Are you getting multiple emails or calls every day from various companies claiming that their Artificial Intelligence (AI) software will transform your business? No doubt that some of those claims are real; however, an increasing number of those claims are actually false, being made by “AI Pinocchios,” who are seeking to cloak their legacy offerings with fashionable AI jargon.
Artificial Intelligence certainly represents one of the most transformational and impactful opportunities available for consumer commerce businesses such as retailers, restaurants and other similar enterprises. However, it can sometimes be hard to know if you’re buying actual AI software…or just a load of BS!
Simply described, Artificial intelligence (AI) is the ability of a computer program or a machine to think and autonomously learn on its own. AI encapsulates a broad set of technical capabilities designed to provide better outcomes, augmenting what humans would otherwise need to do on their own. More clearly, AI is software that mimics and automates the very tasks that humans used to do exclusively. These include learning, reasoning, problem-solving and even understanding language. Common AI terms that you may have heard include machine learning, deep learning, robotics, natural language processing (NLP), and more.
Most consumer commerce businesses will increasingly depend on AI softwares to optimize growth drivers such as pricing, promotion, assortment management, product placement and channel optimization. They will apply AI to analyze and shape customer behavior with customer segmentation and personalized marketing. And they will use AI to understand customer demand to optimize inventory, improve store operations, and even streamline supply chains.
The best way to cut through all of the BS is to understand what AI can and can’t do, and by knowing how to spot red flags in a vendor’s claims. Here are five simple tips that should help in the evaluation of AI opportunities for your business.
#1 The Vendor Is Unable to Demonstrate Their AI
Just like you wouldn’t buy a car that you couldn’t test drive first, why would you spend hundreds of thousands of dollars on an AI software if you couldn’t see it demonstrated first?
You should be very skeptical if the AI software can’t be demonstrated, much like you’d be skeptical when a magician calls upon a ‘random volunteer’ to help with their magic trick.
Certainly, there are some AI technologies such as machine learning or deep learning that are hard to demonstrate in traditional ways since these engines operate in the background embedded in applications or servers. When this occurs, the vendor should try to demonstrate the outcomes or conclusions of these technologies to illustrate why they performed better than a non-AI solution would perform.
Other AI technologies such as Natural Language Processing (NLP) where the system listens to and understands spoken commands, should be visually and audibly demonstrable by the vendor.
#2 Their AI Seems to be Easily Fooled
What good is Artificial Intelligence if that intelligence can be easily fooled?
From job candidates writing prestigious college names or high-demand skills on their resumes in white text or technical experts strategically adding noise to images to fool image-recognition systems, you should be cautious when relying on the insights of a system that can easily be confused. Specifically, AI can be fooled with ‘adversarial examples,’ or small, intentional input modifications meant to confuse or deceive an AI software.
While facial-recognition AI is generally quite reliable, it could be subject to adversarial attacks from informed malicious actors. This fallacy has interesting implications for the retail industry where facial-recognition AI is increasingly being used to identify known shoplifters or to identify high-value customers and ensuring that they receive personalized, red-carpet treatment when they visit the store.
However, as companies become smarter about AI, unfortunately so too will those who hope to take advantage of these advanced systems. So, it’s quite possible that just as AI facial recognition gets better in the future, shoplifters too will get better at disguising their faces from these systems in ways invisible to the human eye.
#3 Vendor Claims their AI Software Can Predict Behavioral Outcomes
What if you could predict exactly how each of your customers would behave in a certain scenario? Imagine being able to look at specific customer profiles and predict with total accuracy their social behaviors; who will be the most likely to hunt for discounts; who will recommend your products to all their friends; and which loyal shoppers might be motivated by personalized offers.
Can the vendor really predict this behavior, and more importantly, can you shape it? While some companies claim that they can make infallible predictions from large amounts of behavioral data, this claim fundamentally misstates how AI, and machine learning in particular, function. This is because AI can sometimes falter with tasks that require some degree of subjectivity or judgment. Tasks like choosing the best job candidate or predicting the most likely time and place for a crime to occur within a given city may be informed by AI, but the use of AI certainly isn’t infallible.
Think of it this way: if you asked ten intelligent people to choose the best job candidate, would they all agree on the same answer? Almost certainly not! So, without having consistent training data on who is the ‘best’ candidate, AI social judgments won’t always be accurate.
An AI Software can sometimes be a poor predictor of human behavior because of its countless variables and inherent subjectivity. For instance, on a given day, each customer’s purchase decision depends on a myriad of variables, from the ones we often consider like price and placement, to the less obvious variables like weather or their current financial situation. Other factors could weigh in like the time of day or whether they’re making their purchase on desktop or mobile could also influence outcomes. Although AI can give truly remarkable insights about customer trends on a store-specific or for a specific customer segment, AI still needs to evolve more around individualized behavioral predictions, and it will get better over time.
#4 Vendor Claims Their AI Can Work Across Any Kind of Company, Question, or Data Set
According to AI writer and researcher Janelle Shaye, “the narrower the problem, the smarter the AI will seem.” So, when you have an AI company that claims that their system can work accurately for any kind of company, question, or data set (no matter how small or incomplete), you should run for the hills.
At this point in their maturity, AI systems are excellent at narrow, specific tasks, but they aren’t able to answer every question, nor work for every kind of company.
Think of it this way: you can hardly ask an AI software that’s only been trained to detect images of cancerous tumors ‘why’ those tumors occurred, just as you certainly wouldn’t ask this AI system to start identifying images of houses suddenly. Just because Google seems to have the answer to every question, this doesn’t mean that your AI software can be anywhere near as smart.
That’s why when seeking out AI software, savvy buyers will choose AI systems that have been trained to focus on their specific industry and have a sense of what kinds of questions the AI can and can’t answer.
More troubling than using AI designed for a different industry, is being told that an AI system can work with any data. This is because an AI system is only as accurate as the data it uses. For example, consumer-centric businesses like retailers and restaurants hoping for extraordinary AI insights from a very small, inconsistent data set will be very disappointed. Data integrity and the abundance of data are as important today as ever before.
That’s why before choosing an AI-powered system, it’s important to understand whether your organization’s data is relevant to the problems you are trying to solve, whether it’s organized, and of course, whether the sources of your data are reliable and can be trusted. These are just the most basic requirements for great AI analysis.
#5 Their Predictive Insights Show Clear Bias
Making predictions without real AI can be inherently biased. One such example is demand forecasting, where you are trying to anticipate customer demand in revenues, profits and units over a time horizon. With less sophisticated regression-based systems, you only learn from data that you have seen in the past. There will always be anomalous data in the past such as weather, events, promotions, and other factors that may affect future prediction accuracy.
Consider another example: trying to decide whom to promote to manager based on past managerial performance data. The challenge is that a regression-based approach might over-weight certain variables, reflect the biases of those in charge of interpreting the data, or even reflect the social biases revealed in the data itself, such as a bias for hiring managers of a certain race, age, or gender.
True AI allows for far better accuracy because these anomalous events in past data can be included or excluded based on the expected future situation and optimized to include awareness of factors such as future pricing, promotions, and other considerations, including possible external variables. Similarly, social biases can also be accounted for by training AI systems with more even-handed examples, identifying other latent variables tied to areas of bias, or even training systems to identify bias and then removing the main effects and interactions of the biased data. That is to say that accuracy with AI can be much higher and forecasting error can be mitigated through filtering, corrections and model training.
Separating AI Innovators from AI Pinocchios
It has become increasingly difficult to separate true AI innovators from AI Pinocchios. Vendor brochures, websites, and their PowerPoint presentations all sound the same even though the reality may not be so. Ask yourself, how did we go from few AI solutions being available in the market to now having so many in such a short period of time? AI-powered solutions built from the ground up have a real advantage versus legacy platforms. If solutions weren’t truly architected to leverage AI, they may never live up to the promise of AI. And importantly, software companies can’t become AI-powered without having real AI-experts on staff.
It is far easier to claim AI leadership in sales and marketing materials than it is to actually build a true AI-powered platform. AI Pinocchios will be exposed by their customers as their offerings fail to yield the promised benefits. As consumer commerce companies become more experienced evaluating AI, vendors will no longer be able “fake it” until they “make it.”
Meet Hypersonix, Inc. – Real AI, Real Results
Hypersonix, a venture-backed San Jose, California based company that offers a comprehensive cloud-based, AI-powered analytics solution enabling decision-makers to analyze disparate data sources and derive actionable insights, simply and quickly, without needing to depend on IT or analysts being involved. Serving Consumer Commerce industries like Retailers, Restaurants, Hospitality, eCommerce, and CPG Brands, Hypersonix helps clients to drive profitable revenue growth, saving money, and delivering exceptional customer experiences.
Harnessing the power of AI innovations like machine learning and natural language processing, Hypersonix offers its customers an exceptionally simple and fast “Google-like” text and voice search experience for their business augmented by “Jarvix” an embedded virtual decision assistant. It offers predictive and prescriptive analytics designed to help users measure results, understand root cause on why results are positive or negative, and what to do next.
Curious to learn how Hypersonix can help your business apply real AI to your business challenges?
✎ by Todd P. Michaud, President & Chief Customer Officer at Hypersonix, Inc.