More companies are claiming their products and services use Artificial Intelligence (AI) than ever before. But Apptus’s Andrew Fowler argues that many of them are not only kidding themselves, but misleading potential customers and investors.
There are many claims made about AI improving our lives and businesses. They include, but are by no means limited to:
Doing the heavy lifting. AI can, in an instant, churn through and act on millions of records of data - a task that a human simply would not have the time to do. In so doing, it dramatically improves efficiency and throughput.
Freeing up humans to do what they do best. By removing the drudge work, people are left to do the strategic, creative and enjoyable tasks.
Making rapid decisions based on masses of evidence in milliseconds rather than weeks, to enhance trend sensitivity and agility.
You may be familiar with some of the applications already using AI, such as:
The benefits are, then, substantial and that’s before the possibilities opened up by autonomous vehicles and medical applications are considered.
Now, of course, there are e-commerce merchandising tools that ensure the right products are in the right place at the right time to ensure the right people buy them and are delighted, but are they using AI?
“Some companies are simply lying, claiming their technologies are rooted in AI when, in fact, they are based on a series of complex, manually generated rules masquerading as AI.”
But all is not as it seems. The truth is, the promises and known benefits are so great that, for many vendors, the temptation to add those two letters - A and I - to their proposition is too great to resist. The belief is that it adds tremendous value to their perceived offering.
But some companies are simply not telling the truth.
Those companies may claim their technologies are rooted in AI but, in fact, they are simply a series of complex, manually generated rules masquerading as AI. Worse still, there are cases where there is technology not even pretending to be AI, it just isn’t there.
As someone who has worked for nearly a decade with a company whose entire offering is built on AI and its component technologies, I have long suspected that some competitors are stretching the AI definition for reasons already established. But to call it out would have the hallmark of sour grapes or just pure malice. So I’ll pass the buck.
Indeed, a report by MMC Ventures published in February this year, tells the story of this ‘Big Lie’ better than I could. It examined 2830 businesses claiming to be “AI Startups”, and the findings could not be more clear. Two in five, or 1250, offered no evidence that AI was in any way material to their company’s value proposition.
A Forbes magazine report also spelled out the stark truth:
“We looked at every company, their materials, their product, the website and product documents,” says David Kelnar, head of research for MMC, which has £300 million ($400 million) under management and a portfolio of 34 companies. “In 40% of cases we could find no mention of evidence of AI.” In such cases, he added, “companies that people assume and think are AI companies are probably not.”
So how do we solve a problem like AI? How do you know if you are being sold duck eggs or a truly business changing solution?
I think the answer lies in going back to the basics, to asking: “what is AI?” and, crucially: “what AI is not?”
Many systems described as AI are, in fact, rule-based. They require intervention to set up triggers based on criteria defined after poring over spreadsheets and other data sources. But rules can conflict, they are not truly scalable, they do not react in real time and creating them requires a lot of manual effort – what’s the point if you are already doing that? This is the antithesis of AI.
What’s more, many e-commerce solutions deal with just one part of a site – search or recommendations or onsite banner ads or navigation. This means retailers can have up to four systems in effect cannibalising promotions from each other as they each run using independent logic and data sets.
For example, a shopper searches for an item they know they want, they get the result but the recommendation shows them a lower margin alternative. In effect, this is the opposite of merchandising - directing willing buyers away from high value purchases to lower margin alternatives. Crazy.
So, AI is not a set of rules set by humans that a computer must follow.
AI, in my view, is game-changing optimisation using machine learning and predictive analytics techniques.
A true AI solution is self-learning, self-operating, consistent, accurate, scalable and efficient. It will run across all aspects of a site, learning and applying that knowledge to every pixel of online real estate.
In fact, as data accumulates it just gets better and better.
Every transaction is a calculated experiment using sophisticated, predictive analytics to model the likely effect of displaying certain items in a certain order, then learning, adapting and optimising in real time as further interactions take place.
Crucially it should require little if any intervention other than a ‘steer’ on overall objectives, be they to maximise conversion, profit or revenue.
“By far, the greatest danger of Artificial Intelligence is that people conclude too early that they understand it.”
In determining whether a solution is using AI, machine learning and predictive analytics it’s worth asking these 5 questions of potential vendors:
How much intervention is required to manage the system? The more intervention, the less likely it is that you will realise the benefits of AI.
What facets does the system control? If it is limited, the risk of cannibalisation between competing data sets is greatly increased and at the very least, the potential that AI offers is severely limited.
Can the system respond in real-time to changes in consumer buying patterns? If it cannot, you’ll still be relying on inefficient human effort and trend sensitivity will be compromised.
How does the solution’s machine learning use algorithms to make sense of historical data? Not all data is helpful, so relying on a system that does not learn the difference leaves you running the risk of being stuck in a ‘junk in, junk out’ paradigm.
What business objectives does the system prioritise? A system focused solely on conversion could leave you shifting lots of units, but realising minimal margin. That’s fine if you just want to run down inventory, but what about profit and revenue? You need the ability to control these high level priorities.
Eliezer Yedowsky, writer and co-founder of the Machine Intelligence Research Institute, is quoted as saying: “By far, the greatest danger of Artificial Intelligence is that people conclude too early that they understand it.”
Don’t fall for the Big AI Lie. Those two letters, AI, are not necessarily a silver bullet, it is what’s under the hood that matters, not what’s on the badge.
So, truly realising the transformative effects of AI on metrics like sales, profit and conversion requires thorough investigation of any tools and technologies promising to deliver it. And,
as Yedowski intimates, that means more than a superficial understanding of AI - it means taking a deep-dive into AI and e-commerce merchandising to ensure that you:
To help you do just that, we have worked with the digital marketing advice company Smart Insights to create a report that sets out the facts without the usual marketing hyperbole. I hope it will give you that knowledge and confidence you need to take an educated view of the market and make your first steps into AI, surefooted and based on a clear business case.