The Criticality of Adopting AI

If you are reading this post, you may be looking for answers about why adopting AI is so hard or how to do it. Many people have written about these topics. Here are two articles that have been published recently — why it’s hard (Forbes) and how to do it (Harvard Business Review). But dig around and you’ll find plenty of others opining on these issues with greater or lesser depth and clarity.

This article is (mostly) not about those things. It’s about why developing an AI mindset and adopting AI techniques is important to the future success of your business and businesses more generally, especially those participating in larger industry ecosystems. Even if you’re a small company, not a technology company, working in the services sector, or on the periphery of your industry, there are many tools and processes in the AI toolkit that can accelerate your business and not break it, either functionally or philosophically. AI will put you ahead of your competitors who are slower to adopt, and it will strengthen your market segment.

What is AI (through the lens of business)?

Artificial intelligence is a general category of technologies and processes that augment humans. (IMO, and others agree, it should be called Augmented Intelligence because that would actually be more accurate and less intimidating sounding than Artificial Intelligence, which evokes dystopian future states with robot overlords.) At its heart, AI is a bunch of math — statistics principles and techniques that have been around for decades, even centuries, that take data and find patterns in order to make generalizations. With the rise of fast, cheap computing and the availability of massive sets of data, new life has been breathed into these techniques, and they have been applied to larger and larger, but fundamentally similar, computations over the past several years.

In the context of a business, these techniques help you solve two types of problems: categorization and prediction. The categorization side of AI is used mostly to identify words and images, such as recognizing a voice command or spotting a potentially cancerous tumor on an X-ray. The prediction side of AI is used to develop models of system behaviors that can then be used to take or recommend actions, such as routing you around traffic or recommending a movie.

Together or separately, categorization and prediction let you automate processes that you might do manually or not at all today. This means that AI can help you accelerate, scale, streamline or simplify work you already do; or it might offer totally new opportunities for your business that you never considered doing before. AIs can either make decisions for you when things are really well understood and very predictable, or they can aid or accelerate decisions when things aren’t.

Let’s take the example of recommending a movie. If you remember going to the video rental store, and you were a frequent visitor, you might also have asked the clerk for recommendations about movies to rent. As a local, small scale process, this worked fine. An attentive clerk might even have gotten better and better at making recommendations for you over time. Enter the internet, streaming services, and the ready availability of virtually every form of filmed entertainment ever made, and the clerk-based recommendation model clearly doesn’t work. Using the data collected from all viewers on a platform, including crowd-sourced recommendations, the process becomes automated and scalable — maybe it’s not perfect, but the video store clerk wasn’t perfect either.

Why should I care about AI (if I work in a business)?

Even if you don’t run a large scale platform like a streaming video service, there are actions to be taken and benefits to be derived from adopting an AI mindset. And, the steps that precede actually applying machine learning methods to your business data are as important to developing a roadmap for AI at your company as performing the eventual analyses and deriving new AI-generated insights. These steps are good practices anyway, and in taking them you will improve your business — while also putting you in a good position to adopt AI and participate with other companies in their AI activities.

Creating a Data Inventory — All AI involves operating on data, typically huge sets of so-called big data, which are so large, complex and unwieldy that they really can’t be manipulated or analyzed by hand. So, you need to take a hard look at your data: the data you have, the data you don’t have, but wish you did, the data you might be able to get from partners or open sources, and how it all goes together (or doesn’t) when you’re trying to solve problems. This data inventory will give you insight into the opportunities and gaps in your company. Though this needn’t be the first step in starting an AI initiative at your company, it can be. It also doesn’t need to be carried out by a data scientist; a business analyst will do. Don’t be surprised if the inventory confirms what you may already know — that your data is a mess — and don’t leave anything out.

Reviewing your Business Processes — Your processes define your business, and you probably have many more of them than are documented or any one person in your company can describe. From the subtle (how you develop a sales lead) to the mundane (how you manage inventory), each process is worth describing and critically examining to see if it would benefit from acceleration, scaling, streamlining or simplification. Further, like the data inventory, this activity may reveal potential opportunities within your own business, or in work you do together with partners, as part of a large supply chain or market ecosystem, or outside your market segment altogether. If you carry out a process really, really well, and it’s a hallmark of how and why you are successful, it may be something you can generalize and codify for other similar types of businesses, and sell as a new service.

Starting Somewhere, Starting Small — Many large companies have embraced AI by scooping up all the data science talent in the country and putting them to work mining their massive troves of data. If you’re not in one of these companies — no big-budget AI initiatives, no resources to fight for data science talent, no massive troves of data — then you’re in the majority! In the United States, most people (99.7%) work for small businesses (companies with 500 or fewer employees), they account for 64% of new job creation, and they produce more patents per employee than large firms (16 times more). That last statistic is especially important. Small businesses innovate at a much higher rate than large ones. This can be you. Pick a small project that involves categorization or prediction, and then use it to exercise your data, process, and analysis pathways. Then build upon that success and expand to larger and more impactful initiatives.

You can also take these steps either separately or concurrently:

• Invest in Software — There are numerous software tools and services that you can use to apply AI to your business. If you have a lot of data but you’re not sure what to do with it, you can even take a “shoebox” approach, named after-tax accountants of yore who had to make sense of “shoeboxes” of bills, receipts, and statements. Further, sparse data AIs are being developed and refined all the time, to help solve problems where large data sets don’t exist or have lots of holes.
• Get Educated — My team at Riff Learning recently delivered an online course on AI adoption for NEXT Canada, helping people learn about implementing AI-inspired business ideas in their companies or spin them out as new companies. The course simultaneously taught AI principles (led by Sandy Pentland), basic Python coding for machine learning, venture formation, and business-building techniques. The course also used the Riff platform’s AI to help small teams work more collaboratively and effectively. By the end of the course, each learner had hands-on experience coming up with and developing a real AI-based initiative.

Maybe these activities don’t feel like AI. They aren’t. But, they are necessary precursors to implementing AI in a way that will actually have an impact on your business.

What has AI done lately (for businesses like mine)?

You may have heard the phrase “AI is data-hungry,” meaning that it needs lots and lots of data to be good at classification or prediction. For example, if you want a machine to identify an armadillo from a photograph, then it needs lots of examples of armadillo photographs to do that reliably. It also needs to be trained to recognize the most important factors that make an armadillo different from, say, a pangolin, a hedgehog, an anteater, or a snake that has just eaten a small mammal. This process of training an AI is called supervised learning, and it’s how natural language processing (your phone recognizing what you’re saying), image recognition (your search engine finding pictures of snails), and other categorization problems have been solved. Unsupervised learning also requires lots of data, but it doesn’t need training to be effective. It works by sorting things into clusters that have common attributes, which results in unlabeled, but categorized data. Transfer learning takes a model developed on one kind of data and uses it to try to sort a similar, but different set of data. And model-based AI begins with a guess about what pattern, structure or attributes a set of data might have and then keeps testing that model with more and more data and refining the model as it goes.

Why do these distinctions matter? Because your categorization and prediction challenges need to fit a certain class of specific problems that AI is good at solving; otherwise, AI is not for you. The good news is that LOTS of problems fit into this class; the even better news is that LOTS of software tools exist for applying machine learning to these types of problems, even when your data isn’t super clean and tidy. Increasingly, new platforms, like Flybits and Endor, make it easier than ever to take your data and figure out what can be known about it.

Can’t I just wait a bit to adopt AI (in my business)?

Um, no. While AIs may still be in their infancy and adopting AI entails some risk, there is an almost unlimited number of interesting problems that machine learning and predictive analytics can solve in your business. For example, look at any manual process you carry out today and decide whether it’s predictable or codifiable — and then decide how many person-hours you spend every week carrying out that process. This is known as process automation, and when AI is used to drive the process automation cycle of prediction, recommendation, action, and measurement, it’s called robotic process automation.

What are the risks? It has been widely reported that AI suffers from biases, typically introduced by using overly narrow data during supervised learning. Your video clerk recommending a title probably had his biases, too. On a small scale, such biases have little effect, and you’re probably even aware of them when dealing with someone face to face. At a large scale, bias is codified and amplified, and potentially hard to see and even harder to reverse. As you may have read or seen (watch Joy Buolamwini video on how facial recognition fails to properly identify black women as women), such biases often affect certain populations more than others.

Another risk is overreliance. Once an AI has been put in place, it must be maintained, just like any other tool or process. AIs are not static, and only work as totally standalone processes when operating over very simple, very predictable activities, or after numerous rounds of training and validating with new and larger data sets. If you’re using an AI, chances are that the problem is complex enough to warrant your continued monitoring and tinkering.

The recap

AI promises to unlock unprecedented opportunities for businesses worldwide — for small, non-technical companies as well as the large technology firms. With the ability to analyze vast amounts of data and reveal hidden, business-shaping insights, AI can transform your business in subtle and dramatic ways. In manufacturing, AI can help find defects in an assembly line before multiple defective items have been made. And when manufactured parts end up in other products, AI can more accurately predict when a part will fail before it causes bigger problems. In government, AI provides insights about tax settlements, predicting which overdue accounts to pursue to maximize the likelihood of payment. In customer service, AI powers chatbots that field routine questions to establish the category of help needed, and then shuttles callers to the right expert to solve the problem. Large corporations use multiple single-purpose AI-powered applications (often developed by smaller companies) to streamline work, such as connecting people who have similar skills and can help each other, or performing image detection on documents, or translating verbal commands into tasks. Wherever you are on this spectrum of size, opportunity, capabilities or needs, don’t wait to implement AI.