How we work: AI judged on the net saving, hours saved minus checking time

Fig. 01Gross → netillustrative numbers
Hours a year
Claimed, not checked People check Odd cases Net
Dollars a year
Claimed, not checked Checking and odd cases Running cost Net
A back-office team handles 12,000 cases a year. What does the system really save?
The pitch: the system takes over all 2,000 hours of this work. $90,000 a year. Net of checking and running costs, it saves 1,580 hours and $59,100 a year.
tap a number or a bar
Fig. 02What you can count on

Small steps, all written down. The know-how stays yours.

Price. Engagements start at $5,000. Research often costs $5,000 to $10,000. You know the price before we start. One service at a time. Nothing is bundled.

Risk. You can stop after any step. Each step ends with something you can use and a clear choice about what comes next.

Ownership. Simple builds are yours outright. For software you own → No monthly fee for software you own. Simple builds are yours outright, source code included. Builds on our existing systems come with a license or as a service.

Running cost. You see what it costs to run before you buy. Hosting, model usage and support, per year, next to the saving. It's already inside the net number above.

Data. Your data is never used to help a competitor. What we build on your data stays your advantage. We design for California privacy law (CCPA) and for GDPR.

Errors. Missing data is shown as missing. Method → The case goes to a person, with the gap marked. Every decision is logged with what the person saw when they made it.

Fig. 03What we measure

Seven numbers, and how each one is measured.

Intervention rate. Share of cases that needed a person. Counted from the log, not estimated.

Exception rate. Share of cases that don't fit the rules at all. They are done by hand and counted as work.

Review time. Seconds a person spends on each pre-filled answer. Timed in a pilot, before any savings are projected.

Errors introduced. Mistakes the system made, caught at review or found later in a sample check. Shown next to the time saved, never left out.

Cost per correct case. Total cost divided by cases completed correctly. People's time plus running cost.

Throughput. Cases completed per day, with the same team. Compared with the weeks before the system went live.

Decision quality. Whether the decisions made with the system hold up later. Checked on a sample, by a person who didn't make them.

Bring your own numbers. We'll work out the net together.