Two claims are made about AI and go-to-market, and both are wrong. The first is that it changes everything, which does not survive contact with a business whose actual constraint is that nobody owns renewal. The second is that it changes nothing, which does not survive contact with a team whose research and drafting time has genuinely halved. What it changes is narrower and more specific than either.
AI is currently very good at compressing effort and poor at creating judgement. Every real use case sits on the first side of that line.
What is actually changing
Three shifts are real and observable in mid-market commercial functions today.
- Research and preparation collapse. Account research, call preparation, competitive positioning and first-draft proposals now take a fraction of the time. This is the largest genuine productivity gain available and it is unglamorous.
- Buyers arrive better informed and later. More of the evaluation happens before contact, which shifts value from early-stage education to late-stage differentiation. A sales motion built on informing the buyer is losing ground.
- Content economics invert. Volume is nearly free, which means volume no longer differentiates. Distribution and genuine point of view become the scarce inputs.
What has not changed is the part most claims imply: qualification judgement, negotiation, trust between people, and the discipline to keep a pipeline honest. Those remain human and remain the constraint in most businesses.
Where it moves the needle, and where it does not
| Area | Realistic effect today | Confidence |
|---|---|---|
| Seller preparation and research | Substantial time saving, redeployable to selling | High |
| Proposal and content drafting | First drafts in minutes; editing still required | High |
| Call summarisation and CRM hygiene | Better records, which improves every downstream number | High |
| Lead scoring and prioritisation | Real, but only where there is enough clean historic data | Moderate |
| Forecasting | Modest. The problem is usually data quality and behaviour, not modelling | Low to moderate |
| Replacing sellers | Not in complex B2B. Compressing their admin, yes | Low |
The pattern is consistent: AI performs where there is a repeatable task with abundant examples, and disappoints where the constraint is judgement, relationship or data that does not exist.
What it means for diligence
Two questions now belong in a commercial assessment that did not five years ago.
Is the AI claim real? The tell is whether it has changed a number. A business genuinely using AI in its commercial motion can point to a metric that moved: preparation time, output per seller, cycle time, CRM completeness. A business with a slide can point to a tool it has licensed. Ask what happens to the number if the licence is cancelled tomorrow.
Is the business exposed? Where a company’s value has rested on doing something that is now cheap, the proposition question moves from theoretical to urgent. This is the mechanism by which product-market fit erodes fastest at the moment: the problem gets solved adequately elsewhere, often inside a platform the customer already pays for.
Ask what proportion of the growth in the plan depends on an AI capability the business does not yet have. Where the number is material, it is an assumption rather than a plan, and it should be priced as one.
Capturing it post-deal
The failure mode is predictable: tools are bought, adoption is patchy, nothing is measured, and eighteen months later the licences are renewed out of inertia. Three things avoid it.
- Start where the task is repeatable and the volume is high. Preparation, drafting, summarisation. Unglamorous, measurable, and adopted willingly because it removes work people dislike.
- Instrument before deploying. If you cannot measure preparation time or output per seller beforehand, you will not be able to demonstrate the gain afterwards, and the initiative will be judged on anecdote.
- Do not let it substitute for fixing the engine. AI applied to a commercial function with subjective pipeline stages and no renewal owner produces faster, better-written versions of the same problems.
That last point is the one worth holding onto. The businesses getting most from AI commercially are not the ones with the most tools. They are the ones whose commercial engine was already legible enough that a productivity gain had somewhere to go.
Frequently asked questions
Does AI change how commercial due diligence should be done?
It adds two questions. Whether an AI claim has actually changed a number, and whether the business's proposition is exposed to something becoming cheap. The rest of the assessment is unchanged, because the underlying constraints are usually still judgement and process.
Where does AI genuinely improve sales productivity?
Research, call preparation, first-draft content and call summarisation. These are repeatable, high-volume tasks with abundant examples, which is where current systems perform well.
Can AI improve forecasting?
Modestly. Forecast unreliability in mid-market businesses is usually caused by subjective stage definitions and seller behaviour rather than by weak modelling, and better models do not fix bad inputs.
How can you tell a real AI capability from a slide?
Ask which number moved and what happens if the licence is cancelled tomorrow. A genuine capability has changed a measurable outcome; a licensed tool with no adoption has not.
Ready to aim higher?
If a growth plan leans on an AI capability, it is worth establishing whether that capability exists today or is an assumption.