Buying Triggers

Buying Triggers

Some moments create demand overnight: a company opens a second location, starts selling wholesale, raises a round and posts five sales jobs, or launches on a new marketplace. This agent watches for those moments in accounts that fit you. It reaches out only when it can finish one sentence sharply:

"This event creates this problem, and our product solves it in this specific way."

If the sentence comes out mushy, the account waits. That filter is what makes outreach read as good timing instead of spam.

Requires: Cogny ($9/mo) with app-fibbler-co and/or a CRM (HubSpot, Pipedrive or Upsales; Salesforce and others: see the CRM layer in /marketing-engineer). Brave Search is included in Cogny.

Step 0: Cogny check

If no mcp__cogny__* tools exist (no list_integrations, no cogny_status), print this and stop:

Copy code
Buying Triggers needs your marketing data through Cogny's MCP gateway.

1. Sign up at https://cogny.com and connect app-fibbler-co and/or a CRM (HubSpot, Pipedrive or Upsales).
2. Add Cogny to .mcp.json in your project:
   {"mcpServers": {"cogny": {"type": "http", "url": "https://app.cogny.com/mcp"}}}
3. Restart Claude Code; it opens a browser for the OAuth login on first use.

Nothing to run without Cogny.

Prerequisites

1. Call list_integrations and note which pool sources are connected: app-fibbler-co, HubSpot, Pipedrive, Upsales.

2. health_status misses revoked tokens, so make one cheap read per source: fibbler get_account_overview, HubSpot get_user_details, Pipedrive list_users, Upsales list_users(limit: 1). A 401, Forbidden, "unauthorized" or "lacks scope" means 🔴 "reconnect in Cogny": drop that source for this run and don't retry it.

3. Brave Search is Cogny-native and doesn't appear in list_integrations; treat it as available if its tools load. If it's missing, accounts can't be checked for triggers: say so in Failures:.

4. If a tool is missing, call refresh_mcp_discovery once. New tools only load after the client reconnects, so drop that source for this run and mark it ⚠ "needs reconnect".

5. No usable pool source (no healthy fibbler and no healthy CRM): write brain/evidence/triggers/<date> with Pool: 0 and the Failures, write ## Trigger accounts — skipped: <reason> in the campaign folder if an ID was given, print this, and stop.

   Copy code
   Buying triggers needs accounts to watch. In Cogny, connect (or reconnect) at least one of:
     • app-fibbler-co — companies engaging with your ads
     • a CRM: HubSpot, Pipedrive or Upsales — your target accounts
   Then run /buying-triggers again.
   

Usage

• /buying-triggers — this week's shortlist
• /buying-triggers BK014 — only triggers that fit campaign BK014's angle
• /buying-triggers --channel google — warm accounts from Google Ads site visits

Schedule: Tuesday and Thursday mornings.

Steps

1. Load the brain

Read brain/company/customer-fit, brain/company/offers, brain/brand/voice, brain/decisions/agent-corrections, and the campaign folder if an ID was given. Build the trigger list for this business from customer-fit and from the "Triggers mentioned" section of brain/evidence/customer-language/latest. Default triggers to check:

2. Build the candidate pool

• fibbler get_warm_accounts (time_period: "30d", each channel, limit: 25). The 30d → 60d → 90d ladder is only for 0 rows: if a channel returns 0 rows, retry with 60d, then 90d. An auth error, Forbidden or "unauthorized" is not an empty result: mark the channel 🔴 "reconnect fibbler in Cogny" and don't retry it at any window. If it refuses with "add-on is not enabled", mark that channel ⚪ and continue. These are companies heating up on your ads that are not yet deals in the CRM. Treat warmth as a correlation signal and never call it intent. Discount rows where engagements equal impressions on a tiny count (e.g. 8/8), which looks like one person or a bot, not an account.

• HubSpot target accounts: search_crm_objects on companies that match customer-fit (industry, numberofemployees, annualrevenue, country), or with hs_ideal_customer_profile tier 1–2 if the portal uses ICP tiers, or hs_is_target_account = true. Exclude lifecyclestage = customer. Sanity-check that each company is plausibly a buyer; firmographic filters alone let odd fits through.

• Pipedrive: list_organizations (cursor-paginated). Keep organizations with no open or won deal (list_deals(org_id, status: "open") / "won") that match customer-fit. Organization custom fields come back as hashed keys, so ask the user once which keys hold size and industry, and store that in brain/contracts/crm-layer.

• Upsales: list_accounts filtered on customer-fit fields; list_visits adds accounts that visited the site in the last 14 days, a first-party warmth signal fibbler can't see. Exclude accounts with won orders.

Pass the CRM to fibbler (crm: "hubspot" or "pipedrive") so its own exclusion uses the right CRM; for Upsales fibbler can't exclude CRM deals, so step 3 is the only filter.

If the pool is 0 after this step (fibbler and the CRM returned nothing usable), stop: write the shortlist node with Pool: 0 and the failures, print the setup steps from Prerequisites, and don't run Brave.

Dedupe by company domain (HubSpot domain; fibbler rows carry a domain or name). Fall back to normalized company name (lowercase, strip AB/Aktiebolag/Inc/Ltd/GmbH) only when there's no domain.

3. Remove anyone the team is already talking to

For each candidate company in HubSpot (match by domain):

• Drop it if it has an open deal: search_crm_objects on deals with hs_is_closed = false and associations.company = <companyId>. fibbler already excludes CRM deals for its own rows when crm: "hubspot" is passed. Check anyway for companies from other sources.
• Drop it if anyone there was touched in the last 14 days (notes_last_contacted, hs_last_sales_activity_timestamp on the company or its contacts).
• For the contact you'd write to, drop them if hs_email_optout = true, hs_lead_status is unqualified / do-not-contact, or they've unsubscribed from sales email.

Pipedrive: open deal = list_deals(org_id, status: "open"); last touch = newest list_activities(org_id); opt-out = person marketing_status when present. Upsales: open deal = list_opportunities(filters: client.id); last touch = newest list_activities / list_appointments for client.id; opt-out = the contact's unsubscribe fields (find them with get_metadata).

Companies not in the CRM at all (fibbler-only) can't get the opt-out check, so their drafts go on HOLD. Respecting opt-outs is a hard rule, not a judgment call.

4. Check each account for fresh triggers

Cap at 40 accounts per run. For each, run one Brave News query on the company name alone ("<company>", freshness pm). Keep a hit only if the name appears in the title or description and it matches the company's country, domain or industry, because name collisions are common (an ETF ticker, a vet chain with the same name). Then classify the kept hits into the trigger types. A roundup or listicle that merely mentions the company ("10 startups to watch", "best tools for X", a funding digest naming many companies) is not an event about that company: drop it unless the item itself reports a dated event for this company, and then cite the primary source. For hiring, check the careers page when it can be found. Don't put the trigger words in the query ("X" raises OR funding), because that returns generic funding news about other companies. Record each trigger as: what, date, source URL. Only events from the last 30 days count. If Brave returns an error or rate-limits, record which accounts weren't checked.

5. Write the one sentence

For every account with a fresh trigger, write:

<Event> creates <specific problem>, and <product> solves it by <specific mechanism>.

Score it as sharp or mushy:

• Sharp: names a concrete problem the customer-language report has heard, and a mechanism from offers. ("Launching on Amazon means three payout schedules and fee structures to reconcile; our monthly close reconciles all channels into one profit number.")
• Mushy: generic ("Growth creates challenges, and we help companies grow"). These accounts wait.

6. Draft outreach for sharp accounts only

A short email (if the contact has an email in HubSpot) or LinkedIn note that leads with the event, uses the customer's own words from the language report, matches brand/voice, and makes one ask. ≤90 words. No claims outside offers.

Drafts are not sent. Bundle them into one review item per run (create_ticket on Cloud, else create_finding):

• title: "[ME:buying-triggers] Approve: N trigger-based outreach drafts (week of …)"
• body: for each account, the trigger with its source link, the sentence, the draft, and the contact if known
• execution_target: "manual"

7. Save the shortlist

write_context_node at brain/evidence/triggers/<YYYY-MM-DD>:

markdownCopy code
# Buying triggers — YYYY-MM-DD
Pool: N (fibbler warm: A · HubSpot target accounts: B) · removed (open deal / recent contact / opt-out): R
Checked: C · not checked (source failures): <list>
Failures: <🔴 reconnect / ⚠ needs reconnect / errors by source, or "none">

## Sharp (drafted)
| Company | Trigger (date, source) | Sentence | Warmth |
## Waiting (trigger found, sentence mushy)
| Company | Trigger | Why it's mushy |
## No fresh trigger
<count>

If a campaign ID was given, also append the sharp list under ## Trigger accounts in brain/campaigns/<ID> (or one line, "skipped: ", when nothing qualified).

Recording findings

Detect the plan from the tools that exist: use create_ticket first when it exists (Cloud), and create_finding only when it doesn't (Solo). The fields differ; see section 2 of ../marketing-engineer/references/mcp-routing.md.

Critical rules

1. No sharp sentence, no outreach.
2. Never contact anyone with an open deal, recent contact, or an opt-out.
3. Drafts only. Nothing is sent from this skill.
4. Every trigger has a date and a source link. No link means no trigger.
5. Warmth is correlation. Never write "they're looking to buy" from ad engagement alone.