The tools usually work. The demos usually work. What is missing is a baseline that someone is willing to be measured against after the pilot ends.
What we see
A platform is chosen because a peer organization bought it. A pilot is marked successful because the exception path worked in a meeting. When someone later asks for the actual volume of exceptions from the last 60 days, the team needs weeks to reconstruct the number. By the time the baseline exists, the license has already been renewed and the original success criterion is no longer being tracked.
McKinsey’s 2026 State of AI survey found that 37% of respondents report any EBIT impact from AI, flat versus the prior year. Gartner reported that at least 50% of generative AI projects were abandoned after proof of concept by the end of 2025. S&P Global found that 42% of companies abandoned the majority of their AI initiatives before production.
What the ROI looks like when the number actually moves
These are anonymized patterns from technology enablement work where the baseline and owner were set before scaling:
Process cycle time
A claims or case-handling process reduced from 11 days to 7 days. At volume, the change freed capacity equivalent to 1.5–2 full-time roles without adding headcount.
Manual hours
Exception handling on a high-volume workflow dropped from 4.2 hours per 100 cases to 1.8 hours. Annualized, the reduction covered the cost of the automation within 7 months.
Error / rework rate
Data-entry or reconciliation errors fell from 6.4% to 2.1%. The downstream rework cost avoided was larger than the technology cost in the first year.
Cost-to-serve
On a specific product or channel, fully loaded cost-to-serve declined 12–18% once the manual steps were removed and the remaining exceptions were owned. The saving appeared in the operating budget, not only in a pilot report.
These outcomes required the metric, baseline, and owner to be clear before the work scaled. When those were missing, the same types of technology produced demos but no sustained change in the number.
Copy-paste checklist
Use this before any scale or renewal decision.
Technology Enablement Pre-Scale Check
Metric
What single number must move?
Current value (last 30-60 days):
Target value:
Data source:
Baseline owner
Who extracted the baseline?
Date baseline was confirmed:
Can it be reproduced in <10 working days? Yes / No
Outcome owner
Name:
Role:
Confirmed they will be measured for 90 days post go‑live? Yes / No
Date confirmed:
Review cadence
Monthly review scheduled for next 90 days? Yes / No
Attendees include outcome owner and funder? Yes / No
Decision: Proceed / Pause / Stop
Decision rules
Apply immediately.
- If item 1 is blank → Stop. Do not spend more money until the metric and target are written down.
- If item 2 cannot be completed in 10 working days → Pause. Assign one person and a deadline. No further investment until the baseline exists.
- If item 3 is blank or unconfirmed → Stop. Do not move to production.
- If item 4 is not scheduled → Pause. Put the three monthly reviews on the calendar before any scaling decision.
What to do in the next 5 working days
If a rule is triggered.
| Trigger | Action this week | Owner | Done when |
|---|---|---|---|
| No metric | 45-min meeting with process owner. Leave with metric + target written. | Process owner | Metric and target documented |
| No baseline | Identify the system/report. Assign one person. Deadline = 10 working days. | Analyst or process owner | Baseline number confirmed |
| No outcome owner | Name the person who runs the process. Get written confirmation they accept 90-day measurement. | Initiative sponsor | Confirmation email or note |
| No review cadence | Schedule three 30-min reviews (day 30, 60, 90). Include owner + funder. | Initiative sponsor | Calendar invites sent |
90-day test
After go-live, review the same metric on day 30, 60, and 90. The only question that matters: does the outcome owner still believe the number has moved? If the answer is no on day 90, stop scaling and return to the checklist.
Close
If your current plan cannot complete the checklist, the next step is not another tool. It is finishing the four items above. Most stalled efforts fail one of the first three.