Annual workload
annual hours = annual units × minutes per unit × people per unit ÷ 60
+ rework hours
AI value measurement
How to establish an AI ROI baseline before implementation
Define one unit of work, observe a representative period, and record the old workflow before the new system changes behavior. Without that contract, almost any later ROI story can sound true.
Short answer: record volume, touch time, wait time, rework, loaded cost, quality, exceptions, and the downstream revenue or capacity constraint for one clearly defined unit of work. Name the measurement window and owner. Use the same definitions before and after implementation.
Baseline contract
Measure the operation from six angles.
| Measure | Definition | Example | Common mistake |
|---|---|---|---|
| Unit and volume | The repeatable case and completed/received count per period | Invoices processed per week | Mixing different case types |
| Touch time | Minutes people actively work on each unit | 18 minutes across two roles | Using elapsed time as labor |
| Wait time | Elapsed time between required steps | 1.6 days waiting for approval | Ignoring customer or revenue delay |
| Rework and exceptions | Share repeated, corrected, escalated, or rejected | 12% need a second pass | Counting only happy-path work |
| Loaded cost and quality | Labor cost plus defects, credits, leakage, or risk | $68/hour and 2.3% error rate | Calling capacity “cash savings” |
| Outcome constraint | The business result the workflow limits | Renewal reviews completed before risk window | Optimizing activity without outcome |
Formulas
Keep the math simple enough to audit.
Loaded labor cost
loaded cost = annual hours × fully loaded hourly cost
Before/after change
improvement % = (baseline − observed result) ÷ baseline × 100
Report capacity released, cash cost avoided, revenue accelerated, and risk reduced separately. They are different economic mechanisms. Capacity becomes cash savings only if spending falls; it becomes revenue only if the released constraint is redeployed and produces revenue.
Collection protocol
A baseline leadership can defend
Choose a representative window
Use enough cycles to include normal volume and meaningful exceptions. Record seasonality, backlog, launches, and policy changes.
Triangulate
Combine system timestamps, a sample of direct observation, and operator review. Any one source can misstate the real work.
Freeze definitions
Name the owner, formulas, exclusions, data source, review window, and acceptable quality floor before the pilot starts.