Direct answer
Measure AI value by comparing an agreed pre-change baseline with accepted post-change performance across cash, capacity, growth, customer value, quality and risk. Separate gross capacity from realised benefit, include the full operating cost and report assumptions and limitations with every result.
A value claim earns confidence in stages
Estimated activity becomes value only after adoption, cost and controls are included.
Build a value tree from the management result
Start with one business result, then identify the drivers the implementation can influence. For example, sales productivity may depend on enquiry response, research time, proposal quality, follow-up and management visibility. The AI component is one intervention inside that operating chain.
Every value case should identify a primary measure, supporting drivers and guardrails. This prevents a faster step from being celebrated when quality, customer trust or downstream workload deteriorates.
Establish the baseline before changing the workflow
- Define the unit of work and the period being compared.
- Measure volume, active effort, elapsed time, quality, exceptions and cost.
- Record the current owner, system path and service or commercial outcome.
- Use a representative sample and preserve evidence links.
- State known gaps instead of converting assumptions into facts.
Use measures that match the value claim
| Value lens | Useful measures | Common mistake |
|---|---|---|
| Cash and cost | Avoided spend, margin, working capital, realised labour cost | Treating all saved minutes as cash |
| Capacity | Active effort, throughput, cycle time, backlog, use of recovered time | Reporting theoretical hours without adoption |
| Growth | Contact, qualified next step, conversion, contribution and sales cycle | Attributing all revenue movement to the tool |
| Customer value | Retention, return rate, resolution, service effort and complaints | Optimising volume while trust declines |
| Risk and quality | Error rate, severity, rework, exceptions, incidents and control coverage | Measuring speed without accuracy |
Calculate the full operating economics
Include discovery, redesign, implementation, integration, licences, model usage, monitoring, review, support, change management and ongoing evaluation. Deduct these from realised benefit, not from a best-case estimate.
Make a stop, improve or scale decision
- Stop
When quality, control, adoption or economics do not meet the agreed minimum.
- Improve
When the value is plausible but one workflow, data or operating constraint remains unresolved.
- Scale
When the result is accepted, controls work under real conditions and the next unit of investment has a credible value case.
Executive reporting should show the baseline, current result, confidence level, full cost, key risks, limitations and next decision. A single ROI percentage is not enough.
Sources and guidance
- McKinsey: How AI creates value in private equity
- Microsoft Work Trend Index: Human-agent teams
- Australian Government: Guidance for AI adoption