Twenty-seven engagements. The numbers, with the stories behind them.
We measure two things: hours reclaimed per week and time-to-first-value. Every engagement gets a public-to-you dashboard from day one. Below: four of them, with permission.
Four engagements, four shapes of problem.
Industries are accurate, names are anonymized except where the client gave us permission. Numbers come from the dashboards we build during the engagement.
Quote drafting agent for inbound RFQs
Tri-State Mechanical Supply
- Problem
- 35-minute average quote turnaround was costing winnable deals — the customer who replies in 5 minutes wins more than the one with the best price.
- Approach
- Built an agent that parses RFQ emails, looks up pricing with margin rules, drafts a Quickbooks estimate, and waits for a sales rep to approve before sending.
- Stack
- n8n · OpenAI GPT-4o · QuickBooks API · Slack · Postgres
- Duration
- 3-week pilot, then a quarter of expansion
Maintenance request triage + dispatch
Westridge Properties (anonymized)
- Problem
- Residents were emailing, calling, and texting maintenance requests to four different inboxes. Average resolution time was 4.2 days; complaints were climbing.
- Approach
- One inbox. An agent that classifies, prioritizes (emergency vs scheduled), and dispatches to the right vendor with a photo and address. Resident gets a status update text.
- Stack
- Twilio · Anthropic Claude · n8n · Supabase · Stripe Connect (vendors)
- Duration
- 6 weeks · then fractional
Sales research agent in Linear
Confidential (Series B SaaS)
- Problem
- SDRs spent 2.5 hours per qualified meeting on pre-call research. New reps took 6+ months to ramp because no one had time to mentor them.
- Approach
- An agent invoked from Linear (where their AEs already live) that produces a structured brief: company moves, recent funding, news, top contacts, suggested angles.
- Stack
- Anthropic Claude · Linear API · Exa search · Clay enrichment · Notion
- Duration
- 8 weeks
Receipt & invoice OCR pipeline
Hartwell & Sons CPAs
- Problem
- Tax season meant 3 staff doing nothing but data entry from PDF receipts and invoices into QuickBooks. Slow, error-prone, and impossible to scale.
- Approach
- An ingest pipeline that OCRs anything dropped in a shared Google Drive folder, classifies it, extracts line items, and stages it in QuickBooks for one-click approval.
- Stack
- Google Drive API · Tesseract + Claude · QuickBooks · Postgres · Retool admin
- Duration
- 5 weeks · then full handoff
Four numbers on every engagement dashboard.
Hours saved / week
Measured against a baseline we capture together in week one. Audited monthly.
Workflow uptime
SLO target: 99.5%. Pages an engineer if it drops, with a clear runbook.
Cost per task
Inference + infra cost amortized across a unit of work, so you can decide what’s worth automating next.
Eval pass rate
A regression suite specific to your workflow. Runs on every change. Pages if it drops more than 3%.