Skip to content
Concepts 7 min read Last updated 3 August 2026

What is conversational AI for IFS Cloud?

The phrase gets used for three quite different things: a search box with better grammar, a documentation assistant, and a system that actually changes records. The difference matters more than the label.

In short

Conversational AI for IFS Cloud is software that lets a person use IFS by typing or speaking plain language instead of navigating screens. The useful ones do two jobs: answer questions from live IFS data, and carry out transactions — creating, updating and releasing records — inside the user’s existing IFS permissions.

  • Conversational AI for IFS replaces navigation, not the ERP.
  • The meaningful dividing line is read-only Q&A versus transactional write access.
  • Permission inheritance is what makes write access acceptable to a security review.
  • Grounding in live IFS metadata is what stops an assistant inventing fields and values.

What does conversational AI for IFS actually mean?

Conversational AI for IFS Cloud is an interface layer that turns a plain-language request into the correct IFS operation. The person describes the outcome — "which work orders at site 100 are overdue" — and the software resolves which IFS projection and entity set that maps to, builds the query or payload, and returns the result in readable form.

The important word is interface. Nothing about the ERP changes. The data model, the permission model, the validation rules and the audit trail all stay exactly where they were. What changes is that reaching them no longer requires knowing that a fault report lives under a particular navigator entry, or that the field is called ObjState.

That distinction is why conversational access is not the same as "AI in the ERP". Predictive maintenance scores, demand forecasting and anomaly detection are AI features inside a process. A conversational interface is a way in — it sits in front of everything and does not care which process you are reaching for.

What is the difference between a chatbot and a conversational AI assistant?

A traditional chatbot matches your sentence to a pre-built intent and runs a scripted flow, so it fails as soon as you phrase something unexpectedly. A modern conversational AI assistant reasons about the request, discovers the relevant API at run time, and composes the calls — which is why it can handle requests nobody scripted in advance.

This is not a marketing distinction. Intent-classification chatbots were the previous generation of ERP conversational tooling, and their failure mode was structural: every new phrasing needed a new training utterance, and every new transaction needed a new flow built by hand. Coverage grew linearly with effort, which meant it never grew far.

Tool-calling assistants invert that. Instead of scripting the flow, you give the model a set of capabilities — search the metadata, read the spec, create the record — and it sequences them for the request in front of it. Adding a new IFS area does not require new dialogue design, because there is no dialogue design.

Can conversational AI change data in IFS, or only read it?

Both are possible, and most products do only the first. Read-only assistants answer questions and stop. Transactional assistants also create, update and release records through the IFS OData REST API. The difference decides whether the tool saves a lookup or actually removes work.

Read-only is the market default for a reason: writing to a system of record is genuinely harder to govern. It requires resolving every mandatory field correctly, respecting concurrency, honouring permissions, and giving the user a chance to stop before anything commits.

For comparison, Acumatica documents its own AI Assistant as unable to "perform actions, update records, or access data outside of the configured inquiries" — a clear, honest statement of a read-only design. Products that do write should be equally explicit about how they make that safe.

How does conversational AI respect IFS permissions?

The right approach is to authenticate each person with their own IFS account and call IFS with that user’s access token. IFS then applies its projection-level permission model exactly as it does in Aurena, so the assistant can never read or write anything the person could not reach through the normal interface.

IFS Cloud grants access to projections through permission sets, with each projection granted one of four levels — none, read-only, custom or full. Because that check happens inside IFS, an assistant that arrives as the real user inherits it for free and cannot circumvent it.

Products in this category differ here more than anywhere else, and it is the first thing worth asking a vendor. Either every call reaches the ERP as the person who typed it, or there is some arrangement involving a single integration account and application logic deciding who may see what — and those two designs have very different conversations ahead of them.

What stops it inventing data?

Grounding. A well-built IFS assistant never fills a field from memory: it reads the required fields, enumeration values and list-of-values from your live IFS metadata, and if it cannot find valid options it says so instead of offering plausible ones. Every figure it reports comes from a query result, not from the model.

The practical rule is a fixed fallback chain for any field with a constrained value: enumeration values first, then list-of-values, then a plain text input. Some IFS fields look like enums but are actually LOV-driven — country and language codes are the usual culprits — so an assistant that only checks one source will present an empty dropdown, and a poorly designed one will fill it in for you.

Note the claim to be sceptical of: "no hallucinations". Nobody can guarantee that. What can be guaranteed is the mechanism — that business values are read from IFS and shown with the record they came from.

Sources

Claims about other vendors are quoted from that vendor’s own published documentation and were checked on 3 August 2026. IFS ships two release updates a year, so treat any comparison as dated.

Questions people also ask

Is conversational AI for IFS the same as IFS.ai?
No. IFS.ai is IFS’s own umbrella brand for AI capabilities across IFS Cloud. Conversational AI for IFS is a category — a way of interacting with the ERP — which several products address in different ways, including IFS’s own and independent products such as NgageChat.
Does it replace IFS Aurena?
No, it sits alongside it. People who work in IFS all day are usually faster in the screens and should stay there. Conversation is aimed at everyone else — and in most organisations everyone else is the majority of the licence count.
Do we need to be on the latest IFS Cloud release?
That depends on the product. An independent assistant that talks to standard IFS OData REST APIs has different version requirements from features shipped inside a specific IFS release. Ask about your exact version before assuming either way.

Early access

See NgageChat answer a question from your own IFS environment

A 30-minute session: bring one real scenario — an overdue work order list, a fault report, a supplier lookup — and watch it run end to end. No slideware.