Challenge
Easy Logistics Management runs a U.S. freight, warehousing, and parcel business on Pipedrive—50,000+ contacts, growing about a thousand a week. There was no reliable way to see who had gone cold for 60–90 days. The sales team could not write personalized follow-ups at that scale, a Make.com workflow never held, and revenue was leaking because contacts fell through the cracks.
Solution
ASSOSIATIX designed and built an AI CRM automation layer on Pipedrive. The system finds contacts idle past a configurable threshold, uses GPT-4 to draft follow-ups from real conversation history, and sends through Office 365—either after human review or on a daily unattended run.
Implementation
We owned the engagement end to end: dashboard and leads surfaces on live Pipedrive data, a paginated stale-contact engine for 50k+ records, draft review and bulk send, a no-code system-prompt editor for brand voice, SMTP sending with a full audit trail, and cron automation that chains pages in the background so no single request times out.
Outcome
The workflow no-code tools could not deliver: cold contacts identified at scale, AI drafts in the client's voice, send with or without review, and a complete history of every attempt—so leads stop disappearing from the pipeline.
What we built
Stale-contact engine
Pipedrive persons are fetched a page at a time, filtered by last activity against a configurable 60–90 day window, then cross-checked against our own send log so a contact emailed recently is never treated as cold.
GPT-4 with real history
Before each draft, the system pulls recent Pipedrive mail threads, strips HTML, and injects that context into GPT-4. Output is parsed through a defensive pipeline—JSON, regex, then a template fallback—so generation never returns empty.
Review or auto-send
Operators can read, edit, and bulk-send drafts from the stale-contacts UI, or flip auto-send and let a daily cron generate and dispatch. Both paths share the same suggestion cache and used-flag so a contact is emailed once.
Scale without timeouts
Three layers of pagination keep 50k+ contacts inside serverless limits: Pipedrive pages of 100, small GPT batches with delays and 429 retry, then fire-and-forget chaining so the next page starts after the current response returns.
Brand voice without deploys
System prompts live in the database and are edited from Settings—company services, tone, CTAs, and Pipedrive field placeholders. The client can change how the AI writes without touching code.
Audit trail
Every send attempt is logged with subject, body, success or error. Cached drafts are reused on reload so OpenAI is not called twice for the same contact.
$./start-project.sh
READY TO BUILD
WHAT'S NEXT?
Tell us where you are and where you want to go. We'll help you choose the right path—storefront, system, product, or a focused growth sprint.