# AI Pilot Charter Template > Source: https://netray.co/tools/ai-pilot-charter-template > Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems. > Section: Free Tools ## Short answer A one-to-two page charter for an AI or agent pilot on your ERP: scope, success metrics, data access, risks, budget and stop rules. ## Category ai-automation ## Audience - CIO - IT manager - Project manager ## Tool **Kind:** template **Output:** document **Fields:** #### Pilot name **Key:** project **Type:** text **Default:** ERP Purchasing Assistant Pilot #### Company **Key:** company **Type:** text **Default:** Acme Precision Components Ltd #### Executive sponsor **Key:** sponsor **Type:** text **Default:** VP Operations #### Business owner **Key:** owner **Type:** text **Default:** Purchasing Manager #### Technical lead **Key:** lead **Type:** text **Default:** ERP Systems Lead #### Start date **Key:** start **Type:** date **Default:** 2026-11-09 #### End date **Key:** end **Type:** date **Default:** 2026-12-18 #### Problem statement **Key:** problem **Type:** textarea **Default:** Buyers spend about two hours a day exporting open PO lines and checking delivery dates across screens and spreadsheets, which delays supplier follow-up. #### In scope **Key:** scope **Type:** textarea **Default:** Read-only natural language questions on open purchase orders, receipts and supplier delivery performance for 12 buyers in one site. #### Out of scope **Key:** out_scope **Type:** textarea **Default:** Any write-back to ERP, supplier-facing communication sent without review, and other modules. #### Primary success metric **Key:** metric **Type:** text **Default:** Buyer time on PO status lookups reduced by at least 40 percent #### Accuracy threshold **Key:** accuracy **Type:** text **Default:** At least 90 percent of 100 test questions answered correctly with the supporting query shown #### Pilot budget **Key:** budget **Type:** text **Default:** $25,000 including infrastructure and effort #### Hosting **Key:** hosting **Type:** select **Options:** - On-premises GPU server - Private cloud - Vendor-hosted API - Hybrid **Default:** On-premises GPU server **Blocks:** #### {{project}}: Pilot Charter **Body:** Company: {{company}} Pilot period: {{start}} to {{end}} Executive sponsor: {{sponsor}} Business owner: {{owner}} Technical lead: {{lead}} #### 1. Problem and objective **Body:** {{problem}} Objective: prove, within the pilot period and budget, whether an AI assistant can deliver the primary success metric below without unacceptable risk, so that a go or no-go decision on wider rollout can be made on evidence. #### 2. Scope **Body:** In scope: {{scope}} Out of scope: {{out_scope}} Hosting: {{hosting}}. #### 3. Success criteria **Body:** Primary metric: {{metric}} Accuracy threshold: {{accuracy}} Additional measures: weekly active users among pilot participants, user satisfaction score, share of answers with a verifiable source, and average response time. #### 4. Data and access **Body:** Data sources, access method, ERP role mapping and any data excluded from the pilot are listed below. The pilot uses read-only credentials and respects ERP roles. No production data leaves the approved environment. **Table:** **Columns:** - Source - Access method - Owner - Restrictions **Rows:** **0:** ERP purchasing tables **1:** Read-only service account **2:** ERP lead **3:** No supplier banking data **0:** Supplier delivery history **1:** Reporting view **2:** Purchasing manager #### 5. Risks and mitigations **Table:** **Columns:** - Risk - Likelihood - Impact - Mitigation **Rows:** **0:** Incorrect answers trusted by users **1:** Medium **2:** High **3:** Show source query, test set, user training **0:** Data exposure beyond role **1:** Low **2:** High **3:** ERP role trimming, access test before launch **0:** Low adoption **1:** Medium **2:** Medium **3:** Weekly feedback, champion buyer **0:** Cost overrun **1:** Low **2:** Medium **3:** Fixed budget, weekly spend review #### 6. Milestones **Table:** **Columns:** - Milestone - Target date - Owner **Rows:** **0:** Kickoff and access granted **2:** {{lead}} **0:** Test question set approved **2:** {{owner}} **0:** First users onboarded **2:** {{lead}} **0:** Mid-pilot review **2:** {{sponsor}} **0:** Final evaluation and decision **1:** {{end}} **2:** {{sponsor}} #### 7. Budget and resources **Body:** Approved budget: {{budget}}. Resource commitments from the business owner, the ERP team and security are agreed before kickoff. Spend is reviewed weekly against the budget. #### 8. Stop and exit rules **Body:** The pilot stops early if a data exposure incident occurs, if accuracy remains below half the threshold after two tuning cycles, or if the budget is exhausted. At the end, the decision options are: stop, extend with changed scope, or move to production planning. #### Approvals **Signatures:** - Executive sponsor - Business owner - Technical lead - Information security **Notes:** Edit the sample risks, data sources and milestones to match your pilot. Keep scope narrow, set the accuracy test before the build starts, and record the baseline for your success metric before the first user logs in. ## Content **Intro:** AI pilots drift when nobody wrote down what success means. A two-page charter fixes scope, evidence and stop rules before the first prompt is written. **Sections:** #### How to use the charter **Body:** Complete the fields with the sponsor and business owner in the same meeting, then export to Word. Be strict with the in-scope statement: one team, one site, one question type. Write the success metric with a number and a baseline, such as minutes per lookup measured before the pilot. Set the accuracy test before development, using real questions from users and answers checked by the business owner. #### What a good charter includes **Body:** A useful pilot charter has eight parts: problem, scope, success criteria, data and access, risks, milestones, budget and exit rules. The data section matters most in ERP work because it states which tables or documents the assistant sees and which role it runs as. The exit rules prevent the common failure of a pilot that never ends and never decides. Name who makes the go or no-go decision and by what date. **Bullets:** - One primary metric, plus two or three supporting measures - Baseline measured before the pilot starts - Read-only access unless a specific write step is approved - A defined stop rule for data exposure and for poor accuracy #### Running the pilot **Body:** Hold a weekly review of usage, accuracy on the test set and cost. Collect failed questions, since they show whether the problem is data, retrieval or the model. Resist adding features mid-pilot. At the end, compare results with the charter and not with the demo: if the metric was missed, decide whether the cause is fixable within a changed scope, or whether the use case should be dropped. #### Common mistakes **Body:** The most frequent mistakes are unmeasured baselines, a scope that grows every week, and a sponsor who only sees demos. Another is choosing success measures that cannot fail, such as user enthusiasm. Finally, plan production early: security review, hosting, support ownership and licensing differ from a pilot, and discovering them after a successful pilot wastes the momentum. **Cta:** If you want help running a read-only pilot against your ERP with the evidence the charter asks for, we can scope one with you. **Faq:** #### How long should an AI pilot last? **Answer:** Four to eight weeks is typical. Shorter pilots rarely collect enough usage data, and longer ones tend to expand in scope. Fix the end date in the charter and decide at that date. #### Who signs the charter? **Answer:** The executive sponsor, the business owner, the technical lead and information security. If the pilot touches regulated or export-controlled data, add compliance or the export control officer. #### What accuracy should we require? **Answer:** It depends on the use case. For read-only lookups with the source shown, 90 percent on a real test set is a common bar. Anything that influences product conformity or financial postings needs a higher bar and human review. #### Should the pilot use real data? **Answer:** Usually yes, since synthetic data hides the messy parts of an ERP. Use read-only access, role trimming and an approved environment, and exclude data classes the policy restricts. ## Search terms ai pilot charter template, ai proof of concept template, ai project charter word, ai pilot plan template, poc success criteria template, llm pilot project template, ai pilot scope document, ai project kickoff template --- Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json