How the Commercial team used AI over the last 14 days. Commercial uses Jira little, so this leans on the AI automation layer — Fin in customer support and AI call-processing in sales — plus measurable tool usage from Honeycomb.
Fin (Intercom) + AI call processing (Aircall→HubSpot) are the AI automation layer for Commercial, which uses Jira little. Last 14 days (16–30 June): Fin was involved in 419 of 636 mobile-chat conversations (66%) and resolved 239 of them. HubSpot logged 6,221 calls, of which 2,605 were AI-transcribed and summarised. Direct Claude tool usage is a measured floor — Chat (claude.ai) is not in OTel. Spend is never shown.
The biggest AI impact in Commercial is systemic, not individual: AI agents and pipelines handling support and sales work directly.
Answers in-app mobile chat (Messenger) support conversations from a 296-article knowledge base plus curated Aiske.ai context snippets; resolves common how-to, registration and account queries without a human. Figures are the mobile chat channel only.
Every sales call is auto-transcribed and AI-summarised (Summary / key notes / topics discussed, with PII redaction) and written back to the HubSpot call record.
Two cloud routines (IE + UK Sales Pipeline Auto-Sync, 06:00 weekdays) read the last 24h of Aircall calls, classify each contact's correct deal stage from the call signal plus Intercom and billing data, and create or sync HubSpot deals directly. The deal owner is set to the rep who took the call; the write is performed by the Claude HubSpot MCP under a connected admin account and each deal is stamped 'Auto-created by … Sales pipeline routine'.
A twice-daily routine triages missed inbound calls on the Irish support lines: it enriches each caller with their Intercom profile and Aircall voicemail transcription, excludes anyone already called back, scores by priority (renewal risk, expiry, urgency keywords), and posts a ranked, click-to-dial action list to Slack so no farmer slips through.
Measured Claude tool usage (Honeycomb: Code / Cowork / Office) per Commercial function. The hatched portion of each bar is the people with measured usage; hover for the % (a floor — see the caveat).
Claude Chat (claude.ai) is not captured in Honeycomb OTel, and Chat is what most Sales, Support and Marketing staff use. So the measured figures (8 of 33 people) are a FLOOR, not the full picture. The Fin and call-processing automation numbers above are exact; the tool-adoption numbers here are conservative.
Two whole functions (USA Growth, Retention & Expansion) show zero measured tool usage — almost certainly Chat-only, not non-users. Confirm via the Teams admin export.
| Function | Surface | Tokens (14d) | What AI is used for |
|---|---|---|---|
| Product Marketing | Cowork + Code Desktop | 1.31B | Content, research and analysis |
| Commercial Leadership (CRO) | Cowork | 61M | Leadership, reporting and account work |
| RevOps | Cowork + Claude Code | 17M | Data analysis and reporting |
| Ent / Vets Ops | Cowork + Code Desktop | 11M | Leadership and account work |
| Sales (B2F) | Claude Code | 8.1M | Pipeline tooling and prototyping |
| Marketing | Cowork | 6.5M | Campaign content |
| Retention & Expansion | Cowork | 1.8M | Renewals and account workflows |
| Onboarding & Support | Cowork | 0.24M | Support workflows |
Concrete, evidence-based moves to push Commercial AI further. Ordered by impact.
Honeycomb only sees Code, Cowork and Office — Claude Chat (claude.ai) is invisible, yet Chat is what most Sales, Support and Marketing staff use. This fortnight only 10 of 29 Commercial people register any measured tool usage, almost certainly an artefact of that blind spot. Pull the Teams admin usage export (as the AI Adoption Leaderboard does) so Commercial adoption can be measured properly rather than appearing as 10/29.
USA Growth shows zero measured AI usage again this fortnight, and Sales, Retention & Expansion and Onboarding & Support each register only a single measured user against much larger headcounts. Run a Cowork/Chat enablement push for these teams — pair each with a power user and measure the lift via the admin export.
The IE and UK Sales Pipeline Auto-Sync routines created 157 deals and categorised 342 more from call transcripts this fortnight. Two improvements: (1) the deals are written by the Claude HubSpot MCP under a personal connected admin account, so HubSpot's 'modified by' shows that account rather than the rep — move it to a dedicated service identity for clean attribution and auditability; (2) the USA team's calls do not flow through it yet — extend coverage.
The USA Growth team shows no measured AI usage and its calls do not appear in the Aircall→HubSpot AI transcription pipeline at the volume the IE and UK teams enjoy. Connect them so US sales gets the same call intelligence, summaries and deal sync as IE and UK.
Fin was involved in 419 of 636 mobile chats (66%) and resolved 239 — 57% of those it engaged. Keep expanding and freshening the 296-article knowledge base and Aiske.ai-context snippets, and tune Fin to push the self-serve resolution rate higher and lift the share of all mobile chats it handles end to end.
The 2,605 AI call summaries this fortnight (plus Aircall sentiment and topics) are an untapped coaching asset. Auto-generate per-team coaching notes, objection libraries and next-best-actions from them for the B2F and USA teams — at team level to keep it developmental rather than surveillance.
Product Marketing is again the standout — heavy Cowork and Code Desktop use this fortnight (1.31B tokens) is the internal proof point that AI works for non-engineering work. Package that pattern as a repeatable Cowork workflow and demonstrate it to Marketing, RevOps and Sales.
Sources: Intercom (Fin, 14-day, exact), HubSpot CALL + DEAL objects (14-day, exact), Honeycomb Claude OTel (tool floor), AI cost-benefit model (Outcome estimate).
commercial-ai-usage-report skill, a companion to the AI Adoption Leaderboard and the P&T outcomes page. AI spend ($) is intentionally not shown. Contains customer/PII-adjacent counts — anonymise before any external sharing.