← All posts
Industry Playbooks ·

How to Evaluate AI Platforms for Industrial Distributors (2026)

Every AI platform demo looks good on the vendor's data, so how do you pick one for an industrial distributor? Run each platform on a sample of your own RFQs and purchase orders, and count what it finishes inside your ERP. InstaLILY built Lily™, its AI forward deployed engineer, to pass that test. She is software: she learns your pricing and approval rules from your own quote history, writes quotes and orders back into Epicor Prophet 21, SAP or Oracle NetSuite with contract pricing, units of measure and substitutions correct, and sends anything she is unsure of to a named person. Most deployments start read-only until your IT team approves write access.

Last updated: 2026-11-06

This guide sets out each test, the question to ask the vendor, and what a good answer looks like. Under each test, a short section shows how Lily handles it, as one worked example.

What should an industrial distributor test before buying an AI platform?

Test what the platform completes in your systems, on your data, under your rules. The checklist below has one row per test, and each test has its own section further down.

What to test Question to ask the vendor What a good answer looks like
It finishes the work Which transactions does it write into our ERP, and which does it only suggest? Named transactions, such as quotes and sales orders, written to the system of record
Read and write scope Which records does it read, which does it write, and which does it leave alone? A written list for your ERP and version, starting read-only
Distribution accuracy How does it score on our messiest RFQs, line by line? Line-level results on your own requests: part numbers, units of measure, pack sizes, substitutions and contract pricing
Uncertainty and approvals What happens when it is not sure? Uncertain or high-risk work goes to a named person before it executes, and every action is logged
Pilot and measurement What will we measure, and against what baseline? One workflow, a baseline recorded before the pilot, and completed work as the measure
Data and ownership Where is our data processed, and does it train anyone else's model? A single-tenant environment, no cross-customer training, and a written statement of what stays yours
Build or buy Who maintains the integration and the exception handling after go-live? A named owner for upkeep, in writing
Your ERP vendor's AI Does it cover the workflow across our inbox, CRM and ERP? A clear line between what the ERP vendor's AI covers and what it leaves to people
Total cost What does one completed workflow cost us, including integration and data work? One number for completed work, with integration and data work included

 

Does it finish the work inside your ERP, or stop at a recommendation?

The test is whether the platform writes a real transaction, such as a quote or a sales order, into your system of record. A suggestion that a person still has to key in saves reading time, not work. Ask the vendor to show a transaction created in a test copy of your ERP. A screenshot of a recommendation is not the same thing.

How Lily handles this. Lily writes quotes, sales orders and the quote PDF to the ERP. In the CRM, she writes notes, tasks and follow-up status. The work is done when the order is in the system, not when someone has been told about it.

Which ERP records can it read, and which can it write?

Ask for the exact read and write workflow on your ERP and version, and for the records it leaves alone. A connector logo tells you nothing about whether it can create a sales order in your setup. Get the answer in writing, and check it with the person who owns ERP access.

How Lily handles this. Lily works with Epicor Prophet 21, SAP, Oracle NetSuite, Infor and Oracle JD Edwards, with Salesforce and HubSpot on the CRM side.

  • Reads: item master, inventory by branch, price lists, contract pricing, quantity breaks, customer and ship-to records, and quote, order and invoice history.
  • Writes: quotes, sales orders and the quote PDF, plus CRM notes, tasks and follow-up status.
  • Leaves alone: your master data, your pricing rules and your approval chain.

On SAP ECC and S/4HANA, she connects through SAP's standard interfaces, OData and RFC/BAPI, so creating a sales order needs no custom ABAP development. She works in Outlook email and calendar through the Microsoft Graph API. Where a system has no API, she can work through its screens. Most deployments start read-only, and write access comes later through a scoped service account that can create quotes and orders and nothing else. Read more on how Lily reads and writes in Prophet 21, SAP and NetSuite.

Does it get distribution details right: contract pricing, units of measure, part numbers and substitutions?

Score accuracy line by line on your messiest real requests, not the vendor's demo. Build the sample from the requests that break demos: a competitor's part number, a superseded part, a vague description, a non-stock line, an order in the wrong unit of measure. Check every quoted line for the right part, unit of measure, pack size and contract price. A wrong unit of measure is worse than no quote.

How Lily handles this. Lily reads requests from email, PDF, fax, web forms and portal cases, and orders already on EDI stay on EDI. She matches each request to your parts, including competitor numbers and superseded parts, and resolves descriptions, verified substitutes and non-stock sourcing. She learns your pricing, product data, branch rules and approval logic, recovers unwritten pricing rules from your quote history, and flags where they conflict. Pricing is applied and exceptions are surfaced before anything is sent.

What happens when it is not sure?

A good platform routes uncertain or high-risk work to a named person before it executes, and logs every action. Ask the vendor to walk you through one exception: a quote below your margin floor, a new customer, or a part match it is not confident about. Then ask to see the audit trail for that exception.

How Lily handles this. Each kind of action is set up front as automatic, approved by a named person, or human-only. Quotes that break a contract or fall below the margin floor go to a person, and so do low-confidence part matches. You set the approval gates yourself, for example a margin below floor, a new customer or a non-stock line. She escalates rather than guesses, and every action she takes is logged. See how Lily routes quotes for approval.

How should you run a pilot, and what should you measure?

Pick one workflow, record a baseline before the pilot starts, and measure completed work, not model accuracy. Useful distributor measures include median first-response time by channel, conversion on quotes created, the share of RFQs answered untouched, end-of-day backlog per rep, PO touchless rate, time from PO to confirmation, and the monthly count of duplicate orders. Measure adoption every week as well. A platform your reps work around will not move any of those numbers.

How Lily handles this. Results are tracked against a target agreed at the start, by week 12, and adoption is measured weekly. For a $600M+ industrial automation distributor, Lily runs RFQ intake, quoting and order attribution around the clock. The results in two months: 11,000+ hours saved annually and $5.5M+ ARR in incremental revenue. See how a first deployment is measured.

Where does your data live, and what do you own?

Ask where your data is processed, whether it trains anyone else's model, and what stays yours if you stop using the platform. Ask for the security report itself, not a summary of it. Then check who is the data controller and who is the processor.

How Lily handles this. Each customer gets a single-tenant environment, in InstaLILY's cloud, your cloud or on premises. No cross-customer models are trained, so your data trains your model only. You own the model and the knowledge inside it, and what Lily learns stays yours. You remain the data controller, and InstaLILY is the processor. InstaLILY holds a SOC 2 Type II attestation for the report period 23 Feb 2025 to 22 Feb 2026, with no exceptions. Sign-in runs through your own identity provider: Okta, Azure AD or another SAML 2.0 provider. See where your data lives and how it is protected.

Should you build it in-house or buy it?

Building means owning the integration, the exception handling and the upkeep, with engineers who know your ERP. Packaged tools cover the common workflow, but rarely your pricing logic and approval chain. So the real question is who adapts the software to your business, and who keeps it working after go-live.

How Lily handles this. Lily is software that works the way a forward deployed engineer works. She learns the work, builds what the work needs and runs it inside the systems you already use, so you get the fit of a build without building and maintaining it yourself. There is no rip-and-replace, no migration and no new system of record. Read what an AI forward deployed engineer does.

Should you wait for your ERP vendor's AI?

Your ERP vendor's AI is worth checking first if you run its current cloud version. The question is whether it covers the whole workflow across your inbox, CRM and ERP, or only the steps inside the ERP. Many RFQs start in an inbox, long before they reach the ERP.

How Lily handles this. Lily sits on top of the ERP, not in front of it, and works alongside the vendor's own features. The ERP stays your system of record. If your ERP vendor's AI already covers your workflow, use it. If you want to see what Lily finishes on your own orders, bring a sample of your RFQs, and if she is not the right fit for your workflow, we will tell you.

Frequently asked questions

How long should an AI pilot take for a distributor?

With Lily, first quotes typically go out within days of data access, and she is fully live in about two weeks. Whatever platform you test, read the results against a target agreed before you start.

Do we need clean data before we start?

No. Messy data is the starting point, not a prerequisite. Test any platform on your real requests, with all their gaps, rather than on a cleaned-up sample.

Will an AI platform replace our ERP or CRM?

It should not. Your ERP stays the system of record, and the platform should work inside it. Lily writes quotes and orders into the ERP, and notes and tasks into the CRM you already use.

What is the difference between an AI agent and an AI sales tool?

An AI agent completes a transaction in your systems, such as a quote or a sales order. An AI sales tool tells a person what to do next, and the person still does the work. Ask every vendor which of the two it sells.

Is an AI forward deployed engineer a person?

No. Lily is software that works the way a forward deployed engineer works. She learns the work, builds what it needs and runs it inside your systems.

Which ERPs does Lily work with?

Lily works with Epicor Prophet 21, SAP, Oracle NetSuite, Infor and Oracle JD Edwards, alongside Salesforce and HubSpot on the CRM side. Most deployments start read-only. Read more on how Lily reads and writes in Prophet 21, SAP and NetSuite.

Share