Skip to main content

​
AI Activity

AI Activity is the page that proves the AI has your back. Every suggestion accepted, every draft sent, every signal acted on, every field updated, every task created, every column run, recorded as a durable row with the proposal, the reasoning, the timestamps, and the resolution all in one place. Scroll back to verify what the AI did and didn’t do, on which records, on which days, and use that record to calibrate your trust over time.

​
What you can do here

  • Audit what the AI has touched. The page is the wide-angle lens on every AI-assisted CRM action Katalyst has taken across your scope: field updates written to Salesforce, emails drafted into Gmail or Outlook, tasks created from meeting and email recognition, and per-account enrichments sourced from external research. Each row carries the type, the entity, the action taken, the resolution, and a View button into the full detail.
  • Scope the view. A scope dropdown above the table flips between My Opportunities, All Opportunities, My Accounts, All Accounts, and All. The default is My Opportunities; flip to the All scopes for a team-wide audit, or stay on My to keep the page focused on your own book.
  • Filter by action type, provider, and date. Action type narrows to Field Updates, Email Drafts, Activities, or Account Enrichment. Provider scopes Email Drafts to Gmail, Outlook, or both. Date range defaults to the last seven days; widen the start date for a quarterly or year-to-date pass.
  • Read the table chronologically. The Time column shows the absolute datetime and the relative phrase (“2 days ago”) on two lines. The Account / Opportunity column carries the company favicon, the name, and an Opp or Account tag chip so you can tell at a glance which entity the action touched. The Type column shows the Salesforce logo for CRM mutations and the Gmail or Outlook logo for emails.
  • Open the per-row dialog for the reasoning. Clicking View opens a per-action dialog with the proposal in full. For a field update: opportunity, field name, proposed value, AI reasoning paragraph, created and accepted timestamps, status. For an email draft: opportunity, recipients, provider, thread ID, AI reasoning, status. For an activity: type, subject, description, status grid, AI reasoning. For an account enrichment: per-field accordion with the field name, proposed value, source URL chips, and a per-field accepted, pending, or rejected pill.
  • Read the status as a final state. Status badges reflect the terminal call: Accepted, Rejected, Action Taken, Suggested, Drafted, Superseded, or the activity’s lifecycle status (Logged, Scheduled, In Progress, Completed). The page is read-only by design; pending suggestions are still actionable on the Bell and on the Actions queue, but once they resolve, they land here permanently.
  • Reconstruct a timeline for a deal review. Filter to the deal’s account name, set the date range to the quarter, and AI Activity reads as a chronological story: every field nudged, every draft generated, every task logged, every enrichment processed, in order, with the reasoning behind each.

​
How to use it

A rep wraps up a Friday afternoon, opens AI Activity, scopes to My Opportunities, sets the date range to the last seven days, and reads the table top-to-bottom. Fourteen rows: six field-update Accepts on Last Activity Date, three drafted-email Reviews she sent to Gmail, two task creations from meetings she ran, two account enrichments where she accepted four of seven proposed fields, and one Stage update on a $480K Negotiation deal that she rejected because the AI proposed Closed Won before the contract was signed. She clicks View on the rejected row, reads the AI reasoning (“Procurement-review email from champion on Jun 9 references signature timeline; suggesting Closed Won”), confirms the call, and notes that the same field on the same column has fired twice with similar reasoning this month. She opens the column’s instruction dialog, tightens the prompt to require an explicit countersignature reference, and saves. Eight minutes from page load to a calibration that prevents the same near-miss next week.

​
Patterns that work

Treat AI Activity as a Friday-afternoon read. The page rewards a weekly habit, not a daily one. Scope to My Opportunities, set the range to the last seven days, and read the table top-to-bottom. You’re not looking for new work; you’re looking for what the AI did on your behalf, where it landed well, and where the next call was close. Fifteen minutes a week, more than enough to keep the agent honest. Use the page to find what should have fired but didn’t. The table answers “what did the AI do” but it also surfaces what the AI didn’t do. If a deal moved Stages last week and there’s no Field Update row against it, the AI didn’t see the move; if a meeting ended and there’s no Task or Email Draft row, the recognition missed it. These gaps are the most valuable signal on the page: they tell you where to tighten the prompts on AI Suggestions columns or where to flag a meeting that wasn’t recognized. Watch your accept-to-reject ratios. A column where every suggestion is accepted is well-tuned; a column where half of every batch is rejected is misconfigured. Filter to one action type, scan a few weeks, and the pattern reads itself. Reject-heavy columns are candidates for prompt tightening on the Update Instruction dialog; accept-heavy columns are candidates for bulk-accept sweeps from the AI Suggestions surface. Save the audit for compliance, not for coaching. AI Activity is the forensic log; Pipeline Hygiene is the coaching surface. When a customer security questionnaire asks “what AI actions have you taken on our account this quarter,” scope to the account, set the range to the quarter, and the table is the answer. When you’re trying to coach a rep on which suggestions to trust, the per-deal review on Pipeline Hygiene is faster than scrolling AI Activity.
  • Signals - signal accept and dismiss decisions surface here as durable rows.
  • Notifications - the Bell and Toast surface the live arrival; AI Activity is the cumulative log of what those notifications represented.
  • AI Suggestions - the per-cell accept-and-reject layer; field-update rows on this page are the audit trail of the same recommendations.
  • Actions - the inbox-style queue for pending suggestions; AI Activity holds the same rows once they resolve.
  • Account detail - the Activity History sub-tab is the per-account slice of this same log.
  • Audit AI activity this week - the Friday-afternoon routine that opens here.
  • Tune your signal-noise budget - using accept-and-dismiss patterns to calibrate the catalog.