One CRM, four retired tools, no cut-over weekend.

About the customer

The customer is an operations-heavy mid-market business whose customer and finance data lived across four systems — a CRM (HubSpot), QuickBooks Online, a spreadsheet AR tracker, and a Plaid-connected bank dashboard — with manual reconciliation between each. The function owner reports to the CFO; the engagement was sponsored at the COO/Head-of-Operations level.

Customer name and per-engagement specifics are templated until written attribution is confirmed. The architecture, rollout cadence, and bill-pay rail decisions are unchanged from the live deployment.

The Challenge

Three converging pain points. First, AR was visible only after invoices were issued from QuickBooks; the CRM (where the actual customer relationship lived) had no view of who owed what, so account managers were blind to past-due conversations they were having in real time. Second, AP was fully manual — bills arrived in a shared inbox, were forwarded to bookkeeping, and only landed in QuickBooks 3–7 days later, with categorization done by a single person whose mental model was the only source of truth. Third, bank-side activity (deposits, fees, transfers, card spend) was reconciled monthly, which meant any anomaly from week 1 wasn’t caught until week 5.

The hard requirement was that any replacement could not require a hard cut-over weekend. Existing automations (Stripe webhooks, the inbound invoice email address, Plaid sandbox linkages) had to keep running while AiTCRM took over view-of-record one module at a time.

The Solution

Intelligent IT deployed AiTCRM under the customer’s tenant on the multi-tenant Clerk + Supabase RLS plumbing. The rollout was sliced into four weekly increments. Week 1: AR view of the existing invoice + invoice-line data, plus shared-inbox intake from the existing forwarding address. Week 2: AP module — vendors, bills, bill lines, payment records — with bill intake driven by a Postmark webhook into a Manny-vision parser that extracts vendor / amount / due-date / line items, dropping a draft Bill into the queue for human approval. Week 3: bank aggregation via Plaid (sandbox first, then production) and statement OCR for credit-card statements that don’t surface clean transactions through Plaid. Week 4: trainable categorization — a rules engine plus a feedback loop that captures every manual re-categorization as a labeled training row.

Bill pay rail is Plaid Transfer (no held funds, no money-transmitter exposure on the platform); until the customer’s KYB completes, the schema falls back to Payment.rail = MANUAL with no migration cost. Stripe is wired for the AR side via Resend-delivered invoice links.

Results

Outcome metrics after the first full quarter on AiTCRM. Numbers below are templated; live metrics are confirmed at end-of-quarter close.

Templated metrics will be replaced with live customer numbers once written attribution is on file.

What’s next

Phase 2 expands beyond AP/AR into customer-specific workflows: P-card reconciliation, multi-entity consolidation, and deal-room → contract → invoice handoff. Phase 3 introduces the embedding-driven semantic search across bills, invoices, and emails (pgvector is already in the schema), and turns categorization v1 into a per-tenant fine-tuned classifier that learns from category_feedback over time.

• Multi-entity / consolidated reporting across the customer’s subsidiaries
• Spend-policy guardrails: auto-flag bills outside category budget envelopes
• Embedded analytics tab driven by the customer’s own reporting questions
• Cross-module embeddings: “show me everything we know about [vendor or customer]”

In their words

About Intelligent IT

Intelligent IT (a brand of Intelligent Group) builds AiTCRM as part of the AiT product suite. The CRM combines customer relationship workflows, AR, AP, bank aggregation, statement OCR, and trainable expense categorization into a single tenant-aware Postgres schema with Clerk-backed RLS.