More of the customer journey now happens where the seller is not present. Before purchase, buyers ask AI to research, compare, and plan. After purchase, that same AI increasingly connects the product, operates it, and gets the customer unstuck.

Both sales-led and product-led motions are losing surface area to this silent part of the journey. The company does not own the machine doing the work. It often cannot see the interaction. And no single leader owns whether the machine's picture of the company matches what the company actually is.

The diagnostic

Five questions the customer's AI keeps asking

These are not stages in a funnel. The machine can consult any of them before or after money moves, and a weak answer at one point contaminates the others.

  1. 01

    Can it find you at all?

    Unprompted, not only when the buyer already knows your name, and with current facts.

  2. 02

    Can it compare you honestly?

    Can it reach your price, limits, security posture, and fit criteria without inventing the missing pieces?

  3. 03

    Can it work out how you connect?

    If it planned the integration this afternoon, would the plan match what your product can actually do?

  4. 04

    Can it run the work?

    Not just describe the product, but complete the customer's core job and verify the result.

  5. 05

    Can it recover when something breaks?

    Can it diagnose the failure, apply the right fix, and confirm that the system recovered?

The job completion flow

Five places the job can stop

Every exit is a loss the seller may never see recorded. The two in the middle are especially expensive: the customer buys a smaller version of what you offer, or pays and then stalls.

The silent losses

The funnel cannot report what never entered it.

Omission

A capability is real but poorly described. The buyer's AI compares you against a smaller version of yourself, and the company never knows the capability was missing from the decision.

Disillusion

The buyer arrives with an implementation plan assembled from public material. The plan is wrong, so activation stalls and the failure becomes evidence for the next buyer.

The vacancy

Five answers, five teams, no owner.

How a company is represented to machines is split across marketing, product, pricing, partnerships, documentation, and support. Each team may be doing its part well, while nobody owns whether the answers add up across the full customer journey.

This is a commercial job because the failure appears as lost consideration, lower conversion, slow activation, higher cost to serve, and quiet churn. The owner needs the authority to coordinate the whole system and the discipline to measure it end to end.

Evidence and forecast

What I have built, and what still needs proving

Built and measured

Discovery

At Lokalise, I built the AI-discovery operating model and measured category leadership across approximately 1,000 high-value prompts.

Operating evidence

The human version

Across Amazon, Taboola, and Rapyd, I worked on the same commercial questions for people using interfaces: compare, connect, run, and recover without expensive help at every step.

To be tested

The machine version

The mechanism is familiar. The machine acting on the buyer's behalf is new. That version needs controlled tests, captured failures, and commercial measurement, not certainty by assertion.