Best ERP for Manufacturing

ERP plus an AI layer: how manufacturers are modernising the office

The slowest steps in many plants happen in inboxes, not in the ERP. A sourced look at what an AI layer can take over, what it must leave alone and how to start.

By The editors · Updated · 4 min read

An estimator on a desk phone in a plant front office with a printed engineering drawing and a quote sheet spread in front of him.
Quote turnaround is often set by one estimator's inbox.
Contents
  1. Where the office loses time
  2. Two ways to add AI
  3. What must stay in the ERP
  4. Controls
  5. A 90-day plan
  6. Questions buyers ask
  7. Sources

A manufacturer can run a capable ERP and still lose orders to slower office work. The RFQ waits for the one estimator who knows the part. Material is ordered when the job is released. Certificates are found the week of the audit. ERP vendors are adding AI to address some of this, and a separate category of AI layers now addresses the rest.

Where the office loses time

OpenLazarus's manufacturing page describes four patterns it designs for: quotes waiting on one estimator, schedules that move faster than the ERP, material ordered at release with mill certs lost in inboxes, and quality records kept in binders⁠1. The page illustrates them with an invented example company, so they are a vendor's framing rather than evidence. They are nonetheless the steps where most of the correspondence in a plant happens, and they sit outside the transaction screens that ERPs are built around.

Two ways to add AI

AI inside the ERP. Business Central's Sales Order Agent monitors a mailbox, reads requests and attachments, checks availability, can promise delivery dates using capable-to-promise logic, and prepares a quote for review⁠2. Epicor's Prism agents give Kinetic users conversational access to live ERP data and documents⁠3. JobBOSS² drafts bills of materials from PDFs and images⁠4. These features work on the ERP's own records and follow its permissions.

An AI layer beside the ERP. OpenLazarus gives each AI teammate an inbox and a phone number and puts it on quotes, purchasing, scheduling and quality, reading jobs, routings and inventory from the ERP and writing back what changes⁠15. The layer works across the ERP, email, phones and documents at once, which suits correspondence-heavy steps.

Office stepInside the ERP (examples)AI layer (OpenLazarus)
Customer RFQs and orders by emailBusiness Central Sales Order Agent⁠2Quotes drafted from past routings for the estimator⁠1
Purchase orders and supplier follow-upNot described by most vendorsOrders material for released jobs, flags price rises⁠1
Certificates and NCRsQuality modules where presentFiles certs and NCRs against job, part and heat number⁠1
BOM creation from drawingsJobBOSS² AI BOM builder⁠4Not described
Costing and the booksERPStays in the ERP⁠1

What must stay in the ERP

MRP, BOMs and routings, inventory valuation, job costing and the general ledger. The rule is simple: if a record feeds the financial statements or the material plan, the ERP owns it, and the AI layer only proposes or writes back through the ERP's own rules. OpenLazarus says it keeps "one record" this way⁠1; ask any vendor to show the write-back on your system.

Controls

Set approval limits before go-live. OpenLazarus's examples are a material order over $20,000 or a change to a promised ship date, both of which wait for a person, with every action logged⁠1. Business Central requires designated users to review every outgoing message from its agent⁠2. For AS9100 or ISO 13485 shops, confirm that agent-filed records carry who did what and when.

A 90-day plan

  1. Weeks 1 to 2. Measure one step today: quote turnaround, late purchase orders or certificates missing per audit.
  2. Weeks 3 to 4. Connect read-only to the ERP and the shared inbox for that step.
  3. Weeks 5 to 8. Switch on drafting with every action waiting for approval.
  4. Weeks 9 to 12. Raise limits only where a month ran cleanly; hold one month-end with the layer live.
  5. Day 90. Compare the metric with the baseline and decide on the next step.

The end state is specific: the estimator reviews drafted quotes instead of building them, purchase orders go out when the rule says rather than when someone remembers, certificates are on file before the auditor asks, and the ERP still holds every cost and every balance. Our main guide ranks the ERPs and the AI layer for this situation.

Questions buyers ask

What is an AI layer for an ERP?

Software whose AI agents do the correspondence and data entry around the ERP, such as reading RFQs and customer emails, drafting quotes and purchase orders, chasing suppliers and filing quality records, and then write the results back into the ERP so it remains the single record.

Does an AI layer replace the ERP?

No. The vendors that sell one say so. OpenLazarus says it reads jobs, routings and inventory from the ERP and writes back what changes, and that most shops keep the ERP for job costing and the books.

What should a manufacturer automate first with AI?

Pick the office step with the highest volume and the clearest measure, such as quote turnaround, purchase orders placed late or certificates missing at audit. Run it with every AI action waiting for approval for a month, then raise the limits where it ran cleanly.

Sources

  1. Manufacturing: every quote, job, machine and part, on one system. OpenLazarus. Accessed 8 Nov 2026.
  2. Sales Order Agent overview: Business Central. Microsoft Learn, 2 Oct 2026. Accessed 8 Nov 2026.
  3. Epicor Prism launches across European markets, bringing embedded, industry-specific AI to manufacturers. Epicor via Business Wire, 23 Jun 2026. Accessed 8 Nov 2026.
  4. JobBOSS²: centralize job shop data and scale operations. ECI Software Solutions. Accessed 8 Nov 2026.
  5. The agent library. OpenLazarus. Accessed 8 Nov 2026.

Updated 8 Nov 2026: First published.

Further reading