This is a guest post for the Computer Weekly Developer Network written by Claus Jepsen in his role as CTO of Unit4, an enterprise software company that offers a suite of functionalities spanning finance management, accounting, ERP, FP&A, HCM and talent management modules.
Jepsen writes as follows…
There is a lot of excitement about technology’s potential to automate tasks making our lives easier and freeing us up to do more fun things.
ChatGPT is the latest in a long line of excited debates – with the CEO of OpenAI now suggesting that his ambition for the technology is for it to become a ‘reasoning engine’, so therefore not just a fact database. But the problem is (and this has been highlighted through various tests and articles) that implementing automation, never mind autonomy, is really quite hard.
It may create interesting experimental tools for consumers, but there are a couple of key considerations.
Firstly, automation will only ever be as good as the available data. In the enterprise tech world, that depends a lot on being able to break down silos between data stores. If an organisation does not have access to a clean, structured data source it will be difficult to make the automation work.
Secondly, it does depend on us humans and all our foibles. We’ve seen how unconscious bias in programming has undermined attempts at artificial intelligence, but the same principles can affect automation in enterprise software.
As IT teams begin to understand why they want to automate business processes and how to go about it, understanding the data fundamentals is crucial: how does an organisation get at the underlying data? What data model is going to be used? How can the data model be made immutable?
Planning the path to automation can produce additional challenges.
Business processes are different depending on the organisation and whether or not the IT team has built up the architecture over many years i.e. there are likely to be different layers of legacy applications with customisations. Automation works best if organisations use immutable data models. This allows the application to trawl as much data as possible from across the enterprise and analyze it in a consistent way.
If automation is to be successful it means putting constraints around the level of configurability that customers have within their IT environments. It means customers must accept more vanilla processes, which may require a change management process to be implemented. Some of this can be overcome by adopting a vertical model for functionality, which will enable IT teams to build automation into applications in an agile manner.
Applying vertical functionality
Given these challenges, it makes more sense for technologies such as Robotic Process Automation (RPA) to sit on top of existing enterprise applications. Low-code orchestration tools allow organisations to integrate RPA functionality with core systems and interact more easily with the data in core systems.
Overall, we are some way off ChatGPT taking over the functions of an Enterprise Resource Planning (ERP) system. For one thing, that level of sophisticated AI is not necessary for a workflow such as evaluating an employee’s holiday accruals. Today, our HCM application can analyse data in the background and push notifications to an employee informing them that they have outstanding annual leave. The conversation between the employee and the HCM system is enhanced using (in our case) an AI-enabled digital assistant, Wanda, but the core instructions can be fulfilled using a standard Yes/No workflow process.
Over time there will be more and more automation in enterprise software. We see a future of 10-second experiences, where automation will simplify user interactions with ERP systems to such a degree, employees barely notice they are interacting with such back-office applications.
However, ChatGPT isn’t going to become sentient any time soon and replace existing enterprise workflows. For one thing, we don’t fully understand how the human brain works, so how could we possibly replicate that with artificial general intelligence?
In truth, the major opportunity for automation will continue at the edge, where we will create task-focused apps that can leverage the vast amounts of data within core systems to improve the user experience.
Of course, this will raise questions around how much intelligence is required to reside at the edge for the application to act autonomously and it will also be important to monitor latency issues to ensure real-time decision-making. However, if the core enterprise applications have been designed to be automated from the bottom up, open APIs and low-code tools can enable developers to integrate autonomous functionality quickly. That will mean 10-second experiences will become the norm for most enterprise application users.