Plateau · €2.4M pipeline recovered

Context

Plateau Inc. runs a B2B SaaS at €18M ARR, with 47 employees and a 4-person RevOps team managing all cross-team handoffs between marketing, sales, and finance by hand. In January 2026, their Head of RevOps reached out after six months of rapid growth had exposed three structural bottlenecks eating time and money every week. First bottleneck: payment reconciliation. Six hours per week of manual matching between Stripe and HubSpot, with a 4% error rate generating incorrect dunning messages sent to customers who were fully up to date. The relationship cost was invisible but compounding. Second: inbound lead triage. 127 inbound per month, sorted by two Account Executives spending 3 hours each per day on intake. Average first-contact delay: 48 hours — in a market where competitors were responding in under 4 hours. Third: dunning. The overdue payment sequence was entirely manual. No personalization, no automatic follow-through. Recovery rate: 34% on payments overdue by less than 30 days.

Audit week

Week 1 was a full data flow mapping. We audited the HubSpot ↔ Stripe ↔ Salesforce integrations, identified failure points in each sequence, and measured the real time cost of each manual task. Findings: 28 automatable hours per week with low regression risk. The three bottlenecks accounted for 22 of those 28 hours. Data quality audit ran in parallel: 3,200 HubSpot contacts with a 9% duplicate rate, inconsistent deal stages across sales reps, 14 custom fields with no populated values. Automating without cleanup would have amplified the noise. Day 3 was spent on a targeted data hygiene pass before writing a single line of agent code.

What we shipped

• **Finance reconciliation agent** · Claude Sonnet 4.6 + Stripe MCP + dbt models · runs every 4 hours, resolves 97% of discrepancies automatically, flags the remaining 3% for human review with full context
• **Triage agent** · Claude Opus 4.7 · scores each inbound against ICP criteria and form content, routes to N1/N2/escalate in 8 seconds, triggers automated first-contact within 1 minute
• **Dunning agent** · Sonnet 4.6 + HubSpot MCP · personalized 4-stage sequences adapted to customer tenure, amount overdue, and relationship tone
• **Audit trail** · all agent decisions logged, signed, exportable CSV/JSON, AI Act Art. 12 compliant, 30-day lookback

Results

Measured over 8 weeks post-deployment versus the preceding 8 weeks. €2.4M pipeline recovered: deals that would have stalled without accelerated triage and personalized dunning. 22 hours per week freed for the RevOps team — equivalent to 0.55 FTE redirected to strategic work. Dunning recovery rate: from 34% to 61% on payments overdue less than 30 days. First-contact delay on N1 leads: from 48 hours to under 2 minutes. Payment reconciliation error rate: from 4% to 0.3%. The finance team stopped receiving calls from customers wrongly dunned.

Trajectory

Audit week 1, MVP sprint weeks 2–5, hardening and adoption weeks 6–10. Plateau now runs 14 active agents, with 92% of weekly RevOps tasks fully automated. They migrated from Claude Pro to the Anthropic API with managed agents in week 8, cutting inference costs by 40%. They are currently exploring an open-source fallback for finance data residency to meet contractual requirements from two enterprise clients. Phase 2 covers forecasting automation and automated weekly board memos.