AI and analytics on the ERP you already have
You do not migrate to an intelligent ERP. You make the one you have intelligent, in a particular order.
Forecasting, anomaly detection, and asking your own data questions in plain language all work now. They work on trustworthy data, and fail quietly on the other kind.
Typical timeline: First working model in 8–12 weeks
Intelligence sits above the ERP, and is built bottom-up
These are the signs this is worth doing
If several of these are true, this is usually where the fastest return sits.
Almost every ERP vendor now sells an intelligent edition. Most of what gets demonstrated is a chat box in front of the reports you already had, and the migration being proposed is a licence upgrade with a new word on it.
That is a shame, because the underlying idea is sound. There genuinely are things a modern system can do with your operational data that a report cannot, and several of them are mature rather than experimental. They just have almost nothing to do with replacing your ERP.
The honest framing is this: your ERP is a system of record, and it is good at that. Intelligence is a layer that sits above it, reads from it, and hands conclusions back. Getting that layer right does not require changing the system underneath — which is fortunate, because changing it is the expensive part.
What actually works today
Each of these is in production somewhere, and each is a defined piece of work rather than a platform.
Demand and inventory forecasting
Reorder points set from demand patterns, seasonality, and lead-time variability rather than from a number somebody set in 2019. The most mature item on this list, and usually the clearest return — stock-outs and dead stock are the same problem measured twice.
Anomaly detection on transactions
Duplicate invoices, prices outside the agreed band, unusual approval patterns, quantities that do not fit the history. Not fraud detection in the dramatic sense — mostly it catches ordinary error, which is far more common and quietly more expensive.
Asking your data questions
Plain-language querying over your own operational data, so a manager can ask what receivables look like by region without waiting for a report. This works well now, with an important caveat covered below.
Documents into the ERP
Invoices, purchase orders and delivery notes read, reconciled against open orders, and posted. This is where automation and the ERP meet, and it removes more keying than anything else here.
None of this requires a new ERP. All of it requires your existing ERP data to be worth trusting — which is where these projects actually succeed or fail.
The data foundation is the project
A forecast built on inconsistent item codes will produce confident numbers about products that do not exist. Anomaly detection across duplicated customer records will flag every second transaction until people stop looking at it. A plain-language query tool will answer fluently and wrongly if the underlying tables encode three different meanings of dispatched.
So the first phase is not modelling. It is a data layer: the ERP data extracted, item and customer masters deduplicated, fields given one agreed meaning, and the definitions written down. Unglamorous, and it is the difference between a system people rely on and one they quietly stop opening.
The useful side effect is that this phase pays for itself before any model exists. Once the data is consistent and queryable, ordinary reporting improves immediately — most businesses find questions they had simply stopped asking.
Two ways this gets sold
The distinction is worth holding on to when a proposal arrives.
The upgrade pitch
- Migrate to the vendor's intelligent edition
- Intelligence arrives as a bundled feature
- Your data problems come along unchanged
- Value depends on finishing the migration
- Locked to one vendor's roadmap
- Priced as a licence uplift
A layer above what you have
- The ERP stays exactly where it is
- One capability at a time, against a real problem
- Data cleaned first, deliberately
- Each phase useful whether or not the next happens
- Portable if you change ERP later
- Priced as work, and stoppable
On asking questions in plain language
This is the demo everyone wants, so it is worth being precise about what makes it safe. A language model translating a question into a database query is genuinely capable, and it is also capable of producing a confident answer from a subtly wrong query — which is far more dangerous than an error message.
The control is a semantic layer: metrics defined once, in one place, with agreed meanings, so the model composes from verified definitions rather than inventing joins across raw tables. Every answer shows the query behind it and the figures it came from. Anything the layer cannot answer says so rather than guessing.
With that in place it is genuinely useful for operational questions. We would still not put an unverified generated number into a statutory filing, and neither should anyone else.
The order we would do it in
- Extract and model the ERP data; agree what each field means, in writing.
- Clean the masters — items, customers, suppliers — with your team, not consultants.
- Ship better ordinary reporting first, so the foundation proves itself.
- Add one predictive capability against a measured problem, usually forecasting.
- Add anomaly detection once there is a clean baseline to detect against.
- Add plain-language querying last, on a semantic layer, never on raw tables.
An intelligent ERP is not a product you buy. It is what an ordinary ERP becomes once its data is trustworthy and something useful is reading it.
We would rather tell you now than three weeks into a project. This work is usually the wrong call if any of the following describes you.
- Businesses whose ERP data is genuinely unreliable and who do not want to fix it first. Models on bad data produce confident nonsense, which is worse than no model.
- Anyone with too little history. Forecasting needs a few cycles of reasonably consistent data; below that a rule beats a model and costs nothing.
- Operations small enough that one person already sees everything accurately. You do not have this problem yet.
- Anyone wanting AI to post to the ledger unattended. We will not build that, and you should be wary of anyone who offers it.
The systems involved
We integrate rather than replace wherever it makes sense. These are the systems this work most commonly touches.
Intelligent ERP — questions we get asked
Do we need to move to an AI-enabled ERP edition?
Usually not. Almost everything worth having reads from your existing ERP and sits alongside it. If a vendor's intelligent edition offers something specific you actually need, that is a fair reason to upgrade — but wanting AI on its own is not one.
How much history do we need?
For forecasting, ideally two to three years of reasonably consistent transactions, and more if your business is seasonal. Anomaly detection needs considerably less, since it learns from a recent baseline. We look at your actual data before promising anything.
Can it read our ERP directly?
Technically often yes, and it is usually the wrong design. Querying a live ERP puts load on the system running your business, and ERP schemas are shaped for transactions rather than analysis. We extract into a data layer built for reading.
Will it tell us something we do not already know?
Sometimes, and honestly, sometimes not. The reliable wins are not revelations — they are consistency and timeliness: the same question answered the same way every time, on current numbers, without waiting for someone to build a report.
The services this work sits inside
Where this comes up most
The sectors where we most often do this work, and where the payback is usually clearest.
Others worth reading
Moving from Tally to a real ERP
Most businesses do not need to leave Tally. They need to stop using it for the things it was never built for, and move those onto something else first.
Read itIntegrationReading and checking documents automatically
Automated document reading works well when it is designed around its own uncertainty — routing what it is not sure about to a person instead of guessing.
Read itThinking about intelligent erp?
Start with a short conversation. We will tell you honestly whether this is the right place to begin, or whether something else pays back faster.