The short answer
In most companies, no. The orders, jobs, purchase orders, receipts, shipments and invoices your ERP records every day are exactly the data AI needs. What is usually missing is not a new system but something that reads that data every day, connects it across tables, and tells people what to do about it.
That is what an AI layer does. It sits on top of the system you already run, reads it without changing it, and turns patterns into a short daily list of decisions: which order is losing money, which job is likely to spoil, which supplier is about to make you late. Replacing the ERP is worth considering only when the system itself is the constraint, and even then it rarely needs to happen in one go.
Why replacement is a hard first step
ERP projects are long, expensive and disruptive, and the value usually arrives only at the end. In Panorama Consulting Group’s 2026 ERP Report, more than a quarter of organisations exceeded their project budgets. The business also has to keep running while the migration happens: orders still need entering, jobs still need printing and trucks still need loading.
Most operations leaders who ask about AI are not asking for new screens. They are asking for answers their current system cannot give. Those are different problems, and the second is usually much smaller.
What your ERP already knows
Every table below exists in a typical manufacturing, distribution or print-on-demand ERP. Each one can answer a question with money attached once it is read alongside the others:
| Data you already record | Question it can answer |
|---|---|
| Sales orders, invoices, freight bills | What is the true margin of each order and customer? |
| Jobs or work orders, reprint and scrap records | Which jobs are likely to spoil, and why? |
| Purchase orders and receipts | Which supplier is slipping, and which orders does that put at risk? |
| Shipments against promised dates | Which open orders will ship late if nothing changes? |
| Inventory transactions | Which items will run out, and which are sitting unused? |
What an AI layer is, and what it is not
- It reads your data, usually from a copy or a scheduled export, so day-to-day performance is not affected and nothing in the ERP is changed.
- It joins tables that reports usually keep apart, for example a job’s blank batch, press, shift and reprint history.
- It learns from your own history which combinations lead to waste, delay or loss.
- It produces a short list each day, with a specific action for each item, delivered where people already work: email, Slack, Teams or a simple screen.
It is not a dashboard project, and it does not ask your team to enter data differently. If a key fact is not being captured at all, for example why a garment was reprinted, the layer will show that gap. Closing it is usually a small change to one screen, not a new system.
Layer, modify or replace: how to decide
| What you see | Usual answer |
|---|---|
| The data exists, but nobody sees the pattern in time to act | Layer: add intelligence on top |
| One important fact is not captured, or a step lives in spreadsheets | Modify: add a field, a screen or an integration |
| A whole process cannot run in the system, or a module is unsupported | Modernize that module, one at a time, while the rest keeps running |
| The platform is end of life, has no way to read data out, and the data cannot be trusted | Replace, in phases, with the old and new running side by side |
Most companies we talk to sit in the first two rows. The third is more common than the fourth, and even the fourth rarely needs a big-bang switchover. (Our Stakes Manufacturing case study shows a module-by-module modernization that never paused production.)
How to test it on your own data
- Pick one question with money attached, such as ‘which orders lose money?’ or ‘which jobs will spoil?’, and agree who will act on the answer.
- Take a read-only extract of the relevant history, ideally a year or more, so seasonal patterns are included.
- Build the first model and test it against a past period: would it have flagged the orders or jobs that actually went wrong?
- Run it alongside your normal process and compare its daily list with what actually happens, before anyone relies on it.
Once that is done, you know, on your own data, whether the answer is useful and what it is worth. That is a much smaller decision than a new ERP.
Are you ready?
- At least a year of order and job history in one system.
- Consistent IDs for customers, items and jobs (they do not have to be perfect).
- Someone who owns the decision the answer is meant to change.
- Read access to the database, a scheduled export or an API.
Want to know what your ERP could tell you? Bring us one problem. We’ll look at your data read-only and show you.
Bring us one problem →Related: The AI Intelligence Layer · Stakes Manufacturing case study · True job costing for print shops
Frequently asked questions
Do I need to replace my ERP to use AI?
Usually not. Most of the value comes from data your ERP already records. An AI layer reads it, connects it and flags decisions, without changing the system.
Will an AI layer slow down my ERP?
It should not. It normally reads from a copy of the database or a scheduled export, outside the system your team works in.
Does it work with older or custom systems?
Yes, as long as the data can be read: directly from the database, through an export or through an API.
What if our data is messy?
Most companies’ data is. Start with the few fields that matter for one question. The layer will also show which gaps are worth fixing.
When does an ERP really need replacing?
When the platform is end of life, data cannot be read out or trusted, or whole processes cannot run in it. Even then, a phased replacement is usually safer than a big-bang switch.
