Phase 1: Process & Scope (Points 1–5)
-
Process boundary defined
Start and end points of the process are documented. You know exactly what goes in and what must come out.
-
Data owner identified
One named person is responsible for the data the AI will use. They have the authority to grant access.
-
Volume / frequency measured
You know how many documents, requests, or cases flow through per day/week. This determines ROI and model cost.
-
Manual steps inventoried
Every manual step in the process is listed. AI can only replace steps you have mapped first.
-
Baseline KPI exists
You can measure the current state (time, error rate, cost per unit). Without a baseline, pilot success cannot be proven.
Phase 2: Data & Compliance (Points 6–9)
-
PII / GDPR scope mapped
You know whether personal data (names, health records, financial data) flows through this process. If yes, a DPIA may be required.
-
Local LLM option assessed
If PII or trade secrets are present, you have evaluated whether a locally hosted model (Ollama, vLLM, Mistral local) is needed to keep data inside your perimeter.
-
Data quality checked
Source documents are in a consistent format, language, and quality. Garbage in, garbage out: clean the data before you build.
-
Integration points listed
You know which systems (CRM, ERP, email, Telegram, etc.) the AI pilot must read from or write to, and API access is confirmed.
Phase 3: Pilot Design (Points 10–13)
-
Pilot success criteria set
You have written down: "The pilot succeeds if X drops by Y% within Z weeks." The decision-maker has agreed to it.
-
Cost ceiling approved
LLM token cost, compute cost, and implementation cost have a budget cap signed off before work starts.
-
Rollback plan documented
If the AI pilot is switched off tomorrow, you know how to revert to the manual process without data loss.
-
Responsible person named
One internal person owns the AI pilot in your organisation. They attend every review call and escalate blockers.
Phase 4: Validation & Go/No-Go (Points 14–15)
-
User acceptance tested
At least 3 real users from the target team have tested the pilot output and given written feedback. Edge cases are documented.
-
Go/No-Go criteria explicit
You know in advance which results would cause you to stop the project. Decision criteria are written, not verbal.
Standard check
Risk area: check thoroughly
Blocker: resolve before you proceed
Not sure how your process scores?
We go through this checklist with you in the first 30-minute discovery call and give you an honest assessment: ready for a pilot, needs preparation, or not the right time for AI.
Request a free 30-min call