How We Automated a Financial Services Deal Pipeline With AI and Make.com

A merchant cash advance funder was spending 30 to 40 minutes of manual work on every deal that came through the door. Email triage, PDF bank statement analysis, financial scoring, underwriter outreach: all of it done by hand, one deal at a time. At 20 deals per day, that consumed the equivalent of 1.5 full-time employees just on data entry and document processing. We built a four-scenario Make.com pipeline powered by AI that reduced that to under 1 minute per deal, saving 10 to 13 hours of analyst time daily.

The challenge

The funder works with a network of 150+ referral partner organisations who source merchant applications and submit them for underwriting. The entire deal flow was manual.

Every submission arrived as an email with PDF attachments: bank statements, application forms, credit reports. Staff had to open each email, identify which partner sent it from 150+ possible sources (complicated by forwarded emails burying the real sender two or three layers deep in HTML), download the attachments, and manually create a deal item in monday.com. That step alone took around 5 minutes per deal.

Then came the bank statement review. An analyst would open each PDF (typically three months of statements), manually read through every transaction, calculate monthly revenue, deposit counts, average daily balances, NSF and overdraft counts, and identify existing cash advance positions from competing funders. This step consumed 15 to 25 minutes per deal.

After that, using a calculator or spreadsheet, analysts computed debt-to-income ratios, risk scores, cash flow strength ratings, and made approve, decline, or review decisions. Results were manually typed into monday.com columns. Another 5 minutes. Finally, staff manually composed emails with the correct attachments and summary data to the assigned underwriter contacts. Another 5 minutes.

The maths was straightforward: the process could not scale without adding headcount. Deal volume was growing, but deals were sitting in inboxes for hours before being reviewed. Slower funding times meant lost competitive advantage.

What we built

We designed and delivered four interconnected Make.com scenarios over 6 weeks, replacing the entire manual workflow from email intake through to underwriter submission.

Scenario 1: Email intake and deal creation

A mailhook watches for incoming partner submissions. An AI agent identifies the sending partner from the 150+ known organisations, handling forwarded emails, signature parsing, and domain matching. A second AI agent classifies each PDF attachment (bank statement vs application form vs credit report vs other) and extracts structured data: company identifiers, business name, address, industry code, start date, and ownership details. The system checks for duplicate merchants by business identifier before creating the monday.com deal item with subitems, uploading each file to the correct file column.

Scenario 2: AI-powered bank statement analysis

This is the technical core of the system. Triggered when a subitem status is set to "Scrubbing", it runs in four sequential stages:

Stage 1 (Bank Metrics): AI extracts every transaction from each bank statement PDF into structured data. A validation module cross-checks extracted totals against statement summary figures using integer-cents arithmetic (avoiding floating-point rounding errors on financial calculations). Calculates total revenue, monthly deposits, average daily balance, negative balance days, NSF counts, and per-month deposit breakdowns.

Stage 2 (Existing Debt Analysis): AI identifies existing cash advance positions by detecting recurring payments to known funders in the transaction history. Calculates total monthly debt obligations and debt-to-income ratio.

Stage 3 (Business Rules and Scoring): Aggregates all metrics and applies configurable underwriting rules to produce a cash flow strength rating, risk score, debt-to-income ratio, and an automated approve, decline, or review decision with decline reasons where applicable.

Stage 4 (Write-back): All 17 computed fields are written back to the monday.com subitem automatically.

Scenario 3: Funding portal reply parsing

Watches for deal package confirmation emails from funding portals. AI extracts company name, partner, status, and deal package details, matches to the correct monday.com subitem, and writes portal status back automatically.

Scenario 4: Underwriter email distribution

Triggered when a subitem status is set to "Sent to UW". Downloads all deal files (application, bank statements, credit report), resolves the assigned underwriter's contact details from the linked contacts board, and sends a formatted email with all attachments and the scrubbing summary data.

The technical approach

Several engineering challenges shaped the design:

Forwarded email parsing. Partners often forward submissions through internal staff, burying the actual sender two or three layers deep in email HTML. The AI agent navigates this chain to identify the true source, which is essential for correctly attributing deals to the right partner.

PDF table extraction accuracy. Multi-page bank statement PDFs come in varying formats from different banks. We built credit/debit column verification, balance reconciliation, and statement total cross-checks to ensure extraction accuracy. When extracted totals do not match the statement summary figures, the system flags the discrepancy for manual review rather than writing incorrect data.

Financial calculation precision. All monetary calculations use integer-cents arithmetic to eliminate floating-point rounding errors. This is not optional in financial services: a rounding error that compounds across hundreds of transactions will produce incorrect underwriting decisions.

Duplicate detection. Pattern-based identification of recurring payments to known funders across transaction histories allows the system to assess existing debt positions. This directly affects the debt-to-income ratio and the automated scoring decision.

The results

At 20 deals per day, that is 10 to 13 hours of analyst time saved daily, equivalent to 1.5 FTE. At 40 deals per day, the saving reaches 3 FTE.

Beyond the time savings, the system delivers consistency (every deal scored by the same configurable rules, no analyst variance or fatigue errors), speed to underwriter (deals reach underwriters in minutes rather than hours), a full audit trail in monday.com, and the ability to scale deal volume without adding headcount.

Key takeaways

AI document processing is production-ready for financial services. The combination of AI extraction with validation checks and integer-precision arithmetic means the system handles the accuracy requirements that financial workflows demand.

Make.com handles orchestration complexity well. Four scenarios with 30+ modules, working across email, AI, monday.com, and custom code, running reliably at volume. The key is designing each scenario with a clear trigger and a defined handoff to the next.

The ROI is immediate. At 20 deals per day, the system saves more analyst time in a single week than the entire implementation cost. For businesses processing high volumes of structured documents, AI automation is not an experiment; it is a competitive necessity.

If your team spends significant time on manual document processing, data entry, or rules-based decision making, the same approach can be applied to your workflow. Our AI-powered automation service builds these pipelines using Make.com, monday.com, and AI, designed around your specific process.

Want to understand what automation could look like for your operation? Book a free consultation and we will walk through the specific opportunities in your workflow.