Fleet AI data needs to be accurate or risk strategic errors
Fleet data uploaded to artificial intelligence (AI) tools needs to be highly accurate or risk strategic errors, says FleetCheck.
Peter Golding, CEO at the fleet software specialist, explained the longstanding IT maxim of ‘garbage in, garbage out’ applied as much to the new technology as any other.
He said, ‘We are encountering some instances where fleets are starting to experiment with AI, which is very much to be encouraged in search of new efficiencies, but the data they are using is fundamentally flawed.
‘At best, this means the output is nonsense and will be ignored by the fleet manager but, at worst, can lead to strategic errors that are expensive and even chaotic.’
He pointed to a discussion at the recent Association of Fleet Professionals Conference where three types of uses of AI were identified – intelligent automation covering repetitive tasks, predictive AI using historical and real time data to identifies patterns, and generative AI that uses data to create recommendations.
‘The effectiveness of AI in fleet situations, especially the predicative and generative applications, depend on your information being reliable, otherwise the whole exercise becomes potentially nonsensical.’
A key problem, he said, was the analysis produced by AI of any garbage data would still deliver highly authoritative sounding advice.
‘The AI has no way of knowing your data is poor and will respond to prompts in exactly the same way as if the numbers are solid,’ he added. ‘It is unlikely it will look at a fuel expenditure upload, for example, and tell you it looks flaky. The technology effectively trusts you to get it right.
‘However, it is also worth remembering that large language model are passive – they are designed to do what you ask, so, if you are unsure about data, it can be a good exercise to pose clarifying questions about anything they are unsure about. In many cases, this can produce a much better final result.’
A further issue with AI output, he said added, is the tone of its output will always appear convincing on the screen.
‘There is a temptation to tend to take it seriously for this reason – but that confidence is simply a product of the way these tools are coded, not the quality of data,’ added Peter. ‘If you then take strategic decisions based on its guidance, problems can obviously follow.’
The answer was simply to ensure that the data you were collecting about your fleet was precise.
‘It is always been critical that your fleet information is dependable but AI arguably makes it more important than ever. These tools have significant potential, and we are increasingly using them at FleetCheck, but their limitations have to be recognised,’ he concluded.
