Artificial intelligence can’t solve all your business needs...

...but that doesn’t mean you shouldn’t embrace it.

One of the most important questions modern businesses face is whether (and how far) to incorporate artificial intelligence into their business models and processes. The rapid explosion of Large Language Models makes it clear that AI already presents enormous challenges and opportunities in more fields than most of us thought possible.

Retail is no exception, and the question of how to best leverage the possibilities AI offers - from forecasting, merchandise planning and replenishment to marketing and customer service - is one a lot of retailers are asking. For every decision maker excited by the possibilities of cost savings and improved accuracy AI claims to offer there are dozens of employees concerned about what that means for their job security. This concern feeds into a distrust of AI specifically and complex algorithms generally.

"The algorithm’s forecasts are wrong. They don’t match my forecasts."

So goes the familiar exchange with business users when implementing a new forecasting algorithm for a retail client.

The conflict between AI and business users is usually framed in terms of job security – the fear that business leaders see AI as a means to replace employees and ultimately reduce costs. Decision makers can be tempted to disregard concerns from their employees about the very real limitations and drawbacks of AI, but in doing so they fail to recognise the very real concerns that often underlie users’ distrust of AI and opaque algorithms. If I don’t know what factors and assumptions have been used to generate an AI forecast, how can I trust it more than my own forecast?

Ultimately the trust issue boils down to transparency. Forecasts produced by inexperienced team members are rightly interrogated by line managers and senior colleagues – we expect people to show their working so we can understand their thought process and anticipate any factors they’ve failed to consider, offering guidance to improve their skills. It follows that we should be able to apply the same scepticism to AI-generated forecasts, but that’s not always possible with more complex models. A simple, replicable forecasting model is easier to interrogate  - and ultimately, to trust - than an opaque one which could produce a different result every time given the exact same inputs.

AI can’t do the job on its own without human input and guidance; there are just too many unknowns. AI forecasting engines are incredibly good at teasing out patterns in historical data, but they’re not psychic. To reliably predict future performance in a complex retail business, algorithms need a lot more than historical data. Feeding the all-important strategic direction, product knowledge and market trends into an AI forecast model is technically possible, but given how quickly and unpredictably these factors can all change, the returns are arguably not worth the time investment for most businesses. Instead we should focus on implementing robust algorithms that use easy-to-follow logic to pick out the key trends in historical data and produce a reliable base forecast which business users – with all their highly specialised knowledge – can refine to align with business goals.

Even AI chat bots agree that artificial intelligence can’t replace human knowledge. We asked ChatGPT whether AI can produce a retail assortment plan without human input. Here’s part of its answer:

While AI algorithms can analyse vast amounts of data and generate optimized plans based on predefined objectives and constraints, they may lack the nuanced understanding of business goals, market dynamics, and customer preferences that human decision-makers possess.

AI clearly presents some big opportunities for retailers, but it needs to be implemented carefully, with realistic expectations. The challenges for business leaders implementing AI-driven solutions are:

• ‍Don’t expect miracles: understand the limitations of the technology as well as its strengths. This is crucial for defining what value AI can actually add and embedding it successfully within your business processes.

• ‍Engage stakeholders: reassure team members that their unique abilities are valued and that the goal is not to replace them with AI, but to arm them with tools to improve business outcomes whilst making their jobs more rewarding.

• ‍Set guardrails: never let AI loose without oversight. Machine learning will improve the hit-rate of AI forecasts over time, but we should never become complacent or too trusting of their accuracy. Put the right business processes in place to ensure important decisions are being made by the right people, not an algorithm.

These considerations are central to Plan IT Retail’s approach to AI. Clients often come to us excited about the possibilities of AI but unsure exactly what its role in the solution looks like. We see AI as a powerful tool in any retailer's arsenal, not a magic wand that can do the whole job for you. We’re always researching the latest AI forecasting engines and incorporate machine learning into our own algorithms to refine outputs, while remembering the key mantra:

When AI is too complex to understand, business users are unable to overlay their expertise intelligently, resulting in a lack of trust and ultimately failure.