How to Write an RFP for ERP AI
how to write an RFP for ERP AI
Also searched as
- ERP RFP AI requirements how to write
- how to include AI in an ERP RFP
- writing an RFP for AI on top of our ERP
- ERP AI request for proposal structure
Short answer
Write the ERP AI RFP as two linked scopes: the ERP platform and the AI layer that reads or acts on its data. Define the business questions the AI must answer, where data may and may not travel, which ERP roles apply, and how you will test answers before you sign. Vague asks like 'must have AI' produce slideware, not comparable bids.
Applies to: Any ERP selection or AI add-on procurement (SAP, Oracle, NetSuite, Infor, Dynamics, Epicor and others), mid-size and enterprise manufacturers
How to structure an ERP AI RFP
- 1Start from the problem list, not the technology: write 8 to 15 concrete questions or tasks users cannot do today, for example 'which open POs will make a promised ship date late' or 'draft the nonconformance report from this inspection record'.
- 2Split the document into Part A (ERP platform, modules, integrations, implementation) and Part B (AI capabilities, models, deployment, data handling). Bidders can then respond to one or both and you can score them separately.
- 3State deployment constraints in Part B as hard gates: on-prem, private cloud, air-gapped, data residency, export-controlled data. Mark them pass/fail so non-compliant bids are removed early.
- 4Require every bidder to state which ERP data the AI can read, by what interface (API, database view, replicated store), and whether it enforces the user's existing ERP roles and row-level security.
- 5Define the write-back policy: read-only by default, and any create or update action requires a named human approval step with an audit entry.
- 6Add a proof-of-value requirement: shortlisted vendors answer your 10 test questions on a sanitized copy of your data, and you grade the answers against the ERP.
- 7Request pricing in a fixed template: licence or subscription, inference or token charges, GPU or hosting assumptions, implementation, support, and year 2 to 5 escalators.
- 8Publish the scoring model and timeline in the RFP itself so bidders know what you value.
Why AI breaks the old RFP template
Classic ERP RFPs are long functional requirement lists: general ledger, MRP, shop floor, quality. Vendors answer 'standard', 'configurable' or 'custom' against each row and the winner is whoever scores the most green cells. AI does not fit that format, because the capability is not a feature you can tick. It depends on your data quality, your permissions model, the model that runs it and where it runs.
Two bidders can both answer 'yes, natural language query' and deliver very different products: one grounded in your live tables with the generated query shown, the other a chat window over a PDF export. The RFP has to force that difference into the open by asking how, not whether.
The practical fix is to keep your functional requirement matrix for the ERP and add a separate AI section built on scenarios, constraints and tests. The /tools/erp-rfp-completeness-checklist tool on this site is a quick way to confirm the ERP half is not missing sections before you add the AI half.
Scenarios beat feature lists
For each AI scenario, write the user, the question, the data involved and the acceptable answer. For example: a planner asks which purchase orders from suppliers with an on-time rate under 85 percent are due in the next 10 days. The acceptance test is that the answer matches a query a power user runs by hand in the ERP.
Mix retrieval scenarios (read data), document scenarios (answer from controlled procedures and drawings), and action scenarios (draft a requisition, propose a reschedule). Keep action scenarios few and explicit about approval.
Keep a held-back set of 5 questions you do not share. Vendors that tuned only to the published scenarios will show it on the hidden ones during the proof of value.
Hard constraints to put in writing
Data location is the most common reason an AI bid fails late in the process. State plainly whether prompts, retrieved ERP rows, embeddings and logs may leave your network or your tenant, and whether the model may be trained on your data. Ask for the answer per component, not per product.
If you handle export-controlled technical data or defence work, say so, and require the bidder to describe how the architecture lets you meet your own obligations. Do not accept a compliance badge as a substitute for an architecture description.
The /tools/ai-vendor-evaluation-checklist and /tools/ai-agent-security-review-checklist tools cover many of the questions worth copying into this section.
Commercial and exit terms
AI adds usage-based cost that classic ERP licences did not have. Ask for the pricing unit (per user, per query, per token, per GPU) and a worked cost for your expected volume, plus what happens at 3x volume.
Require exit terms: export of prompts, configurations, embeddings and fine-tuned weights where applicable, and a statement of which parts depend on a proprietary model you cannot move. This is the AI version of lock-in review, and the /tools/ai-contract-negotiation-checklist tool is a useful companion when you reach contract.
Common pitfalls
- !Writing 'must have generative AI' as a mandatory requirement. Every bidder will answer yes and you learn nothing.
- !Letting the ERP vendor's bundled assistant define your AI scope when your real questions span several systems or documents.
- !Omitting data residency and model training terms, then discovering them in contract redlines.
- !Scoring AI on demo polish. Demos run on curated data; only a test on your data is evidence.
- !Forgetting usage-based AI costs in the five-year total cost comparison.
- !Making write-back actions part of day one scope before read-only answers are proven.
How an ERP-grounded AI assistant handles this
An AI assistant grounded in your ERP can help you write the RFP itself: ERPray-style tooling over live ERP data can list the reports, queries and custom screens users actually run, which become the scenario list for Part B. It is a way to draft requirements from real usage rather than guesses. Treat the output as a draft for your team to review, not a finished specification.
Frequently asked questions
Should AI be a separate RFP from the ERP?
Often yes, or at least a separate scored section. If you are replacing the ERP, put AI in Part B so ERP fit is not distorted by AI marketing. If you are keeping your ERP, run an AI-only RFP against the existing system and its data.
How long should an ERP AI RFP be?
Part B can be short, 10 to 20 pages, if it is scenario based. Length does not add rigor. A clear scenario list, hard constraints, a pricing template and a proof-of-value plan matter more than hundreds of feature rows.
Do we need a proof of value before selecting?
For anything that reads production data, yes. Ask shortlisted vendors to answer your test questions on a sanitized extract and grade the answers against known ERP results. It costs a few weeks and removes most of the demo risk.
Can we reuse a generic AI RFP template?
Use one for structure, but replace its generic questions with your scenarios and your ERP's actual roles, tables and integration paths. Generic templates miss the data access and permission questions that decide ERP AI projects.
Related
ERP Selection in 2026: Where AI Actually Matters
Treat AI as one evaluation column among many, not the deciding factor: fit, data model, industry depth, and total cost of ownership still decide most ERP selections. Test AI claims live against your own data during the demo, not the vendor's canned dataset, and separate "embedded copilot" marketing from features that ship and work today.
CIO briefingERP AI Vendor Due Diligence: What CIOs Should Actually Check
The single highest-signal question in ERP AI vendor due diligence is 'show me it answering a real question against our actual ERP data, live, not a demo dataset.' Beyond that, focus diligence on data grounding method, security and access model, exit/portability terms, and references from customers on your specific ERP platform - not on feature checklists, which most vendors can match on paper.
How-toERP RFP Template with AI Requirements
Add six AI requirement blocks to your ERP RFP: data access and grounding, security and roles, deployment and data residency, model governance, user experience and explainability, and commercial terms. Each block below has requirement wording you can paste and a way to mark it mandatory, weighted or informational.
CIO briefingQuestions to Ask AI Vendors in an ERP RFP
Ask AI vendors how, not whether: how the assistant reaches ERP data, whose permissions it uses, where each model runs, how answers are tested, what it costs at volume and what you can take with you on exit. Below are the questions grouped by topic, with what a good and a worrying answer sounds like.
CIO briefingERP Upgrade or AI First? A CIO Decision Framework
In most cases, add a grounded AI layer on top of your current ERP first, because it proves value in weeks and shows exactly what an upgrade would need to fix. Reserve a full ERP replacement for cases where the platform itself is end of support, unsupported, or structurally blocking the business, not simply because it feels dated.
CIO briefingHow Much Does ERP AI Actually Cost?
A scoped pilot against one ERP module typically runs 15,000 to 50,000 USD; a single-department production deployment runs 50,000 to 150,000; a multi-module enterprise rollout runs 150,000 to 500,000 or more, plus ongoing hosting and usage costs. Native vendor copilots are usually priced per active user per month on top of existing licensing, not included free.
AI for ERPBuild vs Buy: Should You Use Your ERP Vendor's AI Copilot, or Build Your Own?
SAP Joule, Copilot for D365, Infor GenAI, Oracle AI Agent Studio, or a private LLM on your own data. A CIO framework for the build vs buy decision, with real trade-offs.
AI for ERPA Private LLM Grounded on Your ERP Data
How a private LLM answers questions on your ERP data: RAG plus text-to-SQL, role-based permissions inherited from the ERP, and where each fits.
Stuck on ERP selection and AI?
Talk to engineers who work inside ERP selection and AI every week, and who build private AI that answers these questions from your own ERP data.