Start here
The Baseline
Before you sign an AI contract or a headcount plan, know what your team’s work actually is. In two to three weeks you get your real customer interactions measured as work, not tickets: hours by type of work, what AI can take net of the new work it creates, your costliest failures priced in dollars, and a staged path your finance team can check.
When to run it
- Someone has a number (“cut 20%”) and you need to test it before you commit.
- You’re about to buy or expand AI and want the “before” measured. You can buy AI in any quarter; you can only measure the before once.
- AI is live, the savings aren’t showing up, and you need to know why.
What you get
Everything is yours, in open formats, whether or not you continue.
- Your work, measured. Twelve months of tickets, calls or success notes, labeled for cause, effort, difficulty, avoidability and outcome, with hours put against each type of work. Accuracy is checked and shown next to every label.
- What AI should take, in hours. Which types of work AI can handle, what share of your team’s hours that is (not tickets), and the new work it creates: knowledge, quality review and escalation design, each with an owner.
- Your top five recurring failures, priced in dollars. Frequency times cost per incident, with the sample and window shown.
- Capture gaps: what you can't see today because of how data is captured, with evidence.
- The path and its value sheet. Staged decisions with gates, built only from inputs your finance and HR teams sign off: what a cut would actually save, and when, with year-one value shown apart from longer-term value.
- A recommendation and a 60-minute readout with you and your CFO or CEO, including which of the five decisions on the path your data supports now.
How it works
Access and definitions
Read-only access, agreed definitions, and 30–45 minute conversations with you, two agents and your finance partner. Your full history is pulled and matched to accounts. Week 1 starts once access is confirmed.
Labels you can trust
Every interaction is labeled, then checked against your senior agent's hand-labeled set until each label clears the agreed accuracy bar. Accuracy is published next to every label.
Priced and mapped
The top failures priced, the value sheet built from your inputs, which decisions your data supports now, and the readout.
What it takes from your team
No IT project and no new systems. If you have no analyst support, I handle the account join from exports.
| Who | What | Hours |
|---|---|---|
| Executive sponsor | Kickoff and readout | ~2 |
| Help desk admin | Read-only export or API token | 1–2 |
| Security or legal | Data agreement review | varies |
| Finance partner | Sign off headcount, salaries, burden and next year's plan | 1–2 |
| One senior agent | Hand-label a small set, check label accuracy | 3–4 |
| Analyst (optional) | Account join | 2–4 |
What your data needs
- Full conversation text for at least 12 months, exportable read-only.
- A way to tie most conversations to a customer account: an account ID, a domain, or a contact record.
If either is missing, we start with a short Capture Sprint to fix it, credited in full toward the Baseline. You'll know on the first call.
Not included
Revenue, cost and people joins, the module analysis, dashboards and implementation. Those come in the modules, if and when you want them.
Talk it through: 30 minutes, no pitch.
Bring the decision on your desk. You'll leave knowing whether your data can answer it.