There is no denying the impact artificial intelligence (AI) – in particular, generative AI (GenAI) – is having on business applications. In many ways, the intensity of industry excitement over AI suggests the technology is at the peak of the Gartner hype cycle, which means businesses are about to realise how complex doing AI well really is.

The challenge for many is finding use cases that can deliver a fast return on investment and scale beyond a pilot roll-out cost-effectively. This is particularly relevant when deploying applications that use large language models (LLMs) as costs can quickly escalate.

From what Computer Weekly is seeing, there appears to be a shift in business away from public LLMs to small language models and AI models that can be tuned and run on-premise.

A key message coming from industry experts with regards to training AI using internal data sources is that the data needs to be of high quality. Enterprise data is often spread across multiple applications and there is generally no single version of the truth. Trying to consolidate data into one store can be a challenge as applications are owned by different areas of the business and a single data architecture may not seem like a high priority.

AI is often linked to enterprise resource planning (ERP) providers’ strategies to migrate customers onto their cloud services. This can be a challenge for many organisations, which tend to run heavily customised versions of an ERP system. Migrating these customisations is extremely costly, even when the project goal is not to customise the software. For instance, when Birmingham City Council began a migration from an SAP ERP system it had been running for years, the project to move to Oracle cost the council millions more than had been budgeted for. The migration also needed manual steps, which prevented the council from completing its annual financial report, leaving a gaping hole in its budget.

One of the issues IT leaders face during a multi-year ERP upgrade project is that the software acquired at the start of the project may not be what is needed a year or so later. IT leaders have always needed to deal with shelfware. Sometimes, software providers offer tempting discounts on product bundles. However, while the purchase price may be tempting, ongoing software maintenance can quickly erode these savings.

Third-party support providers can offer a way to reduce these ongoing costs and help IT departments keep older enterprise software platforms running for longer. In effect, the third party software support provider has access to any of the patches and support documentation a customer has downloaded, before they switch over. After switching, subsequent patches and support are no longer available from the software provider. However, given that many of the products that move to third-party support have been run for several years, entirely new issues that require code changes are unlikely.

Here are Computer Weekly’s top 10 business applications articles this year.