Process improvement & AI
Process improvement with practical AI: from baseline to a 90-day plan
Where AI genuinely helps, where a simpler fix is better, and how to prove the result.
Illustrative example · fictional data
The situation
A fictional manufacturer wants to cut manual administrative work around quality and logistics. The question is where AI genuinely helps, and where a simpler fix is better.
Project phase
Baseline
Done
Prioritise
Done
Pilot
In progress
Embed
Open
Hand over
Open
Baseline versus target
Index values, baseline = 100. Targets are illustrative planning values, not a promise.
Manual handling time per batch
10055Rework from data-entry errors
10040Lead time for deviation reports
10060
Use cases by impact and effort
- AAutomatic intake of supplier documents
- BDraft weekly status reports from existing data
- CClassification of deviation reports
- DAssistant for questions on procedures
- EDemand forecasting for spare parts
- FFully automated invoice matching
Findings
1. High
Document intake is repeated at three steps
The same supplier data is typed in at receipt, at inspection and at invoicing. This is where most time and most errors sit.
2. Medium
Data lives in four spreadsheets without an owner
Any automation built on top inherits the inconsistencies. One owned data source comes first.
3. Medium
No baseline measurement exists yet
Without two weeks of measurement, improvements cannot be proven afterwards.
Action plan
Quick wins
- Measure the baseline for two weeks.
- Pilot automatic document intake with one supplier group.
Medium effort
- Consolidate the four spreadsheets into one owned data source.
Higher effort
- Decide on deviation classification once pilot results are measured.
Risk register
Likelihood →
Impact →
- 1
AI output trusted without review Score 12 (3×4)
Mitigation: Every AI output keeps a documented human check.
- 2
Pilot data not representative Score 9 (3×3)
Mitigation: Choose pilot suppliers across volume and document quality.
- 3
Users do not adopt the tool Score 9 (3×3)
Mitigation: Involve the people who do the work, and keep the old route open during the pilot.
- 4
Data quality too low Score 12 (4×3)
Mitigation: Fix the data source before any forecasting use case.
Recommendation
Start with the two low-effort, high-impact cases (A and B). Keep E out of scope until data quality is fixed. Every AI output keeps a human check.
What you receive
- Baseline measurement and target values
- Prioritised use-case list with a clear go / no-go per case
- 90-day plan with owners and checkpoints
- Documentation the owner can run independently
Have a project like this?
Tell me what you need assured, built or improved. I come back with a concrete proposal.
