Hardest Part

Achieving strict deterministic accuracy on probabilistic LLM outputs to guarantee financial compliance, preventing hallucinated matches from corrupting the general ledger.

Min Viable Scope

Focus exclusively on matching bank feed deposits to open accounts receivable invoices in NetSuite for US-based B2B software companies. Deliberately leave out multi-currency support, inventory reconciliation, accounts payable, and secondary ERP integrations.

Cold Start Problem

Models fail on idiosyncratic accounting codes and messy vendor strings without prior context. Break this by securing three mid-market design partners to extract historical manual reconciliation logs for initial supervised fine-tuning.

Time To First Value

1 full month-end close cycle to validate shadow-mode matches against human output

Data Moat Available

true

Technical Difficulty

High

Wedge

Target mid-sized e-commerce brands handling high volumes of Shopify, Stripe, and bank transactions. This niche experiences acute reconciliation pain daily due to hidden platform fee deductions, currency fluctuations, and delayed multi-gateway payouts. Expansion moves from e-commerce payment reconciliation into B2B SaaS revenue matching, and finally into full general ledger closing for traditional retail.

Timing

Foundational models now reliably extract context from unstructured remittance advice and map it directly to rigid database schemas. This allows automated resolution of the anomalous transactions that consistently break legacy deterministic rule engines.

Why This ICP

Mid-market finance teams face transaction volumes high enough to break manual spreadsheets but lack the internal IT resources required to implement heavy enterprise integrations.

Size Of Prize

Approximately 200,000 mid-market companies in the US and Europe spend an average of $35,000 annually in labor hours and basic software on ledger reconciliation, yielding a total addressable prize of roughly $7B.

Gap Narrative

Mid-market finance teams manually match thousands of transactions across bank feeds, ERPs, and payment gateways at month-end to find missing cents and categorization errors. Current rules-based systems fail on unstructured invoice data and multi-currency anomalies, forcing accountants to dump data into Excel for line-by-line reconciliation.

Defensibility

Defensibility compounds through workflow lock-in as the system internalizes a client's specific chart-of-accounts logic and custom vendor naming conventions. Over time, the system builds a proprietary mapping graph of how edge-case transactions resolve across different ERP setups, continually driving the exception rate toward zero and making the switching cost prohibitively high.

Why This Thesis

A Service-as-Software approach replaces the reconciliation workflow entirely, delivering fully reconciled ledgers with isolated exceptions as the final output. This direct labor replacement fits an ICP that wants to eliminate headcount rather than purchase another dashboard to manage.

SAM

~$400-600M US mid-sized accounting firms and outsourced CFO practices

SOM

~$15-30M

TAM

~100k US accounting firms × ~$15k/yr software equivalent ≈ $1.5B

Growth Rate

~12-18%/yr, driven by ongoing shortages of qualified CPAs and increasing client transaction volumes

Paid Comparable Spend

~$40k-60k/yr per firm spent on offshore junior bookkeepers and data-prep spreadsheet add-ons

Kill Thresholds

• Auto-match rate remains below 60 percent after 30 days
• Human escalation exceeds 30 percent per reconciliation batch
• D60 retention drops below 50 percent for pilot firms
• Customer acquisition cost exceeds $8,000 for a $15,000 ACV

Leading Metrics

• time-to-first-reconciled-ledger
• auto-match rate for incoming transactions
• human-in-loop escalation percentage per batch
• integration setup time in hours
• number of ledgers connected per firm

What Proves Right

Firms process 80 percent of daily transactions without human intervention within the first 30 days of deployment. Cohorts retain at over 90 percent after three months, proving the system effectively replaces junior offshore labor. Firms pay $1,500 per month per client entity because the software cuts data-prep time by 40 hours per month.

What Proves Wrong

Firms revert to manual Excel macros because the AI flags too many false positives during reconciliation, requiring more human review than standard processing. Security and compliance barriers block API access to client bank feeds, capping the automated ingestion rate below what is needed for positive return on investment. The product fails to handle complex general ledger codes, causing the human-in-the-loop escalation rate to stall above 40 percent.