Skip to main content

​
Audit AI activity this week

Friday afternoon. The week is mostly closed; the calendar is light; the rep has 10 minutes before the last call. This is the moment to look back at what the AI did across the book this week and calibrate trust: what did Katalyst get right, what did it miss, and what one or two tunes will make next week sharper. This play turns the AI Activity tab into that end-of-week ritual: scan accepts versus rejects by type, drill into the reasoning on a few entries that stand out, identify a pattern, and take a small calibration action before signing off.

​
What to expect

  • Timing. Roughly 10 minutes, once a week (Friday-afternoon or end-of-week ritual).
  • Prerequisite. The rep has had AI Suggestions, Drafts, signal generation, or account enrichment running for the week. They land on the AI Activity tab via the side nav (the Sparkles icon, labeled AI Recommendations Activity).
  • Outcome. A sharper sense of what the AI got right and what it missed this week; one or two tuning actions taken (a column instruction edit, a noisy signal type muted, an enrichment field reviewed) that compound into a calmer next week.

​
Step-by-step

  1. Open AI Activity. Click the Sparkles icon in the side nav. The page lands with the default filter: scope My Opportunities, action type All, provider All, date range the last 7 days. The Total Activities KPI card at the top shows the row count for the current window; the table below lists every AI-assisted CRM action chronologically with type icon, time, account or opportunity, action label, and status badge.
  2. Set the scope and date. The default last-7-days is the right window for the Friday ritual. If the rep also wants account-enrichment activity in view, flip the scope to All (combines My Opportunities and My Accounts). For a wider managerial pass, the All Opportunities and All Accounts scopes show every rep’s records in the org.
  3. Scan accepts versus rejects versus pending by type. Read the table top-to-bottom and let the Status column tell the story. The patterns the rep is looking for: a type firing a lot of Suggested rows that never got Accepted (the AE is letting them age out, why), a type firing Rejected at a higher rate than usual (the AI is reading the deal wrong, where), Email Drafts that landed but were never sent (the prompt is off, in what way). Filter the Action type dropdown to each of the four shapes one at a time (Field Updates, Email Drafts, Activities, Account Enrichment) for a clean per-type read.
  4. Drill into a specific entry. For a row that stands out (a rejected field update on a key deal, an enrichment with a value that looks wrong, a draft email that was never sent), click View in the Details column. The dialog opens with the row’s shape-specific renderer: for a field update, the proposed value plus the AI reasoning paragraph; for an enrichment, the per-field accordion with source URLs; for a draft, recipients, provider, thread ID, and reasoning. Read the reasoning. It usually names the AI’s hypothesis explicitly, which is what makes the calibration possible.
  5. Identify a pattern. Two or three drill-ins is usually enough to see one. The shape the rep is looking for: “Column X keeps proposing values that ignore Y,” or “The hiring signal type keeps firing on backfill roles that do not matter for my segment,” or “Account enrichment keeps overwriting the Industry field with a less-specific value.” Name the pattern in one sentence before moving to the next step.
  6. Take one calibration action. Three common moves. If the pattern is on a column the AI keeps getting wrong, open Opportunities or Accounts, find the column, edit the AI column instructions to name what the rep wants instead. If the pattern is a noisy signal type, run Tune your signal noise budget and mute the type or replace it with a custom one. If the pattern is on a draft prompt or an AI Suggestions setting, adjust the corresponding setting. One action is the discipline; two is fine, three is the rep avoiding the next call.
  7. Repeat for one or two more patterns, then close out. If time allows, take a second pass with a different action-type filter (e.g. Account Enrichment if the first pass was Field Updates) and run the same pattern-then-action loop. End by closing the tab. The next week’s AI activity will reflect the changes; the next Friday-afternoon ritual will read against a slightly cleaner baseline.

​
Variations

If the rep is a manager auditing a team rather than their own book, flip the scope to All Opportunities or All Accounts and use the Account / Opportunity column to spot per-rep patterns; today there is no per-rep filter, so the scan is visual. If the rep is doing a quarterly review rather than a weekly one, widen the date range to 90 days at the start and treat the play as a 30-minute deeper audit rather than a 10-minute Friday ritual; the same pattern-then-action loop applies, just with more drill-ins.
A dedicated screenshot of the AI Activity tab (ai-activity.png) is pending. When it lands, it should anchor step 1 (the page shell with the KPI card, the four filters, and the Recent Activity table) and step 4 (the Activity Details dialog showing one of the four shape-specific renderers).