What this is
An AI-powered extraction and reconciliation engine that automatically ingests financial data from Google Sheets, reconciles it against Ramp's spend data, and surfaces discrepancies, missing entries, and reconciliation intelligence. The product sits between a company's ad-hoc sheet accounting and Ramp's structured spend management system, automatically bridging the gap that exists in most SMB finance teams where sheets are the source of truth and Ramp is the transaction engine.
Core flow: user authorizes Ramp + Google Sheets access, AI reads sheets monthly (expense reports, vendor ledgers, accrual logs, manual journals), maps rows to Ramp transactions by vendor/amount/date, flags mismatches, suggests categorizations, and exports a reconciliation report. Think of it as an automated monthly close assistant that speaks both sheet and structured finance languages.
Why it's interesting
This solves a real pain point for Ramp's mid-market segment (Series A/B companies, $10M-$100M ARR). Finance teams at this stage have already adopted Ramp for spend control but still maintain parallel sheets for accruals, recharges, client billing, and cross-functional reconciliation. They run month-end close manually: exporting Ramp data, comparing to sheets, flagging discrepancies via email, chasing ops teams. A 3-5 person finance function loses 2-3 days per month to this.
Ramp owns the customer relationship and spend data but not the reconciliation workflow. Building this natively gives them:
- Higher product stickiness (finance team becomes dependent on the reconciliation output).
- A new moat: normalized financial data becomes training data for spend pattern detection, anomaly alerts, and predictive categorization.
- A wedge into the broader close/consolidation problem (eventually, integration with bill.com, accounting software, consolidation stacks).
- API upsell: expose reconciliation data to ERP/accounting integrations.
The market is real. Every mid-market finance team we've talked to does some version of this manually. Ramp has direct access to 10K+ potential users.
Why a landing page would fail
A landing page implies "easy adoption." It isn't.
First, Google Sheets layouts are not standardized. Company A's expense report has vendor in column B, amount in column D, date in column A. Company B has a completely different structure across multiple sheets. Company C has a hybrid: some data in Sheets, some in a separate tool. The AI has to learn each company's schema, which requires either extensive upfront documentation or multi-turn interaction with the finance person to validate correctness. Neither scales via a landing page.
Second, the value is entirely in the reconciliation delta. If a company's Sheets and Ramp are already in sync (few mismatches), the tool feels like noise. If they're wildly out of sync, it means the finance team has trust issues with one source or the other, and an automated tool suggesting fixes can do more harm than good if it's wrong. Early users will be skeptical. Building trust requires explainability, audit trails, and high accuracy from day one, not iterative improvement post-launch.
Third, Ramp's existing sales motion doesn't have a hook for this. It's not a module in the onboarding flow. It's a monthly workflow add-on. Selling it requires a champion in finance (not procurement, not the CFO) who has enough pain and authority to approve a new tool. That's a different sales conversation than "Ramp controls your spend."
A landing page would generate interest from tire kickers and procrastinators, not buyers. You'd need a full sales team working tight partnerships with Ramp's onboarding and customer success teams to place this correctly.
The realistic shape
Team: 1 founding engineer (AI/LLM ops), 1 product/growth person, 1 part-time Ramp liaison. Contractor support for initial customer onboarding and sheet schema mapping.
Architecture:
- Multi-modal LLM pipeline (vision for complex sheets, structured prompts for clean ones).
- Per-company schema store and validation layer.
- Ramp API integrations (read spend, categories, vendors).
- Google Sheets API + OAuth.
- Monthly batch reconciliation with streaming results to a dashboard.
- Explainability layer: every match/mismatch decision annotated with confidence and reasoning.
Capital: $500K seed to cover 12 months of payroll, infrastructure, and 10-15 customer onboarding projects.
6-month milestones:
- Months 1-2: Core extraction pipeline working on 5 customer sheet schemas. Manual schema design, not auto-detection.
- Months 3-4: Semi-automated schema learning. Three paying customers, each doing one month of reconciliation with human review.
- Months 5-6: Dashboard live, 8-10 paying pilots, first accounts reporting 80%+ automated reconciliation (>300 transactions matched per month). Start building explainability features.
Honest 12-month case
Best case: 25-30 paying customers, $3K-$5K MRR by month 12. Customers see 15-20 hours/month saved. Net retention trends positive as teams expand usage. Platform achieves 85%+ match accuracy. Ramp moves to acquisition or integration discussions.
Base case: 10-15 customers, $1.5K-$2K MRR. Adoption stalls because schema variance is higher than expected and sales motion didn't scale. Team is acquiring customers but not fast enough to justify next capital round. Platform needs either deeper verticalization (specific to e-commerce, SaaS, etc.) or integration into Ramp's core product to survive.
Kill criteria:
- Schema detection doesn't improve beyond 40% accuracy within 4 months.
- Fewer than 5 paying customers by month 6.
- Reconciliation accuracy plateaus below 75% despite increasing training data.
- Ramp signals they won't integrate or co-sell.
Five questions to answer before committing
1. Can Ramp's sales team actually sell this, or does it need independent GTM? If Ramp sees it as a feature opportunity rather than a product line, you're building on uncertain distribution.
2. How much of the value is AI automation vs. the schema mapping itself? If 70% of the win is just knowing how to read the customer's sheets, the AI moat evaporates and this becomes a services business.
3. What's the unit economics on customer onboarding? How many hours of manual schema work per customer to get to 80% accuracy? If it's more than 20 hours at $150/hour, the model doesn't scale.
4. Does Ramp already have 100+ customers asking for this, or are we creating demand? Validation matters. If the problem is acute and Ramp's sales team hears it regularly, this is easier to sell. If not, you're fighting inertia.
5. Who else might build this first? Bill.com, Sage, Stripe? The window to own this category closes fast if a larger player decides it's worth the engineering lift.