process that used to require six people to remember six things, and now requires none of them.
Nobody asks for a demo of that.
An AI Systems Review looks at your revenue and operations.
Where work depends on someone remembering, what it costs, and what would change first.
Book an AI Systems Review →