01 · JUNK
"Missed call" × 3,000+
Telephony creates a record for every call without a name. The base bloats, search breaks, reports lie. Real leads hide among empty shells.
Clean the base of junk and duplicates, assemble a 4-layer architecture (foundation → products → operations → dashboard), turn on AI where it saves human hours. Your sales team stops spending a day on "there was this customer somewhere in the CRM".
Inside — LOOP · operating layer, ECHO · inbound response engine
Telephony creates a record for every call without a name. The base bloats, search breaks, reports lie. Real leads hide among empty shells.
"Acme Corp" = 315 records, big accounts scattered around. One rep works one record, another works the second, nobody knows about the other two.
Half the deals sit at the first stage forever. Nobody moves them, nobody closes them, nobody remembers what was there.
Lead arrives from the site but UTM tags don't land in the CRM. Ads aren't linked to sales. Ads say "many leads", sales say "no clients".
Single entry point for all channels. Dedup by tax ID and root email domain. Auto-close of zombie deals. UTM capture in contact and deal fields.
Not "sales-1 / sales-2 / sales-3", but by client type and deal type. Each product with its own touch cadence, its own fields, its own automation level.
Twilio / Aircall / Dialpad → record with call recording and task. Stripe / QuickBooks ↔ CRM for payment status and A/R. Eventbrite / Hopin for event registrations and QR check-in.
Once a week — owner digest: where big deals sit, what missed SLA, who's overloaded. Role-based dashboards. Access controls per ISO 27002.
Reminder cadences, membership renewal, A/R dunning, event registration intake, inbound triage. Auto-runs without asking.
Invoices, contracts, banking details, client replies. AI writes a draft — rep reads, edits, hits send. Speed without errors.
Big partnerships, large accounts, finance, edge cases. Automation only hurts here — we keep it manual.
3,753 records with names like "Missed call" / "No name given" — sorted into three buckets. 2,575 without contact — deleted (phone numbers saved to CSV). 1,022 with contact — restored the company name from the corporate email domain and correspondence, assigned owner = whoever actually communicated.
Built a "Missed calls" pipeline, seeded it with 148 fresh deals from recent missed inbounds. Filled tax IDs via a lookup API with validation (blocked mismatches like "zara→Global Metal Zaraysk"). Deduped by domain: 159 groups → 125 merges, removed 559 extra duplicates. A live inbound leak on the phone line (88% missed on two reps) — documented for the owner.
Money and outbound to the client — always human sign-off. Automation makes expensive mistakes here. Even "just send the invoice" — with a checklist.
We don't fill tax IDs blindly. Lookup APIs often guess wrong — we only write when a significant word matches. No match — we leave it blank. A gap beats a lie.
Nothing gets deleted without a backup. All cleanup actions — in a CSV with before/after fields. If something's wrong — we restore from the CRM's Trash within 30 days.
The client's data stays with the client. Raw data — local. Reports and repositories carry only aggregates without PII. GATE layer over the LLM guarantees names don't leak (gate).
You send read-only guest access — next day you get a breakdown: where the junk is, where the duplicates are, where the inbounds leak. No commitment.