Key takeaways
- A pilot is one segment, one workflow and one named owner.
- Success criteria are written by the team before the first run.
- Start where a person approves every action that reaches the outside world.
- Loosen control step by step, never for the whole workflow at once.
What should an agentic GTM pilot cover
An agentic GTM pilot should cover one segment, one workflow and one owner, and nothing else. A segment might be, for illustration, finance directors at mid-sized logistics firms in two countries; the workflow might be research the account, draft a first email and save it as a draft.
Keep the segment small enough that the owner can read every output. The point of a pilot is to learn where the assistants are reliable, and you only learn that by looking at the work itself.
How do you choose the starting execution mode
Start in the mode that keeps a person between the assistant and the outside world. Sends, CRM writes and new lead searches should wait for approval until the owner has seen enough of them to trust the pattern.
In Pebbles Ai the modes are Manual, Plan and Auto. Manual asks for approval before actions that reach the outside world; Plan queues those actions instead of running them, so you can see the whole intended batch; Auto runs them, except template send and calendar delete, which always ask.
When should you loosen control over a step
Loosen control over a step only when it has run cleanly many times with few edits. Clean means the owner rarely changed an approval, drafts needed light edits at most, and no record was written to the wrong place.
Loosen one step at a time, not the whole workflow. Research and internal notes are usually first; outbound sends come last, and some teams keep them behind a person for good. Before any change, Plan is a useful check: it queues the actions the assistant intends instead of running them, so the owner can look over the whole batch.
What success criteria should the team define
Write criteria the team can check by reading the work, and agree them before the pilot starts. Criteria written afterwards tend to fit whatever happened.
- Quality: would the owner send the draft with light edits?
- Accuracy: do research facts hold up against their sources?
- Control: did any action run that should have waited?
- Time: does the owner spend less of their week on this workflow, in their own judgement?
- Response: how does the segment reply compared with how it usually responds?
How do you test AI agents in sales without risking key accounts
Keep the pilot away from your most valuable accounts and keep sends behind a person. Exclude current customers and open opportunities from the pilot segment, so a weak draft cannot land in a live deal.
Agree a stop rule before you start, such as pausing the pilot if a wrong message is sent or a record is overwritten. A clear stop rule makes it easier for sceptical colleagues to agree to the pilot at all.
How to pilot an agentic GTM motion, step by step
Plan the pilot in one sitting, then review it weekly until the end date.
- Pick one segment and one workflow, and name the owner who will read every output.
- Write the success criteria and the end date before the first run.
- Start in the most cautious mode, so every outside action waits for approval.
- Run the workflow on a small batch, and record every edit and rejected approval.
- Loosen the mode one step at a time, only for steps that have run cleanly.
- At the end date, decide to widen, change or stop, and write down why.
Where Pebbles Ai fits
- CapabilityAI workspaceExecution modes are labelled Manual, Plan and Auto, and the web app starts in Manual. Approval cards show each action that needs confirmation, and template send and calendar delete always ask, in every mode.
- CapabilityProspecting & customer recordsFor the pilot segment, name up to ten target companies with a role and a location; up to 100 leads a run come from Pebbles Ai's own growing B2B contact database, GDPR compliant, and land as contacts in their own segment of the built-in CRM (in Beta).
- FoundationAgent memoryFile the pilot's chats in one project, so Project memory keeps the facts you state about the segment for those chats only.
Common questions
How long should an agentic sales pilot run?
Long enough for the segment to reply and for the owner to see patterns in the edits, often a few weeks. Fix the end date at the start so the pilot cannot drift.
Who should own an agentic GTM pilot?
One person who does the workflow today and has time to read every output. A manager can sponsor it, but the owner should be close to the work.
Should a pilot run in Auto mode from the start?
No. Starting in Auto removes the review that tells you whether the assistants are reliable. Earn Auto step by step from the evidence of the pilot.
What comes after a successful pilot?
Widen one thing at a time: a second segment with the same workflow, or a second workflow in the same segment. Keep the same review habits while you do.
About this page. Written by the Pebbles Ai team for people who run sales, marketing and growth. Last updated .
