AI in ERP systems — practical applications, not promises

According to a KPMG study from 2025, 77% of Polish enterprises declare an intention to increase spending on artificial intelligence within the next two years. At the same time, more than half of them admit they do not know where to begin. This is a gap we know from first-hand experience — not from conference presentations but from daily work with our clients' ERP systems.

What artificial intelligence actually does in an ERP system

Before we turn to specifics, it is worth separating two realities. The first is a marketing reality in which AI “revolutionises”, “transforms” and “automates everything”. The second is an operational reality in which AI performs strictly defined tasks faster, more accurately and more repeatably than a human — provided that someone has properly defined those tasks.

At KJ Consulting we work with Odoo 18 and 19. Every day. Not as a trade-fair demonstration but as a production tool for industrial, trading and service companies. Here is what AI actually does in these systems.

Report and analysis automation

An ERP system generates vast quantities of data — sales transactions, stock levels, production settlements, margins on individual products. The problem is not a lack of data but the fact that nobody has time to analyse it regularly.

AI in the role of an analytical assistant changes this situation fundamentally. Instead of asking an employee to prepare a weekly report, we create a scheduled action in Odoo that independently aggregates data, identifies deviations from the norm and generates a clear summary. Computed fields allow continuous monitoring of indicators that previously required manual compilation in a spreadsheet.

The crucial difference: this is not a report “on demand”. It is a report that creates itself, reaches the right person automatically and signals when something requires attention.

Intelligent notifications

One of the implementations that delivered an immediate, measurable effect was a system of weekly price-list notifications for the sales team. The mechanism is simple to describe but remarkably effective in practice: a scheduled action in Odoo analyses price lists — in PLN and EUR — every Monday at 07:00 for new items added within the previous seven days. If new pricing templates have appeared, the sales team receives an organised summary in their company inbox.

The result? Instead of a situation in which a salesperson learns about a new pricing entry by chance (or does not learn about it at all), every team member begins the week with up-to-date knowledge of available prices. Response time to quotation requests has shortened, and the number of quotations with manually entered “approximate” prices has dropped to virtually zero.

Quotation and pricing support

Product configurators in Odoo — particularly in industries with high variability — are an area in which AI delivers clear value. Algorithms can suggest product variants based on the customer's order history, automatically apply discounts in accordance with pricing policy and verify configuration consistency before a quotation is sent.

This is not “AI that sells on its own”. It is a tool that eliminates the most time-consuming and error-prone stages of quotation preparation — finding the correct price list, calculating the margin, verifying availability. The salesperson still makes the decision. But they make it faster and on the basis of complete data.

Sales data analysis and predictions

This is the area in which expectations most frequently diverge from reality. Predictive models can identify sales trends, demand seasonality and customer churn risk. But — and this is an important caveat — the quality of predictions is directly proportional to the quality of data in the system.

A company that diligently maintains its CRM, records the reasons behind won and lost sales opportunities and keeps its contact database current can expect valuable forecasts. A company in which the CRM is “a box to tick” will receive nothing more than attractive-looking charts devoid of informational value.

What AI will not do for you

This section is as important as the previous one — perhaps even more so, because it guards against costly disappointments.

AI will not replace a properly configured ERP

Artificial intelligence is a layer that operates on top of an ERP system — not instead of it. If the warehouse structure is chaotic, logistics routes are undefined and BOMs (bills of materials) are incomplete, no algorithm will fix that. AI can optimise processes that are already working. It cannot invent them from scratch on the basis of disordered data.

AI will not fix bad processes

This is a trap many organisations fall into: “we will implement AI and the processes will fix themselves”. They will not. If the purchase order approval process requires five signatures and takes three weeks, AI will at best accelerate the forwarding of documents between five inboxes. The proper solution is to simplify the process — and that requires a management decision, not a technological one.

AI requires clean data

The principle of “garbage in, garbage out” takes on particular severity in the context of artificial intelligence. A model trained on incomplete or inconsistent data does not merely produce wrong results — it produces them in a confident tone that can lull users into a false sense of security. Before any AI deployment, data must be put in order: product naming standardised, missing fields populated, duplicates eliminated.

What AI implementation in Odoo looks like in practice

Based on our implementations, we have developed a four-phase model that minimises risk and allows a rapid assessment of whether an investment in AI makes sense for a given area of the business.

Phase 1: Process audit

We identify tasks that are repetitive, time-consuming and error-prone. We do not look for “processes to automate” — we look for specific activities that an employee performs according to a fixed pattern, spending a disproportionate amount of time on them. Typical examples: generating weekly reports, manually transferring data between modules, checking stock levels before confirming an order.

Phase 2: Pilot on a single module

We never deploy AI “across the entire system at once”. We select one module — most often CRM, pricing or inventory management — and test the first solution there. The pilot typically lasts between two and four weeks. That is long enough to assess real-world utility and short enough to contain costs in the event of failure.

Phase 3: Measuring results

Before launching the pilot, we establish measurable success criteria: by how much has quotation preparation time decreased? How many pricing errors were detected automatically? How has the frequency of CRM data updates changed? Without hard comparative data, the assessment of an implementation reduces to subjective impressions — and those can be misleading.

Phase 4: Scaling

If the pilot confirms the value of the solution, we extend it to additional modules and teams. Scaling encompasses not only technical configuration but also user training — because even the best automation will not deliver results if people do not trust it or do not know how to use it.

Case study: price-list notifications

It is worth examining one of our implementations more closely, as it illustrates both the capabilities and the limitations of AI in an ERP environment.

A client — a trading company with a sales team of around a dozen people — was struggling with a seemingly trivial problem: salespeople did not know which products already had automatic pricing in the price configurator and which required manual calculation. The consequences were quotations with outdated prices, unnecessary enquiries to the procurement department and delays in customer service.

The technical solution is relatively straightforward: a scheduled action in Odoo, triggered every Monday, monitors the date_start field in price-list items from the previous seven days. The output — a tabular summary of new items, split by PLN and EUR price lists — is sent to the company mailing list and simultaneously recorded in the price list's communication history.

The code must be compatible with the Odoo sandbox environment, which imposes certain constraints: no imports, no function definitions, no Markup objects. This requires experience — not every integrator is aware that message_post escapes HTML and that mail.message.create with an Html-type field must be used instead.

The implementation result: the time from adding a new price-list item to the moment the sales team learns about it shortened from “several days to several weeks” (depending on chance) to “a maximum of seven days” (guaranteed). Implementation cost — a few hours of consultant work. Return on investment — immediate.

Points of caution: security and limitations

Deploying AI in an ERP system entails risks that are worth knowing about in advance.

Data security

An ERP system contains sensitive data: purchase prices, margins, personal data of employees and counterparties, contract terms. Every integration with an external AI model — whether through an API or through tools such as Claude Code — requires a conscious decision about which data leaves the company's infrastructure. In our practice, we apply the principle of minimum access: AI receives only the data strictly necessary to perform a specific task.

Odoo sandbox

The Odoo execution environment — in both the on-premise and SaaS versions — imposes significant constraints on code executed within scheduled actions and automation rules. These constraints exist by design: they protect the system against unauthorised operations. An integrator must understand these constraints and be able to build effective solutions within them. Attempting to circumvent the sandbox is a direct path to system instability.

API keys and permissions

Integration with external AI services requires API keys that must be stored securely, rotated regularly and monitored for excessive usage. In Odoo, API keys should be kept in System Parameters or in the server's environment variables — never in module source code or in fields accessible to ordinary users.

Conclusion: AI as a tool, not a goal

Artificial intelligence in an ERP system is not a goal in itself. It is a tool — effective but demanding. It requires clean data, a properly configured system, well-defined processes and realistic expectations. Meeting these conditions is not trivial, but the reward is proportional to the effort: measurably shorter operating times, fewer errors and better decisions based on current data.

At KJ Consulting, we implement AI in Odoo not because it is fashionable but because it delivers concrete, quantifiable results for our clients. If you are considering the use of artificial intelligence in your ERP system — let us talk. Not about visions of the future, but about what we can do right now, in your system, with your data.