Check process readiness, not only the team
Buying an AI service does not make a business AI-ready. If the process is different every time, data sits in five places, and decisions are made “by feel,” the model will only accelerate the mess.
The mistake I would remove first: Buying a tool before describing the process.
What to prepare and what result to expect
- Outcome: You know what can launch now and what first needs order in data or rules.
- Describe trigger, input, actions, result, exceptions, timeline, and owner. Then build a data registry with owners and access rights.
- Keep source data and access rights separate from the output so you can verify what business AI readiness work actually did.
- Score areas 0–3: data 30%, process 25%, security 20%, integrations 15%, people 10%. Readiness = sum((score / 3) × weight).
Build a data registry and the first safe pilot
- Describe one process
Do not run an abstract AI audit of the whole business. Choose a repeatable task with a clear input, result, and cost of error.
Проверьте: A newcomer can understand the process in five minutes.
Если не сработало: Break it into stages and pick one.
- Build a data registry
For CRM, spreadsheets, and email note owner, refresh, completeness, access, and personal data. Data in a CRM is not ready automatically.
Проверьте: You know which fields are complete and who will fix errors.
Если не сработало: Start with a test copy and do not move the whole archive.
- Check roles and training
Assign a process owner, data owner, exception handler, and feedback channel. Give the team three real can/cannot examples.
Проверьте: Employees know where to send a doubtful result.
Если не сработало: Write a one-page rule instead of a long presentation.
- Run a small pilot
Take real but reversible examples with manual review. Measure accuracy, fixes, time, and incidents before connecting a second channel.
Проверьте: There is a baseline and a decision date.
Если не сработало: Return to data or process instead of adding a new model.
A one-page readiness scale
Score one process 0–5: money impact 20%, repeatability 15%, data quality 15%, integrations 10%, observability 10%, risk control 15%, owner 10%, team 5%. Formula: `readiness = sum(score × weight) / 5`.
75–100 — ready for a pilot; 55–74 — fix data and controls first; below 55 — postpone automation. This is an internal scale: lock it before launch and do not fit it after a bad result.
readiness = sum(score * weight) / 5
if risk_is_high && approval_gate == false: block = trueWhat belongs in the process registry
Fields: `process_id`, `owner`, `trigger`, `input_source`, `systems`, `manual_steps`, `decision_rules`, `exception_types`, `monthly_volume`, `manual_time`, `error_rate`, `pii_present`, `approval_required`, `rollback_method`, `baseline_metric`, `target_metric`. Separately answer: what does a person do if AI is unsure. If there is no answer, the process is not ready.
What to do in one workday
In the morning ask an employee to show the last ten real cases, not retell the process. Midday collect sources, fields, exceptions, and owners. End of day take one case and manually walk the future workflow from input to result.
If you cannot name baseline, rollback, and the person for `needs_human`, do not buy a new tool. Fix those three gaps first, keep a test data copy, and only then recalculate readiness.
What must be ready before rollout
| Criterion | Question | Good sign |
|---|---|---|
| Input | What exactly enters business AI readiness? | Describe trigger, input, actions, result, exceptions, timeline, and owner. Then build a data registry with owners and access rights. |
| Action | What is the system allowed to do on its own? | Only pre-listed actions, with no access to the whole account |
| Verification | How do you know the result can be accepted? | Score areas 0–3: data 30%, process 25%, security 20%, integrations 15%, people 10%. Readiness = sum((score / 3) × weight). |
| Failure | Where does an unclear case go? | Pause the pilot if there is no data owner, permission to use data, or baseline metric. |
What should change after setup
You know what can launch now and what first needs order in data or rules.
Why subscriptions do not turn into adoption
Buying a tool before describing the process.
Treating data as ready only because it sits in a CRM.
Not assigning an owner for exceptions.
Running training without a real task and measurement.
When readiness needs a change program
Bring in a specialist if readiness spans multiple departments, access rights, regulated data, or a major process change.
What to check before the first task
Do you need perfect data?
No. But you need to know where errors are, who fixes them, and which fields cannot be used without permission.
Where do you start without an AI team?
With one repeating task with a clear result and a low cost of error.
What matters more: data or team?
Both. Good data without an owner and a team without a clear process do not produce a stable launch.
How often should readiness be revisited?
Before a new pilot and after a material change in process, data, or access rights.
