Author: Andrew Burgess, Adviser on AI in business, AJBurgess Ltd
Is artificial intelligence only useful to big corporates, or can ‘ordinary’ businesses exploit this most disruptive of technologies? Andrew Burgess explores the key considerations medium-sized businesses need to think about if they are to maximise value from AI. He looks at the different sorts of data that can be exploited and the risks that need to be understood and mitigated. AI is not just for companies with thousands of staff and huge budgets – as long as the data is plentiful and of the right sort, then every sort of company should consider how AI can transform their business.
There are many examples of companies adopting artificial intelligence technologies to enhance or transform their businesses. Deutsche Bank, for example, use AI-enabled speech recognition to listen to every recording, rather than just a small sample, of their dealers’ calls with clients to identify potential cases of non-compliance or fraud. Google have used it to reduce the cooling requirements of their data centres by 40%. Virgin Trains use AI to read and classify all of their incoming emails so that they can be processed much faster and more effectively. And Paypal uses AI to identify fraudulent transactions almost in real time - they now have a fraud rate of just 0.32% compared to the industry average of 1.32%.
But all of these examples are from very big companies - ones with thousands of staff and multi-million pound turnovers. So, is it the case that AI is only applicable to the big corporates who have the scale and money to deploy this new technology? What about the medium sized enterprises that make up the bulk of the businesses in the UK? Should they just forget about AI as something that is only for the big guns? The short answer to these questions is no: AI can be very relevant and applicable to SMEs. But, as always, there are some caveats to this, and some quite specific actions that they need to take so that they can exploit AI to the full.
Don’t believe the hype
The first and probably most important thing to do is not to believe all of the hype. The excitement around AI is a double-edged sword - it certainly gets people’s attention, but can also encourage over-inflated expectations. Most of what is written in the press and journals concerns work that is being done in labs and universities - all very exciting but not yet relevant to the business world. And those businesses that are doing interesting stuff with AI tend to keep it to themselves so they don’t give away any competitive advantage. The worst case is where companies claim to be doing something with AI but are stretching its definition so they can look ‘cool’ and interesting. So, it’s tricky to get a good feel with what is really happening with AI in the business world, unless you speak directly to those companies, or bring in external advice. As a general rule it is best to water down anything that is written, and assume that the benefits are less than claimed and the risks slightly higher. And one can almost guarantee that it will have been more difficult to implement than anyone claims.
Understand its purpose
The second thing to think about when considering AI is to understand your AI ambitions. There are those who simply want to ‘tick the AI box’ and will do the minimum amount of work to achieve that. There is nothing inherently wrong in this approach - the marketing opportunities are usually worthwhile - but it won’t fundamentally change your business. The first meaningful stage of AI adoption is to use it to improve specific processes or functions; make them faster, more efficient or more accurate. AI has some good capabilities here but, for a business without scale, the ROI can be less than desired. SMEs therefore should really look at how AI can transform a whole function or department, or even start to transform the whole business. This could open up opportunities to offer new services or products to your customers and accelerate growth into new areas. There is no right and wrong answer here - each company will have different ambitions. The important thing is to have a good idea what yours are before you start.
The case for SME adoption of AI
The reason that SMEs don’t usually consider AI and think it is only relevant to big corporates is because they don’t have thousands of employees, and therefore the opportunities for savings will be limited. There are two important points to understand from this that should hopefully dispel that myth. The first is that AI is not really about delivering efficiency savings. Although it can do a pretty good job of it, efficiency savings can generally be left to the more traditional automation technologies such as Robotic Process Automation. Rather than replacing people, AI is actually really good at augmenting their capabilities - making people better informed, and helping them make better decisions through delivering insights and value from the data that already exists in the organisation.
Which brings me onto the second important aspect to understand about implementing AI in business: it’s not about the people, it’s about the data. So, it’s quite reasonable for a company with only a handful of employees to extract huge amounts of value from AI as long as they have plenty of relevant data available. But what is ‘relevant data’? There are, of course, the obvious examples of big databases of structured data - AI is adept at extracting insights from these data sets that might have been impossible for a human to comprehend. It can look for correlations between different data sets to identify patterns and root causes that can provide information for targeted marketing campaigns, reducing customer churn, recommending up-sell opportunities or creating optimised preventative maintenance schedules. The value in this sort of data can also create completely new revenue streams in their own right, such as from targeted advertising.
But the data doesn’t have to be just numbers - it could also be images or sounds. AI, and deep learning specifically, is very good at turning these unstructured data sources into structured data - identifying what is in pictures, or what words are being spoken. Image recognition can be used to identify objects on production lines, or recognise faces or predict activity from satellite images of ports and car parks. Speech recognition can be used for voice activation or, as in the Deutsche Bank case cited at the beginning of this piece, for identifying risks or opportunities in telephone calls.
Another type of unstructured data source is documents. Unless these are very structured tabular documents that OCR can read and extract the information, then AI needs to be called upon to ‘read’ what is there and turn it into something structured and useful. Even invoices, which can be considered as semi-structured data because the information is almost the same but can vary quite a bit from document to document, are good candidates for AI; rather than create a new template for each type of invoice that needs to be processed, the AI is trained on a sample so that it can cope with any subsequent differences. If the difference is too large, and the AI is not confident enough, then it can escalate that to a human - the AI will then learn from this interaction and become better trained. The Virgin Trains example given earlier on, uses this type of AI capability (called Natural Language Processing) on free-form emails to classify them, but even in smaller business as long as there is a reasonable volume of documents (emails, invoices, etc) then AI can probably add value here.
So the focus for any AI opportunity should be on the data that is available, and identifying the appropriate AI capability to extract that value. And that could mean drawing on a number of different capabilities to create an overall solution. Chatbots are a good example of this - a good solution may have speech recognition to provide the inputs and then a Natural Language Processing engine to understand the actual intent of the words, and finally a reasoning capability to provide an appropriate answer. Chatbots are actually a very popular approach for many businesses to enhance their interactions with their customers. But they can also be a very good way of quickly destroying your reputation if they are done badly. Chatbots are useful to highlight some of the risks of AI if it is not approached in the right way. Chatbots work best when they are focused on narrow and specific capabilities, such as processing customer orders - the more you try and get a chatbot to do, the more difficult it gets to provide a good service, and then it usually takes only one bad chatbot experience for the customer’s trust to completely evaporate.
Other risks that SME businesses need to look out for are very similar to those for the big corporates. Eliminating bias in any source data is very important if the outcome from the process impacts people’s lives, such as recruitment or loan approvals. For regulated businesses, the need for explainability is crucial, but some AI algorithms can be ‘black boxes’, so these need to be selected carefully. And if the AI capability being developed is going to be the basis for the whole business model, then smaller businesses in particular need to be careful they don’t become over-dependent on the system, especially if the model’s decision-making process is not transparent.
So, AI is not just for the big corporates to exploit. There are certainly opportunities for smaller businesses if they have enough of the right sort of data available to them and can mitigate the risks satisfactorily. Understanding your AI ambition and matching those to your data sources and the available AI capabilities is the first step in the journey to exploit those promised benefits of AI.
Andrew Burgess provides insight and thought leadership on the practical application of Artificial Intelligence in business, and is the author of 'The Executive Guide to Artificial Intelligence' (Palgrave MacMillan, 2018).