Most ERP AI projects fail because somebody picked the wrong first use case. These five questions take five minutes and will kill about half your shortlist — which is the point.
Every field exists and is entered reliably
Exists but inconsistently maintained
Lives in a private spreadsheet or isn't collected
Note: B1 customers pass this more often than they expect — structured, reconciled, current data is what most companies attempting AI don't have. Master-data hygiene (item groups, BP groups, lead times) is the usual weak point.
Annual cost comfortably below annual cost of the problem
Close
Not calculated
Note: Put a yearly value on the problem first; the tool must cost clearly less.
One sentence you'd type into chat
Three sentences
A paragraph
Note: If you cannot phrase it, the AI cannot answer it.
A wasted ten minutes
An awkward internal conversation
A shipped order, posted document, regulatory exposure or customer-visible error
Note: Your first four or five attempts should all score 2 here — suggest, don't act.
Metric named + today's value known
Metric named, value unknown
No metric
Note: No baseline, no proof. Write the number down before you start.
8–10 — start here. 5–7 — fix the weak criterion first. Below 5 — not this quarter.
If everything on your list scored eight or above, you have not been honest — a healthy shortlist has failures on it.
Name up to five use cases, pick 0 / 1 / 2 per question, and read the verdicts. Runs entirely in your browser — no data is sent anywhere.
| Use case | Q1 | Q2 | Q3 | Q4 | Q5 | Total | Verdict |
|---|---|---|---|---|---|---|---|
| — | — | ||||||
| — | — | ||||||
| — | — | ||||||
| — | — | ||||||
| — | — |
Want the printable version plus the 12 use-case tables and the cost model? Download the free workbook.
Ask. Diagnose against live data, read-only. Your first weeks are questions, not actions — receivables ageing, dead stock, margin leaks, answered from the database you already have.
Teach. Load the SOPs you already own — credit policy, reorder rules, discount ceilings, the approval matrix — so answers follow your rules, not generic ones.
Prove. The smallest usable prototype plus a measured before/after on the metric from Question 5. One page, real numbers, no faith required.
Automate. Scheduled delivery where people already are — WhatsApp, Teams, email, inside SAP Business One. The answer arrives; nobody has to remember to ask.
Scale. Skills compound and users expand. The second use case is cheaper than the first because the data hygiene, policies and habits already exist.
The free 30-Day Workbook has the printable scorecard, the exact B1 tables behind 12 use cases, and the build-vs-subscribe cost model.
Thinking on staged AI adoption sharpened by a Think School episode with Gayatri Agrawal; the framework, scorecard and SAP Business One specifics are TEKROI's own.