Why AI Stalls in Enterprise Sales and Marketing, and How to Fix It

Walk into any large sales or marketing organisation and you will find AI everywhere: licences for a chat assistant, a copilot inside the CRM, a pilot for AI-written emails, a vendor demo of agents booked for next quarter. Ask what it has changed in pipeline, win rate or cost of acquisition, and the room goes quiet.

That silence has data behind it. MIT's NANDA initiative reported in 2025 that around 95 percent of the generative AI pilots it studied showed no measurable effect on profit and loss. The figure has fair critics, since it used a strict definition of impact and a short window, but it matches what many leaders say in private: plenty of activity, little that shows up in the numbers.

I have seen the other side at a smaller scale: a B2B sales team of around thirty people where an AI layer read and scored every inbound lead before anyone picked up the phone, inside a CRM I built for them. It worked not because the model was special, but because it changed one workflow end to end, with an owner and a number. That lesson is the one that scales.

Adoption is not productivity

Most AI programmes measure adoption: seats activated, prompts sent, people trained. Those numbers climb easily, because using AI is pleasant and mostly harmless. Productivity is something else. It means a workflow now produces more, better or cheaper output, and someone can prove it. The gap between the two is where most enterprise AI budgets disappear.

Ethan Mollick has written about the secret cyborgs: employees who quietly use AI to do their work faster and tell nobody, because they fear more work, or for their jobs. The gains are real but private, so the organisation never captures them. A programme that only rolls tools out from the top misses where the learning is actually happening.

The jagged frontier

In a 2023 field experiment with 758 BCG consultants, researchers from Harvard, Wharton and BCG found that on tasks inside AI's capabilities, consultants using GPT-4 completed 12.2 percent more tasks, about 25 percent faster, with work rated 40 percent higher in quality. On a task just outside that frontier, those using AI were 19 percentage points less likely to reach the right answer than those working without it (Dell'Acqua et al., Harvard Business School).

For sales and marketing this may be the most practical idea in the research. A first draft of an account brief sits inside the frontier. Judging from a few emails whether a large customer is about to churn may not. Rolling AI out evenly across every task guarantees both gains and expensive mistakes. Mapping the frontier task by task is the real work.

The 70 percent nobody budgets for

BCG's research on where the value in AI comes from found that the companies getting results put people and processes ahead of technology and algorithms, and recommends a 10-20-70 split: roughly 10 percent of the effort on algorithms, 20 percent on technology and data, and 70 percent on people and processes. Most enterprise budgets invert it. They buy the 10 and the 20 and hope the 70 happens by itself.

In sales and marketing the 70 percent is concrete. Who owns the workflow. How quotas and incentives treat AI-assisted work. What a rep may send without review. What the manager inspects. Which meeting changes because the report now writes itself.

One workflow at a time

The pattern I trust is narrow and deep. Choose one workflow that touches revenue: inbound lead handling, account research before a meeting, campaign reporting, proposal drafting. Give it one owner. Write down how it works today, with numbers. Then rebuild it: AI takes the steps inside its frontier, people keep the judgement, and every decision the system makes is logged so someone can check it.

Where AI earns its place

Agents, carefully

The next wave of vendor pitches is agents that act on their own. Gartner predicts that more than 40 percent of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls. That is not a reason to wait. It is a reason to give agents narrow jobs with clear permissions: read and draft freely, act on customers or money only with approval, and log everything.

Measuring what matters

When leaders cannot quantify AI's effect, it is usually because nobody wrote down the baseline. For each rebuilt workflow keep three numbers before and after: time per unit of work, quality or conversion at the end of the workflow, and the cost of running it. Review them monthly with the workflow owner. Scale what moves. Stop what does not, without embarrassment.

Bring the secret cyborgs into the open

The fastest learning in most companies is already happening at desks, unofficially. Make it safe to share: a monthly session where people show a workflow they improved with AI, an internal library of prompts and agents that work, and a clear policy on what data can go where. When I ran AI adoption training inside a sales and marketing team, the most useful hour was always the one where colleagues showed each other what they had quietly built.

AI works in sales and marketing. What stalls is the organisation around it. Fix one workflow, with an owner and a number, and let the results make the case for the next one.