ROI
How to tell whether automation saves time
Updated: 5 min read
“This saves a lot of time” begins as an expectation. Checking it requires a baseline and comparable measurements. Otherwise it is unclear whether work actually decreases or merely moves elsewhere.
Record what you will later compare
A baseline describes the current process. Record transaction volume, active handling time and overall duration. Use samples, timestamps or system logs where possible, and identify estimates as estimates.
A compact measurement sheet contains the baseline, target and later value for each metric. Add the data source, method, period, owner, assumptions and exclusions. Empty fields are unresolved measurement tasks, not zero values.
- Transaction volume and case types
- Active handling time per transaction
- Lead time between clearly defined start and end points
- Error, rework and exception rates
- Cost per transaction, including review and operations
- Quality, such as complete records or on-time processing
Handling time is not lead time
Handling time is active human work on a case. Lead time also includes waiting, questions, handovers and approvals. Automation may improve one measure without materially changing the other.
Fictional example: An invoice takes three days to approve but needs only twelve minutes of active work. Saving two minutes of data entry reduces handling time. If approval still waits two days, that bottleneck remains.
Assess speed alongside errors and exceptions
A fast output is not a success if checking it takes more time than before. Record rework even when it occurs later or in a different department. Also define what counts as a fully automated case and as an exception.
The following rates are calculation rules for your own measurement model. Numerators and denominators must refer to the same period and case population; no rate can be calculated when the transaction count is zero.
- Error rate = cases containing errors ÷ all cases × 100
- Rework rate = cases requiring rework ÷ all cases × 100
- Automation rate = fully automated cases ÷ all cases × 100
- Exception rate = exceptions requiring manual handling ÷ all cases × 100
Show calculated time savings transparently
Gross time saved in hours = transaction volume × (minutes before − minutes after) ÷ 60. Additional review, exception handling or operating effort must be recorded separately unless already included in the after measurement.
Fictional model calculation: 12,000 transactions per year, with eight minutes of active work before and three minutes after. 12,000 × five minutes equals 60,000 minutes or 1,000 hours of calculated gross savings. This is not a verified client result or a guaranteed outcome.
Check how many hours become usable in practice. Minutes distributed across many roles and days cannot automatically be treated as eliminated payroll costs.
Separate capacity, financial benefits and costs
Freed capacity may enable more transactions, less overtime, faster responses or other work. The actual effect must be observed separately. A costing rate does not automatically turn time into cash savings.
Include design, integration, data cleanup, testing, training, licences, maintenance, quality review and error handling. One-off investment and ongoing costs must neither be omitted nor counted twice.
For a simple model that explicitly does not discount future amounts: ROI = (total monetised benefit − total costs) ÷ total costs × 100. Benefits and costs must cover the same evaluation period, and total costs must exceed zero. Multi-year or complex decisions may require a more detailed economic assessment.
The Green Book is an appraisal framework for the UK public sector. Its structured consideration of options, benefits, costs and uncertainty can inform your method; it does not establish a particular business ROI.
Examine averages and comparison conditions
An average can conceal exceptions. In a fictional example, 80% of cases take two minutes and 20% take forty minutes. The average is 9.6 minutes, while the typical case still takes two minutes. Examine the median, distribution and slower case types too.
Compare the same process boundaries, case types and quality requirements over comparable periods. Record seasonal effects, onboarding, changed volumes and additional tasks. Useful measurement points are before the project, after the pilot stabilises and during ongoing operation.
Use measurable process changes as the criterion
Ask what changed in the complete process without silently reducing quality or control. This may mean less active work, shorter lead time, less rework or more predictable operations.
Keep uncertain assumptions separate from measurements. A small pilot can show whether the model is viable before you expand it. The existing ROI calculator helps explore your assumptions; it does not demonstrate actual savings.