ERP migration is not the first move. AI agents work on top of the on-prem ERP you have — migrate when the data proves what you use.
Every equipment dealer we sit with has the same sentence ready: "we can't rip it out." The ERP is on-prem, installed when the second location opened, decades deep in parts history, serial numbers, and warranty records. The vendor has a cloud version — these days usually owned by a private-equity firm — and the ERP cloud migration quote keeps getting worse. So the system stays, and the work around it stays manual: claims re-keyed into manufacturer portals, parts leaking out of warehouses and the backs of service trucks, dispatch living in one person's head. The problem was never where the ERP lives. It is that the business rules live in people, not in the system.
What an ERP migration actually moves
A migration moves data and screens. It does not move judgment. Dealers who migrate discover, a year or two and a seven-figure invoice later, that the new system runs the same workarounds as the old one — the claims are still re-keyed, the parts still leak, the dispatcher's head is still the dispatch board. Worse, the decades of history turn out to be mostly dead weight: SKUs unsold since the last recession, customers acquired and renamed twice, price files from vendors nobody stocks anymore. Migrating everything means paying to move that weight, then paying again to clean it up on the other side.
SAP data migration is the cleanup, not the move
The same story runs at SAP scale. One mid-market B2B company we work with runs SAP ECC on-prem — decades of customers, materials, and transactions in a system carrying years of compromises. Their first phase is SAP data extraction into a cloud data platform — Google Cloud, in their case — to feed a customer-facing AI application. Their executive sponsor said it in one sentence: "the complexity for us is to clean up the data and make sure we filter it properly." Not the move. The filter.
Open an SAP estate that has run for twenty years and the extract tells the truth. Customers entered three times under three spellings. Material masters renamed with every acquisition. Custom fields bolted on for processes nobody runs anymore. Tables nobody has read since the last decade. Moving all of it raw is not a migration — it is relocating the mess at seven figures into a platform that then bills you monthly to store it. The cleanup is the migration: dedupe the master records, decide which fields are real, write down which tables are live and which are history. The rest is plumbing.
There is a second reason to do the cleanup first, and it is the one that decides whether AI works on top of any of this: agents need governed schemas, not raw dumps. An agent pointed at a raw extract inherits every compromise as fact — it quotes the dead customer, orders the renamed material, trusts the field nobody has maintained since 2019. An agent pointed at a governed copy — deduped, defined, owned, filtered to what the business actually uses — does the work you hired it for. The cleanup is not preparation for the AI project. It is the AI project, and it starts wherever the data already lives, on-prem ERP included.
Is SAP data extraction enough, or do we need a full legacy data migration?
Usually enough, for years. Extraction reads the system where it sits — scheduled exports, the tables underneath — and lands the data, cleaned and governed, in your cloud platform. The agents run against that copy while the ERP keeps running the business. Legacy data migration becomes a real question only when the estate itself is being retired — and by then the cleanup is done, the live data is identified, and what moves is the governed set instead of three decades of everything. Extraction first; migration if and when it earns its invoice.
What do AI agents on top of the ERP actually do?
They work on the system you already run. An on-prem ERP still exposes its data — scheduled reports, exports, the SQL database underneath — and that is enough. AI agents read it and do the work around it: assemble each warranty claim against that manufacturer's rules and file it, reconcile parts across warehouses and service trucks, chase the reimbursement that was earned and never collected. As one dealer put it, a warrantied part they follow up on is "100% margin" — which is exactly the work that slips when it depends on a person logging into a fifth portal. The ERP stays where it is. The agents run in your cloud account, on your data, and a person reviews whatever the rules cannot clear.
Then the interesting thing happens: the migration question answers itself. The encoded rule set is now the written record of what the business actually does — which workflows run every day, which data they touch, which corners of the old system nobody has opened in years. If a migration ever makes sense, it moves what is proven to be used instead of three decades of everything, and the integration map already exists. Most dealers find the urgency simply leaves: the system they refused to rip out is doing the work they wanted the new one for.
You do not migrate the ERP to get the work done. You get the work done, and the migration question answers itself.
This is implementation work, not migration consulting. Zaigo is an AI implementation firm: we build the agents, they run in your cloud account against your data, and the encoded rule set is your IP — written down, yours to keep, portable to whatever ERP you eventually land on. Your people stay in the loop throughout; anything the rules cannot clear lands in an exception queue with the reason attached. The claims side of this is on the warranty claims processing page (zaigo.ai/workflows/warranty-claims-processing); the parts side — warehouses, branches, technician vehicles — is on the inventory discrepancy page (zaigo.ai/workflows/inventory-discrepancy).
If the ERP migration quote is sitting in a drawer while the manual work piles up, the next step is a 30-minute working call: zaigo.ai/book-a-call. We will tell you on the call what agents on top of your current system would take — and what they would pay back.
All insights