Start with the right deployment
Open Close Forge and select the deployment you want to improve. Ask for an analysis of that deployment, not a workspace-wide guess. The analysis reads the workflow, permitted settings and recent run/message counts and returns recommendations bound to the selected configuration.
The activity counts describe records created during the reported period. They are not a conversion attribution report, a count of delivered messages, or proof that a recommendation will increase revenue. Ask Work to preserve those definitions in its explanation.
Request analysis before changes
Analyze this Close Forge deployment. Summarize the follow-up sequence, delays, message copy and recent run/message states. Prioritize improvements, but do not edit, publish, activate or enroll anyone.
Separate implementation-ready recommendations returned by the analysis from additional general writing or business advice. If you want a change that is not represented by a recommendation, state its scope explicitly and ask for a fresh review.
Apply the recommendations you actually selected
- Review the ordered recommendations and choose the specific ones you want.
- Keep the original conversation and deployment context. When following up, identify the recommendation instead of saying “do all of it” after unrelated discussion.
- Review the exact changes proposed for the workflow draft.
- Approve the scoped action and follow the resulting workflow link. Check the new draft version before continuing.
Understand the draft boundary
Supported draft edits include delays, SMS/email wording and supported AI review-mode settings. They preserve consent, routing, source connectors, triggers and stop conditions. If the source workflow is published, Work prepares the editable draft path without silently changing the live workflow and queued messages.
Editing a workflow draft is not publishing it. Publishing is not resuming a disabled deployment. Use a separate explicit request for each effect when it is actually intended.
Do not move recommendations to another deployment
A saved analysis belongs to its original deployment, workflow version, user, workspace and conversation context. Opening a different page does not retarget the recommendation. If the workflow changed, request a new analysis rather than forcing stale suggestions through.
Try: “Use the saved analysis for Estimate check-in in this conversation. Prepare only its second recommendation; if that analysis is stale, stop and tell me.” This gives the assistant a precise recovery path.
Judge the finished result
Review tone, factual accuracy, offer conditions and the customer’s next step in the draft. Then inspect the affected version and saved receipt. Do not infer a performance gain from successful saving. To measure outcomes later, compare consistent periods and clearly separate sends, replies, bookings and completed jobs.