As if the big data hype of the recent years wasn’t enough, artificial intelligence (AI) is now taking over the scene: the media paints pictures of an AI revolution in the consumer world; there are professional AI conferences for every business area; more and more companies now position themselves as AI-based, or have announced their newly-formed AI strategies.
Of course, this has spurred c-level executives across multiple industries to ask themselves this question – “what does the AI hype mean for me and my company?” Yet, for the Chief Financial Officer (CFO) this only translates into serious concern. As so often happened with big data projects, the CFO will wonder whether this new technology will become yet another IT spend that carries no clear financial outcome.
Well, we have some good news. When implemented properly, AI is actually the CFO’s best friend.
Using your data intelligently
Machine learning algorithms are able to recognise underlying patterns in the historical data about certain processes and events, making successful predictions and prescriptions in real-time, on a large scale, with high precision. In practical terms, this opens the possibility for businesses to optimise existing processes; switching from ‘best practice’ systems, reliant upon statistics and human decisions based on experience, to highly precise recommendations based on accurate predictions.
In internet companies, where practical applications of machine learning were developed, they were perfected to do exactly that: learn from vast amounts of data to deliver iterative improvement on every possible task, powering real-time decision making not with simple rules, but with smart algorithms. Online advertising is a great example of this. When online each us is shown an ad that he or she is mostly likely to click on, as predicted from the past behaviour and that of similar users. This model drives direct business results – money earned – and there is no human decision making involved.
But there’s a catch.
As humans, we are used to seeing data as a source of knowledge. This creates a temptation to think of machine learning as a smarter, shinier version of the business intelligence tools already in use. However, expecting machine learning to serve you with dashboards dramatically underplays its power and speed; as well as the usefulness of the data.
This powerful technology is able to recognise all the correlations between thousands of factors, and build very precise prediction or make very efficient operational decisions based on the data. When, instead, it is asked to produce explanations to a level that a human could understand, it results in wasteful simplification.
Just think about the advertising example mentioned earlier. What if, instead of using the algorithms to choose which ad to show, to each specific visitor, at a certain time; it was asked to provide explanation to the marketing managers, so that they could decide themselves? This would be dramatically inefficient, harking back to the era of rough “audience segmentation” instead of personalised targeting.
It works the same way for any other application of AI: you must choose between getting knowledge and getting actual results, and usually for businesses, the latter is much more preferential
Machine learning that pays
Indeed, machine learning is a CFO’s dream tool – a helpful black box that doesn’t require capital investments, but leads to positive results in P&L that same year. Once you grasp the concept, one question is left – where exactly to plug it?
One suggestion to aid this search – let’s call it a “crystal ball factor” – is to think about which of your business processes could bring the biggest gains, if only you could see the future with some extra clarity. Maybe its stock replenishment, and you’d find it useful to know with more precision how many items of each product to store in the warehouse. Or perhaps the factory floor is better suited, and you need to optimise the raw material mix to hit the production target without any extra waste. Established processes that rely on some assessment of the likely future – demand prediction for retail goods or choice of raw material use – are very good candidates for machine learning, as they can quickly bring measurable cost optimisation for a business.
Another thing to think about is the company’s ability to measure the isolated influence of machine learning. Experimentation is a crucial part of applying machine learning as it allows you to estimate and further improve its value – go from 2 to 5 per cent improvement, for example. Once the predictive model is put to practice and integrated in a stable process, with all other factors unchanged, you should be able to measure the actual business metrics achieved.
In some cases not being able to isolate influence can be a major challenge. For example, in retail, if several promotions are running simultaneously it is hard to measure the optimisation effect reached through better targeting for just one of those. However, for other cases it is much more straightforward. For example, with predicting cash demand for ATM, even a historical test can serve for demonstration.
Last but not least make sure that your machine learning project brings its first results quickly. By applying machine learning to an existing, well-established process, where you know exactly what you want to improve, and likely already have enough data – you should see results within a few months, if not weeks.
A misunderstanding of ‘big data’ and its value often leads businesses to place such projects somewhere in between IT and BI: the former to take care of the data storage, and the latter to try and make sense of it. This frequently leaves c-suite, most notably the CFO, disappointed with the return on investment.
With the AI hype in full swing, this is a good chance to embrace the new technology to benefit the business. With every existing process, where historical data is available, measurement is possible, and a KPI is known, machine learning can bring both optimisation and savings. Perhaps, that is why lately CFOs start appearing as main stakeholders of AI projects. Maybe it’s time to include it into your agenda?