Interviews Suggested About 90% Less Time. More Capacity?

The subject was monthly internet-research work at an anonymized company. Roughly two or three people were involved depending on the month. After an RPA workflow incorporating AI was introduced, retrospective interviews suggested that time spent on the target work had fallen by about 90%. That is large enough to investigate further, but not evidence by itself for approving another investment.

A follow-up described as occurring two to three months later attributed only about 10–20% of the estimated freed time to new work or additional production. The available material does not establish whether that interval began at introduction, operational start, or stable operation.

The right response is neither to dismiss the 90% nor to label the remainder waste. The time-reduction hypothesis warrants further measurement, but it is not evidence of success or a basis for scaling until quality and hidden work are checked.

For a department head or transformation owner, the decision is not merely whether the automation ran. It is whether the conditions for copying the design elsewhere have been met. Because quality and hidden work remain unconfirmed, continuation, scaling, workforce-plan changes, and recognition of financial benefits all require evidence beyond these figures.

The roughly 90% figure is a retrospective estimate of how much time spent on the target workflow fell relative to pre-introduction time. The 10–20% figure uses the estimated freed time as its base and describes the share that follow-up interviews could attribute to new work or additional production. One concerns process time and the other freed time, so placing them on a common bar chart would falsely imply comparable success rates.

The two or three people are the number involved in the work, not necessarily the number interviewed or a full-time-equivalent measure. The material available does not establish the baseline month, workflow boundary, or case volume behind the 90%, nor whether the 10–20% range reflects respondent differences, monthly variation, or rough rounding. We therefore do not multiply the percentages to claim that a share of original time became value.

The defensible statement is narrower: retrospective interviews reported a large reduction in target-workflow time, while the follow-up attributed only a limited share to new uses after two or three months. The case does not establish the causal effect of AI-enabled RPA, a productivity gain, a profit rate, or a rate that will recur elsewhere.

• About 90%: interview estimate of the decline relative to pre-introduction time spent on the target workflow
• 10–20%: share of estimated freed time that follow-up could attribute to new work or additional production
• Unconfirmed: baseline period, case volume, quality, respondent count, and time for review, correction, exceptions, and maintenance

The remainder that follow-up interviews did not attribute to new work or additional production is not necessarily vanished, idle, or worthless time. Minutes may have been scattered through the day and never formed a block large enough for another task. There may also have been insufficient demand, assignment authority, skill, budget, or downstream capacity.

The capacity may have supported other forms of value. Peak-load resilience, deadline stability, avoided overtime, rest, higher quality in existing work, clearing backlogs, and readiness for unexpected requests may not have counted as new work or additional production. Conversely, review and correction of AI output, exception handling, RPA maintenance, incident recovery, or search-condition updates may have shifted to another person and inflated the apparent reduction.

It is therefore premature to blame individual adoption or management's failure to reassign work. The 10–20% is only the approximate range attributed in this follow-up and may overstate or understate actual reallocation because of recall error and uncertain calculation boundaries. A missing explanation for the remainder should not be replaced with an explanation of failure.

First, add review and correction of AI output, exception handling, reruns, and RPA maintenance back to post-introduction human time. These remain part of post-introduction human work. Compare before and after at the same workload and quality standard; net freed time is the difference remaining after these inputs are included. Trace its destination only after that comparison.

Separate the destination of net freed time into three categories: time used for new work or additional production, another use supported by records, and time whose destination remains unknown. Record whether the time formed an assignable block or appeared in fragments as a separate dimension, because fragmented time can still have a known or unknown destination.

The purpose is not to fill the remainder with guesses. It is to assign the next observation for the monthly cycle. Known destinations need outputs or records, time shape needs duration and timing, and an unknown destination needs another question or log and a responsible owner.

• Net freed time: Add review, correction, exceptions, reruns, and maintenance back to post-introduction human time
• Destination: Separate new work, other uses supported by records, and unknown destinations
• Time shape: Record assignable blocks and fragmented time in a separate field
• Unknown destination: Leave it unknown and assign the next question, log, and owner

As a Catalysphere starting example, use different questions immediately after introduction and several months later. During roughly the first 30 days, hold the workflow boundary constant and record case volume, outputs, time spent, accuracy, coverage, rework, and timeliness. Link review and correction of AI output, reruns, exception handling, RPA maintenance, incident recovery, and search-condition updates to the same workflow ID as hidden work that offsets saved time.

At roughly 90 days, trace the destination of freed time. Whose hours moved to which work, output, and volume? If their destination cannot be confirmed, record whether demand, authority, skills, time fragmentation, downstream constraints, resilience use, or measurement gaps explain the difference. Thirty and 90 days are starting points suggested for this case, not universal observation windows; set them in advance to fit the work cycle and time to outcome.

Keep interviews because they reveal perceived burden and uses of time. But retain the questions, respondents, dates, and calculation method, then reconcile them where possible with work logs, case management, and output records. The purpose is not to dismiss memory; it is to keep estimates and continuous measurement in separate fields.

The capacity-tracking record starts with a workflow ID, baseline period, case volume, before-and-after time, hidden work, and net freed time. It then records the destination of that time and, separately, whether it formed usable blocks or appeared in fragments. Add the measurement basis, confidence, owner, 30-day and 90-day judgments, next review date, a quality boundary, and separate fields for continue, scale, narrow, or stop. Never blend estimates, logged values, and unknowns.

The record does not automate a decision; it separates information for different decisions. Process time, quality, and hidden work inform continuation. Destination, demand, and outputs inform scaling. Quality breaches, exceptions, and maintenance inform a scope change. Hold scaling when no destination can be supported by records. The relevant owners must approve the boundaries for continuing, scaling, narrowing, or stopping according to the workflow's impact.

The tracking record is not meant to create a precise-looking composite score. It keeps unknown quality, hidden work, and capacity destinations next to the attractive 90% figure, then assigns the observations needed for the next decision.

• Value: Keep volume, time, quality, outcomes, and spending in separate fields
• Method: Distinguish interviews, work logs, case records, and financial records
• Confidence: Mark estimates, confirmed values, and unknowns explicitly
• Owner: Do not collapse measurement, improvement, and approval into one ambiguous responsibility
• Next review: Reassess process stability and the destination of freed time on different dates

This case does not prove that AI-enabled RPA failed to generate profit. Nor does the estimated 90% reduction establish a 90% productivity gain or profit. It shows a gap in observation between reduced workflow time and another outcome; whether a destination for capacity had been designed was also not confirmed.

For the next automation, define quality, hidden work, reassignable time blocks, destination demand, outputs, and the next review date before introduction. As a starting example, check the predefined process and quality boundaries at 30 days, then use only the capacity destinations that can be explained at 90 days as evidence for the next decision. Treat unexplained capacity as neither failure nor profit, but as a target for the next measurement.

For this case, the interview estimate alone does not justify scaling, changing the workforce plan, or recognizing a financial benefit. Approve only the next measurement phase: calculate net freed time at equal workload and quality, then trace its destination. Decide whether to scale only after those records are available.