Outside-in before AI: the sequencing mistake

Most teams I watch adopt AI in GTM the same way: buy the seat, connect the data, turn up the volume, then ask why the pipeline looks the same. The diagnosis that follows is usually "we need a better tool", "the data is bad" or "the copy is generic". Those complaints are real enough, and they are usually downstream of a skipped stage. The tool is rarely the problem; the order is.

Here is the sequence I now use before any AI touches a go-to-market motion, and what broke each time I skipped a step.

Market → Motion → Proof → Signals → AI

Outside-in means you start from the buyer's world and work inwards to your tooling. Inside-out starts from what the software can do and goes looking for a use. That second path feels faster, and it is the expensive route.

What I learned (transferable, not tool-specific)

1. Market first: if you cannot write the ICP in one sentence, AI will invent as many as you leave room for.

Enrichment and generation fill whatever space you leave. A vague market definition does not stay vague; it gets scaled into lists, copy and dashboards that all look plausible and disagree with each other. The fix is boring: one sentence naming who, what job they are trying to get done, and what changes for them if they do.

2. Motion second: decide how you win before you decide how fast.

Outbound, inbound, partner, product-led and expansion each need different data, different hand-offs and different definitions of a good week. AI tuned for one motion quietly damages another, for example volume outbound landing on accounts an account manager is already working. Pick the primary motion and write down what it is not allowed to touch.

3. Proof third: no evidence, no scale.

Proof is the smallest believable reason a buyer should listen: a result, a pattern you can show, a before-and-after you can defend. Personalisation without proof is just a nicer way to say nothing. If you cannot point to the proof a message rests on, that message is not ready for volume.

4. Signals fourth: timing is evidence, not a vendor score.

A signal earns a place only when you can say why it predicts a conversation for your market and motion. A job change, a funding round or a pricing-page visit can mean something or nothing depending on steps 1 to 3. Track which signals preceded your own wins before you trust anyone else's ranking of them. Whether third-party intent scores hold up across segments is unknown for most teams until they check against their own outcomes.

5. AI last: it is a multiplier, so check what it is multiplying.

By the time you reach AI, the job should already be narrow and safe: draft, research, route and summarise inside gates that exist. Put AI first and it becomes the strategy by default, because whatever it produces fastest is what the team ends up doing. So AI works only inside gates that have a named owner and a pass/stop rule.

What I would do on Monday

Write five lines, one per stage, and put an owner and a pass/stop rule under each.

• Market: the ICP in one sentence. Stop if you cannot write it.
• Motion: the primary motion and what it must not touch. Stop if two motions share the same accounts with no rule.
• Proof: the proof each message rests on. Stop: no message goes to a segment without a named proof point.
• Signals: the two or three signals you can justify from your own wins. Stop if the only justification is someone else's score.
• AI: the one job you will let AI do inside those lines. Stop if any line above it is blank.

Before the next purchase, name the earliest blank stage. That blank is the broken layer, and it is this week's work rather than a new seat.

What this is not

Not a tool list, not a stage diagram you need to buy, and not an argument against AI. The argument is about order.