Connect
We connect Claude to every source that holds a key number, not just your ERP, with per-user permissions.
Teams across your company are building dashboards, Skills and agents on Claude, each defining your key numbers its own way. AMW gives Claude one tested definition for each, so what every team builds compounds instead of drifting apart.
We work with finance teams already on Claude Team or Claude Enterprise. Everything we build lives in your own Claude tenant.
Adam Ware, founder, on why every team's numbers drift apart, and how we fix it.
Adjusted EBITDA, ARR, management and business-unit P&Ls live in analysts' spreadsheets. So every prompt, Skill and agent has to work out the definition on its own, and each one lands somewhere different. Your finance team can't review every number in everything the company builds on Claude, so the wrong ones reach decisions before anyone catches them.
Locking Claude down isn't the answer. Your teams should be able to move fast, on numbers Claude gets right.
We connect Claude to every source that holds a key number, not just your ERP, with per-user permissions.
We build the semantic layer: one definition for each metric that matters, written so Claude knows exactly what to pull.
Our evaluation suite tests a bank of finance questions with known answers against your official reports, then again after every month-end close and every new version of Claude.
From our work with a healthcare company. Read the case study →
Keystead Systems is a fictional $300M company we built to show Tick & Tie end to end. Its data is synthetic, but its books carry the problems real ones do: six entities in five currencies, a fiscal year ending in January, intercompany sales, partner rebates that moved accounts mid-year, an acquisition and a leverage covenant. Each one is resolved by a definition in the semantic layer.
Is Keystead inside its 5.75x leverage covenant? On reported EBITDA, net leverage is 7.04x, which reads as a breach. On management adjusted EBITDA it is 5.75x, exactly at the limit. On covenant EBITDA, the one the credit agreement actually tests, it is 4.94x, with room to spare. The semantic layer keeps all three apart and tells Claude which one each question needs.
50% → 100%
Board-pack questions Claude answered correctly with the vendors' own connectors, versus the same connectors plus the semantic layer. 61 questions, each asked three times, graded by code.
Keystead is a fictional company with synthetic data.
We map where your key metrics live, recommend the approach that fits your stack, and send an affordable fixed-fee quote.
Connect, Define, Certify, inside your Claude tenant.
Optional monthly recertification, plus add-ons when you want them.
Claude connected to your financial data, a semantic layer that defines your key metrics once, and an evaluation suite that certifies them. Three weeks for most stacks, at an affordable fixed fee.
We rerun the suite after every month-end close and every new Claude version, and fix what breaks.
Fixed-fee builds on the foundation: reporting packs, forecasts, close workflows.
Hackathons and builder programs so your team builds on the foundation itself.
Add a warehouse, planning tool or billing system to the foundation and the suite.
Each is an optional, separately priced add-on.
Not sure where to start? We also run a one-week finance AI roadmap with your leadership team. Ask about it on the intro call.
In 30 minutes we'll map where your key metrics live, determine which approach fits your stack, and provide an affordable fixed-fee quote.