Search

What are you looking for?

Search our services, use cases and practical insights.

Enter at least 2 characters

Popular starting points

AI decisions

How do you measure the value and ROI of an AI project?

Time saved in a demo is not yet business value. Impact appears only when people use the solution in the real process, outputs are good enough and released capacity is put to productive use.

The short answer

The ROI of an AI project compares measurable additional value against a documented baseline with the full investment and operating cost. Financial effects should be reported alongside time, quality, risk, adoption and effects on employees.

In brief

  • Before a pilot, establish a credible baseline for effort, quality, volume and current cost.
  • Value may come from time saved, better quality, avoided errors, increased capacity or additional revenue.
  • Theoretical minutes saved become valuable only when adoption, quality and actual process change are demonstrated.
  • ROI should be reported as a range with transparent assumptions and alongside non-financial outcomes.

Which baseline should be measured first?

Without a comparison point, every improvement remains an assertion. Before launch, capture case volumes, processing time by case type, errors, rework and waiting time. Seasonal variation and differences between teams also belong in the baseline.

  • Volume and share of different case types
  • Active handling time, elapsed time and waiting time
  • Error rate, questions, rework and escalations
  • Current people, system and opportunity costs
  • User and customer experience and relevant risk events

Which kinds of value should be assessed separately?

Time saving is only one effect. A solution may produce more complete outputs, absorb peaks or improve access to expertise. Benefits should not be converted hastily into money when the causal link is not evidenced.

  • Efficiency: less active work or shorter elapsed time
  • Quality: fewer errors, greater completeness or consistency
  • Capacity: more cases handled with existing resources
  • Risk: avoided incidents, better traceability or earlier detection
  • Revenue and service: new offerings, faster response or higher conversion

How does observed impact become a defensible calculation?

The conventional formula is financial benefit minus total cost, divided by total cost. The difficult part is not the arithmetic but attribution. An observed change may have resulted from another process or demand shift.

  1. Step 1

    Define the comparison

    Use before-and-after data, a control group or phased introduction so other effects remain visible.

  2. Step 2

    Account for adoption

    Count only outputs that are actually used and accepted in realised value.

  3. Step 3

    Subtract fully loaded cost

    Include development, licences, operations, review, training, rework and change effort.

  4. Step 4

    Present a range

    Report conservative, expected and favourable scenarios with their assumptions.

How does adoption change the result?

A service process handles 50,000 cases per year. Each draft that is both used and quality-compliant releases six minutes, with labour valued at CHF 75 per hour. The theoretical maximum benefit is therefore CHF 375,000. Only the share that is adopted and passes review becomes realised value. ROI = (realised benefit minus fully loaded cost) / fully loaded cost.

How does adoption change the result?
ScenarioAdoptionUsable shareRealised annual benefitFully loaded costROI
Conservative45 %65 % of used draftsCHF 109,688CHF 130,000−16 %
Expected70 %80 % of used draftsCHF 210,000CHF 150,000+40 %
Optimistic85 %90 % of used draftsCHF 286,875CHF 175,000+64 %

How does value remain visible after the pilot?

A successful pilot does not guarantee impact at scale. Usage may fall, the case mix and model cost may change, or review may consume the released time. Owners should therefore manage impact and cost together as product metrics.

  • Observe usage and abandonment by team and case type
  • Compare acceptance, correction and failure rates with the baseline
  • Measure cost per successful outcome, not only total spend
  • Revisit benefit assumptions after three, six and twelve months
Example from day-to-day business

Example: drafting service responses

Before the pilot, a standard case takes twelve minutes on average, seven of which are spent on research and drafting. In a phased pilot, active time falls by three minutes with the same quality review, and 70 per cent of drafts are used. The company counts only realised time on those cases, subtracts model, platform, review and support costs, and also monitors errors and customer satisfaction. The reported value remains traceable.

What to remember

Define the baseline, value logic and fully loaded cost before the pilot. Report realised impact, adoption and uncertainty transparently instead of deriving theoretical ROI from demo timings.

Sources and further reading

These primary sources provide further detail on definitions, technical foundations or responsible use.

Content reviewed

Would you like to apply this to your situation?

Together, we clarify what makes sense for your process, data and systems – in plain language and without unnecessary complexity.

Discuss Your Project