How to Tell a Real AI GTM Claim From a Sales Pitch

The state of AI in GTM right now

Two things are true at the same time. The underlying capability is real and improving quickly, and the marketing around it has run far ahead of what most teams actually get into production. That gap is where budgets get wasted and expectations get burned.

The important thing to understand is that in go-to-market, the model is rarely the bottleneck. The bottleneck is everything around it: whether your data is trustworthy, whether your systems are connected, and whether there is a clear workflow for the AI to slot into. A capable model pointed at messy data in a disconnected stack produces fast, confident, wrong answers. The same model inside a clean, well-designed GTM engineering setup quietly removes hours of work a week.

What's real: where AI already earns its keep

The pattern is consistent. AI earns its keep on work that comes at you constantly and does not hinge on a hard call, where a person still edits and owns whatever goes out.

Call prep is the clearest example. A rep opening an account cold usually burns twenty minutes reading history before a call, and most of that reading is reconstruction rather than thinking. Handing that to a model gets them a brief instead. The same logic covers meeting notes afterwards: a transcript is not useful, but the two or three lines the next person actually needs is.

Enrichment is less interesting and probably worth more. Routing, scoring, and targeting all read from fields that are frequently blank, and every blank one silently changes where a lead ends up. Filling those gaps automatically fixes a class of problem most teams do not know they have.

Then there is inbound. Reading a form fill, working out what the person actually wants, and getting it to the right owner is a task that takes a model seconds and a queue several hours. Drafting is similar, as long as everyone understands a draft is a starting point and not something that goes out unread. The edit is the entire point.

Last, and quietly the most valuable: watching data nobody has time to watch. Intent, churn risk, and expansion signals are usually sitting in a system already. They just never surface, because surfacing them was somebody's fourth priority.

None of this is glamorous, and that is exactly why it works. It is bounded, it is measurable, and a person stays accountable for the output.

What's hype: claims that don't survive contact with a real funnel

The claims that reliably disappoint share a trait. They promise to remove human judgment, or to skip the unglamorous foundation, and usually both.

Start with the autonomous SDR that books meetings while you sleep. It will draft and send at volume, that part is true. Left unattended it also personalizes badly, misreads intent, and burns the sender reputation that decides whether your good email reaches anyone. Volume was never the hard part of pipeline.

"AI replaces your sales team" is the same error one level up. It replaces tasks. It does not replace the relationship, the negotiation, or the judgment about which deal is actually real, and teams that believe otherwise tend to cut the wrong people. Close behind it sits set-and-forget AI that supposedly needs no review, which holds right up until the first confident mistake, at which point the team stops trusting the tool and quietly goes back to doing it by hand.

Two more worth naming. Anything sold as fixing your pipeline without fixing your data is selling you faster bad decisions, because no model turns bad inputs into good calls. And any pitch to replace your entire stack with one platform is skipping the integration and change-management work that actually determines whether people use the thing.

Why most AI GTM projects underdeliver

When an AI initiative in go-to-market disappoints, it is almost never because the model was not smart enough. The usual causes are mundane, and every one of them is fixable:

• The data underneath was full of duplicates, gaps, and definitions three teams disagreed about, so the model was reasoning over garbage from day one.
• Nobody had decided which specific step in the workflow the tool was supposed to own, so it never got inserted anywhere in particular.
• Somebody bought a subscription when what was needed was a design decision. Being AI-native is the second thing, not the first, which is the whole argument in what AI-native RevOps actually means.
• There was no review step, so the first confidently wrong output killed adoption before the thing had a chance.
• Success got measured as activity. Emails sent went up. Pipeline did not.

How to evaluate an AI GTM claim before you buy

You do not need to be technical to pressure-test a pitch. Ask five questions:

• On whose data was this demo built? Clean demo data is not your data. Ask to see it run on a messy, real example.
• What does it do when it is unsure? Good tools flag uncertainty and defer. Bad ones guess confidently.
• Where is the human review step? If the answer is ‘you do not need one,’ be careful.
• Does it need my systems connected first? If it does, that integration work is the real project, not the AI.
• Is the win high-volume and low-judgment? That is where the return is real. Anything promising to make the judgment call for you deserves scrutiny.

Then start with a single workflow, keep a person in the loop, and measure the outcome rather than the activity. That is the whole game: the teams that win with AI in GTM are not the ones with the best model, they are the ones with the cleanest data, the clearest process, and the most honest expectations. For where this fits inside operations specifically, read AI in RevOps: where it works and where it doesn't.

Keep reading: AI in RevOps: where it works and where it doesn't, realistic AI expectations for revenue teams, and how we run GTM operations with Claude.