kortix-ai Read-only and report-only: the agent never edits a deal and never contacts a prospect. The output is a Slack message; a human decides what to change in the playbook.
Step 1 — Pull the closed deals
Through the hubspot connector, read-only, pull every deal whose stage is
closed-won or closed-lost, with: close date, amount, segment/plan (or
company size if no segment field), the deal's stage-history,
{{close_reason_property}}, and {{competitor_property}}.
Step 2 — Isolate the window
Using the close-date field, keep only deals closed in the last {{lookback_days}} days. Because the session is fresh each run, this window is derived entirely from HubSpot's own dates — there is no external ledger to consult, and nothing carries over from the prior week.
Step 3 — Compute the headline numbers
Count closed-won and closed-lost in the window, and the win rate
(won / (won + lost)). Compare total won amount to total lost amount (deal
value at risk, not just deal count).
Step 4 — Break down by segment
Group by segment/plan/company-size band. For each segment, report win rate and deal count. Flag any segment whose win rate is materially below the overall rate — that's where the playbook is weakest.
Step 5 — Break down by competitor
From {{competitor_property}}, group losses (and wins where a competitor was
named and we still won) by named competitor. For each competitor with enough
volume to matter, report the loss count and, where the close reason
mentions it, the reason we lost to them specifically (price, feature gap,
relationship, timing). Deals with no competitor recorded go in an
"unrecorded / no competitor" bucket — don't force a competitor name that
isn't there.
Step 6 — Break down by price
Band deal amount (e.g. small / mid / large, using this account's own distribution rather than fixed dollar cutoffs) and report win rate per band. Note whether losses cluster at the high end (a pricing/negotiation problem) or low end (a fit/qualification problem).
Step 7 — Find where lost deals die
For closed-lost deals only, read the stage history and identify the stage the deal was in immediately before it was marked lost. Rank stages by how many lost deals died there. A stage that kills a disproportionate share of deals relative to how many pass through it is a process problem, not just bad luck.
Step 8 — Cluster close reasons into themes
Read {{close_reason_property}} across all closed-lost deals in the window
and group by what the rep meant, not exact wording — e.g. "lost on price,"
"missing feature," "competitor chosen," "champion left / no decision,"
"bad timing." For each theme, note the count and one representative
verbatim reason. If a reason field is blank, count it under "reason not
recorded" rather than guessing.
Step 9 — Turn themes into recommendations
Pair every named theme (from segment, competitor, price, stage-of-death, or close-reason clustering) with one concrete, actionable recommendation — e.g. "three of five losses to Competitor X cited pricing on the enterprise tier; consider a packaging review for that segment." A theme without a recommendation doesn't make the post.
Step 10 — Post the weekly summary
Post one message to {{alert_channel}}:
- Win rate and won/lost counts and value for the window, vs. what the window looked like the last time this much data closed.
- Breakdown by segment, by competitor, and by price band.
- The stage where lost deals most often die.
- The top themes, each with a count, a representative reason, and a recommendation.
Nothing else is written anywhere. No HubSpot record is modified, and no rep or prospect is contacted.