4 minute read
USING AI AND ECONOMICS TO PLAN FOR THE FUTURE
Niels Bohr, the Nobel Prize-winning physicist, allegedly said, “Prediction is very difficult, especially about the future.” What makes forecasting the future particularly difficult is that outcomes are a combination of a number of factors that may combine in complex ways. These include economic trends, consumer behaviour, and random chance; how these impact a specific business, product, customer, or location; but also the beliefs and prejudices of, and the information available to, the person making the predictions.
Usually when we think about predictions we think about ‘big calls’, like flying cars, rather than the myriad of small predictions that businesses need to make on a day-today basis. These small predictions can be for things like “How many people will visit a specific outlet at a specific?”, “What will they consider buying?”, and “How many of the scheduled staff will be absent for a specific shift?”. Improving predictions of these ‘smaller’, but essential, processes can lead to more certainty and better consistency in decision making, adding up to a more effective and profitable business.
Currently, most glances into the future are based on ‘rules of thumb’, backward looking calculations and extrapolations of current trends, or gut feel. Using these types of approaches to make forecasts introduce a set of avoidable errors. Humans are plagued by behavioural biases, including a recency bias where we overvalue things that have happened recently, and a tendency to see patterns where there are none. People struggle to process large amounts of information, and to identify vague but consistent patterns. Predictions and extrapolations which only look backwards can also miss changing economic and social conditions, particularly in cases like the past three years with the Covid-pandemic. These forecast errors filter through to the whole business. Having a bad forecast for your business is like having a bad map – you’ll end up getting lost and wasting time and energy going the wrong way. There is thus a large opportunity for ‘machine-assisted’ approaches to forecasting the future.
Making forecasts for things like revenue, more than three months into the future, can be difficult, but forecasts at this time horizon are often essential for things like budgeting, stock ordering and strategic planning. Longer-term forecast methodologies have been developed and used by institutions like Central Banks to make forecasts for inflation and GDP but these approaches are underutilised by businesses. Over the past year, Predictive Insights has worked on adapting these long-term forecasting approaches to business and budget forecasting. Combining curated real-time data on the economy, with machine learning and behavioural insights can enable restaurants and retailers to more accurately predict demand up to 18 months ahead.
Neil Rankin, CEO at Predictive Insights, said: “Businesses often base forecasts on activity from the same week or period of the previous year. Due to Covid, behavioural changes and other disruptions, this historic data is dirty. Forecasts based only on this data are even more inaccurate now.”
The team of economists and data scientists at Predictive Insights use a combination of data points to improve the accuracy of forecasting including real-time data, transactional data and relevant external data. Through the Predictive Insights machine learning platform, customers are provided with actionable insights for strategic planning, budgeting, demand forecasts, staffing, stock optimisation, lead scoring, promotions, pricing and behavioural nudges.
Long-term forecast accuracy
Over the last year Predictive Insights has implemented this medium-to-long term Demand Forecasting solution at multipleoutlet stores, which often sell through multiple channels, in the United Kingdom, South Africa, Botswana, United Arab Emirates, and Australia. This approach has led to an improvement in branch managers’ predictions by up to 30 percent for forecasts up to 18 months ahead.
These longer-term forecasts can be combined with Predictive Insights shortterm Demand Forecasting solution (providing hourly, daily and weekly forecasts up to eight weeks in advance) which can be used to drive operational aspects like staff scheduling or stock ordering, which also dramatically increase operational efficiency.
Allow the machines to do the heavy lifting
Neil commented on the added benefits: “In addition to improving accuracy, a big bonus for our customers is that machines are doing the heavy lifting. Where data gathering and analysis can be a full-time job, machine learning can help to provide analysts with a baseline for review. This takes the tedious work away, allowing people to add additional insights and do so much more with their data.”