By Xavier Fernandes, Analytics Director at Metapraxis, the financial analytics firm

2020 has been an unpredictable year, bringing volatility to already uncertain markets. Many businesses are feeling the strain of the effects of the pandemic and have spent the last nine months trying to keep the company going. Not only has it been necessary to adapt and react in real-time to ever-changing circumstances, but companies have also been looking to bolster their finances with an eye to surviving future challenges. With arguably the worst of the economic fallout of Covid-19 still to come, it’s key that business leaders use all the tools and technology at their disposal to deal with the challenges ahead and secure a successful 2021 and beyond. Fortunately, businesses can take the lessons and data from the tough past few months and use that to prosper whilst navigating further market uncertainty before we see calmer waters. One striking lesson is that algorithms which make room for human predictions can far out-perform both AI and human expert-only forecasts.

Changing demand post-COVID

For years, AI has delivered benefits to businesses, including more insightful analysis to support management decisions, creating better customer experiences, and reducing costs by automating time consuming processes. Yet, despite these benefits, many firms have still been hesitant even just to run trials let alone implement production capabilities. Since the crisis hit, many of these businesses have realised that the benefits far outweigh their reservations. They want to be better able to mitigate black swan-type events, and they want access to real-time, higher-quality management information, and therefore the ability to make better and faster decisions.

But whilst interest has increased, many firms are still not confident in the capabilities of AI tools, or don’t understand what data is needed for successful implementation. There is certainly more awareness that most AI algorithms are trained on past performance of the business – and the more historical data there is, the better those algorithms can perform. There’s also growing understanding that the predictions of algorithms are just guidance and should still be questioned as to how precise the forecasts are. But with this growing understanding can come new doubts for AI-wary businesses: if the algorithms are trained on historical business patterns, what happens when those business ‘norms’ suffer shock scenarios?

Firms are right to want to know how AI tools will help them produce effective forecasts for a whole range of scenarios and help them better deal with unpredictable situations – like the pandemic and its aftermath.

What can AI offer?

Xavier Fernandes

Historically, AI has been particularly useful for supporting tasks with repetitive activity, for example, performing financial checks and analysing large sets of data. AI performs particularly well within this context, spotting outliers before a human expert would notice them and allowing impending problems to be flagged to avoid costly mistakes.

By doing so, it can also be used to predict existing customers’ future needs as well as track trends in their financial circumstances, supercharging the ability to cross and up-sell products. With potential benefits like these on offer, management teams of innovative businesses can rely on AI to help them with some of the heavy lifting of predictive and prescriptive analysis.

Whilst those advantages are always helpful, the most relevant advantage AI can offer in the aftermath of Covid-19 is scenario modelling. Using advanced data capabilities and learned behaviours, AI can analyse market trends and historical business performance in relation to external factors to provide early warning forecasts of future performance. These insights can be truly invaluable in our current climate, allowing management teams to modify direction early and take calibrated small steps to mitigate any risks and avoid management later feeling cornered into taking drastic and potentially further-destabilising action.

Helping with the COVID aftermath

Clearly, in uncertain times, a tool that allows CFOs to see how certain factors would impact performance is invaluable. But to utilise it properly, businesses must understand what the technology needs in order to produce accurate, useful forecasts.

The pandemic was such a rare event that businesses did not have any previous data on the impact of anything remotely similar. This meant that many scenario modelling tools struggled to accurately predict the actual business outcomes because they had no relevant historical data to utilise. Now, as long as businesses have collated relevant data from 2020, there is 9 months’ worth of ‘pandemic’ business performance data to analyse and train algorithms, meaning forecasts can be generated to model future shock events.

It should be noted that algorithms like large amounts of data – and nine months isn’t necessarily that long in the context of business performance. Therefore, larger data sets may need to be synthesised with some data science wizardry from the ‘pandemic’ dataset and the historical business-as-usual datasets.

It must also be remembered that the tool can only work if the previous information is representative of business decisions and outcomes, so all relevant factors must be taken into consideration. For example, if the tool is basing its decisions on data from the UK and on a certain product, it should not be assumed it will give accurate business forecasts for the US market with a different product.

There is still one massive hurdle that typical AI-generated algorithms face – they generally fare poorly if the information they have on the market no longer represents, or encodes, the drivers of that market. This is where humans’ amazing ability to synthesise all sorts of different inputs into a decision can be applied to great advantage.

Human-AI collaboration

We often hear talk about how algorithms can augment human capabilities. Indeed, the ability of algorithms to free up the business from the soul-destroying task of creating monthly forecasts and instead focus its energy on achieving better company performance has been a game-changer for many organisations. Some of the best-performing of those algorithms take proxies for human forecasts – say sales pipelines or raw material orders – and then correct for the human biases they have discovered by analysing historical performance.

What we saw in the immediate aftermath of the lockdowns was that where algorithms incorporated such proxies, they adapted swiftly to the changed market situations despite not having any equivalent situations in the historical data they utilise.

So, whilst we often think of algorithms augmenting human capabilities, human ability to assess the impacts of a black swan event can effectively augment algorithms’ capabilities resulting in agile, trusted forecasts which would have been unachievable by algorithms or humans alone.

Such human-AI collaboration will be critical for many businesses in helping them navigate their way out of the pandemic and through the economic uncertainty that still awaits.