Simon Paris, Unit4
Midmarket ERP vendor Unit4 today launches a try-before-you-buy offer of its AI capabilities that will see customers who sign up this year getting free access through the end of August 2027, subject to a fair usage cap. At the same time, the vendor has been evolving its cloud-native ERPx platform to support new agentic capabilities, and is aiming to work with a variety of foundation models, including those from European model providers, to cater for customers with AI sovereignty concerns.
Simon Paris, CEO of Unit4, explains that the aim of the free trial period is to encourage adoption across service-centric industries, and therefore the vendor will also be supporting customers with how-to advice and services. He goes on:
We’re saying, hey, you can have all of the agents free of charge for a considerable amount of time. You can take all the way through until the end of August next year, so you can really play with this technology, you can really experiment safely with this technology. But also our people — professional services, customer success, account executives, executive sponsors, partners, marketing — will help you adopt, they’ll help you learn.
The AI offering is available to customers running on the Azure-hosted version of Unit4’s older ERP software as well as those on its cloud-native ERPx platform. Once the free access period ends, participating ERPx customers will pay 20% more than the base subscription for the AI features, while a 5% uplift will apply to those still on the Azure-hosted ERP CR. Paris says that “around 70%” of its agents work with both platforms, with the remainder solely available on ERPx, along with more advanced reasoning capabilities.
European LLMs
The decision to source from European Large Language Model (LLM) providers is designed to appeal to customers with data sovereignty concerns, particularly in the public sector. It also aligns with Paris’s ambition to build Unit4 into a European software champion that continues to grow its customer base in the region. He elaborates:
We’re trying to make sure we remain agnostic across the different Large Language Models. We’re also trying to make sure that we build some European partnerships in there with people like Cohere and Mistral and others. Back to this notion of building the European software champion…
It’s not just around the model, it’s also around the infrastructure location, the data residency, the jurisdiction, the lack of a back door, the fact that no one outside of the EU or the UK can have access to the data. It’s quite a sophisticated topic, and obviously we’re compliant with the EU AI Data Act, and so on. But we do want to make sure that there’s this sovereign answer for customers who so choose it, in particular, obviously, public service.
The vendor’s been putting a lot of work into building what it calls a business context layer — what diginomica calls a a System of Knowledge — to tailor its AI capabilities to the specific industries, geographies and business requirements of its customers. Claus Jepsen, Unit4’s CTO, says:
AI will only be effective for midmarket organizations if it is embedded in an intelligent core that is fluent in the language of our customers’ industries.
With customers able to interact with the ERP system conversationally through applications such as Microsoft Teams, that business context layer becomes the main differentiator for an ERP vendor, says Jepsen:
What is a general ledger? What is a procurement? What is the sales order? What are these things, and how do they relate to each other? That’s the business context piece. We strongly believe that the value moves from the transactional into the meaning. The one who can provide the best knowledge about your business is the one who’s going to have the most value to offer to customers going forward, and if you don’t do that, somebody else would do it. You’ll still be around, but you’ll just be a database with an MCP in front of it, which is not kind of interesting. It’s still valuable, but it’s not differentiating.
Context archeology
Much of that contextual understanding is implicit in the data model and business logic already built into the ERP software, but more work has been needed to make it intelligible when passing instructions to a general-purpose Large Language Model (LLM) to perform specific tasks or queries. Jepsen explains:
Already with ERPx, we start building this very rich description of all the data. All the objects in our system have been described through semantics and an ontology. Then on top of that, we have now layered in all the business processes in a format that allows an LLM to actually reason about it… We have literally gone out and done a huge archeological piece of work to fundamentally understand, why is the ERP system doing as it is doing? Because everybody can look at an ERP system and we can see what it’s doing, but very few understand why is it actually doing what it’s doing, what was the reason we produced this?
So we refactored all that back into a corpus of information as well, and that is what allows the LLMs to actually create automation, because they understand — or they can deduce based on this corpus of information — what is the next step in the workflow.
There’s also a control plane and various checks that the vendor has built in to ensure compliance with permissions, authorizations, regulatory frameworks and transactional integrity. He comments:
Confidence is not the same as correctness… You need something to check that what you’re confident about is actually also correct.
Unit4 says that its AI will also learn over time and adapt to the specific context and processes of each individual customer’s business and operations.
My take
There’s increasing recognition that LLMs depend on accurate context to be able function effectively in the enterprise. This gives established application vendors a clear advantage, provided they’re able to harness all of the knowledge they’ve built up over the years of the industries and customers they serve. Unit4 appear to have done this work and now the next challenge is driving customer adoption and trust in their AI agent offering. Providing free access for as much as a year in the case of those customers that take advantage of the offer straightaway gives the vendor breathing space to achieve that goal.