What’s harder than training an algorithm to detect images or automate a process? Collecting and cleaning the data in the first place.
And what’s even harder than that? Integrating AI systems into old technology, old processes, and old skillsets that exist in most enterprises and midsize businesses today.
This sort of cultural shift is arguably vastly more challenging than the technical problems of AI. Computer vision and different kinds of NLP are rather proven use cases. They provide great value when you have the right data and the right setup, but baking that into an existing business is where the challenges arise.
This is what the vendor landscape of AI companies are trying to solve. They’re all trying to become more accessible.
Because of this accessibility issue, enterprises often jump the gun and make poor decisions, and there are factors that they simply don’t consider or factors that they often over consider. We talked to vendors, we talked to consultants and we talked to the buyers in big enterprises, and I’ve come up with three broad topics that really do seem like the top pitfalls of AI adoption.
These are things that the C-suite just misses or is overlooking:
- The Insidious Force of AI Novelty
- Underestimating Integration Needs
- Underestimating an AI Application’s Time to Value
This article series will address each of these three factors, the first of which is what I call the insidious force of AI novelty. In this article, I’ll talk about the worst case scenario of falling for AI novelty and other common ways that a company could do so, as well as, most importantly, how to overcome it.
Pitfall 1: The Insidious Force of AI Novelty
The worst manifestation of the insidious force of AI Novelty is in looking for AI for its own sake. We saw a lot more of this 18 months ago than we do today because there was a lot more ignorance around the topic of artificial intelligence in general. I think we’re still probably at the same level of hype around AI, but we’re not at the same level of ignorance, which is a good thing.
Many business leaders attend AI events and do their research on websites like ours that dive into AI enterprise adoption and all that comes along with it. Again, this is a very good thing.
Although not every executive needs to drag themselves and their team away from work to attend an AI event, as we outline in our executive guide on the topic, on the aggregate, continued interest in how AI is making a difference in one’s sector is going to allow one to make much better decisions when they’re ready to implement AI at their company.
The opposite of all of that is looking for a place in your business to use AI when AI might not be an answer to your business problem nor the key to driving value.
Looking for AI for its own sake involves starting with the question “Where can we use AI?” instead of asking about the best technology to solve the problem or increase revenue is. Luckily, this doesn’t happen as often as it used to because business leaders are becoming more informed.
More often than not, executives start to look for AI for its own sake after coming back from an event or a meeting where they found out their competitors are using AI. Sometimes it comes after reading a competitor’s press release. Maybe they just announced a new chatbot application, for example. The C-suite then feels they need to jump on so-called AI “toy applications” for their own sake.
We’ve written about AI toy applications extensively in the past. In a nutshell, we consider an AI application a toy application when the company that’s implementing it is doing so just so they can say they’re doing AI. An executive at the company will see their competitor’s press release about a new predictive analytics system for fraud detection, and they’ll divert resources to procuring one from a vendor, too.
When they do this, they often fail to account for the extensive integration process that comes along with adopting AI in the enterprise and the time to value, which we’ll discuss in a later article in this series.
They ask, “Where can we use AI?” instead of asking the better question, “What’s the right technology to adopt into our business that could drive the most value?”
AI vendors will often play up the fact that their software leverages AI (even when in many cases it doesn’t) because they know business leaders believe they should be using AI at their companies. Although some vendors might say that their clients don’t care whether their product leverages AI or not, they do know that saying that their product does in fact use AI will get them more sales calls because the word “AI” drums up interest.
The vendor companies that we speak with often will be quite frank about the fact that a lot of the competitors in their space are just faking the use of AI. They’re doing so because it’s going to get them a writeup in a Forbes article.
But most columnists writing about AI don’t necessarily know how to separate hype from reality at an AI company. As a result, these publications end up unintentionally steering some executives down the wrong path.
Business leaders might want to know who to work with for AI-based marketing software, but the current zeitgeist isn’t one in which executives and journalists and aptly equipped to vet AI companies on their legitimacy nor AI products on their integration requirements and time to value.
Generally speaking, that requires two different kinds of knowledge bases: that from data scientists and machine learning engineers (the people who know how AI software works and the data and time it requires) and that from businesses who have bought and/or used AI software to varying levels of success and having spent varying amounts of resources.
That said, there are ways for business leaders to level up their skills when it comes to vetting AI companies and avoiding the lure of Ai novelty whenever a competitor releases an AI product or whenever people on stage at AI events hype up how disruptive AI software is going to be in their sector. We discuss this in the next section.
How to Avoid Falling for AI Novelty
The ability to avoid the pitfall of AI novelty is best achieved when one can determine where in their business AI might bring the most value and if AI is worth implementing in that business area at all. Then, business leaders will need to meet with the appropriate people on their teams to figure out answers to hard questions about AI’s integration requirements. In other words, they’ll have to figure out if they have enough data for the project and where it is.
Determining Whether AI is the Right Tool for the Job
Businesses should probably not even look at vendor companies if they don’t have the capacity to leverage AI or a good enough understanding of where it could be used at their business. This requires a basic understanding of knowing what kinds of problems AI can solve.
For example, a hospital may want to find an AI software for diagnosing cancer in its patients. Machine vision software may be able to help, allowing radiologists to upload patient medical scans into the software and having it point out where in the image a tumor may exist.
The problem is that without the proper research, a hospital might believe that the machine vision software is flawlessly going to point out a tumor within a patient scan every time. That isn’t the case. The accuracy of machine vision software can vary drastically depending on how many patient scans it was trained on.
In addition, even if the hospital understands that the software might not accurately point out a tumor every time, hospital staff might think that a machine vision software that can diagnose cancer in the brain can necessarily diagnose cancer in the lungs. That also isn’t the case.
Right now, most machine vision software used for assisting oncologists in diagnosing cancer are trained to detect only one or two specific types of cancer. If the hospital wanted a software that could diagnose several different kinds of cancer, they’d likely need to build a software in-house, and they’d then have to ask themselves if they have enough of a history of patient medical scans to train the software on. They might even need to get patient consent to do so.
Not only that, but there are serious ethical concerns with adopting AI for medical diagnostics, so much so that vendors for this kind of software don’t even want us here at Emerj to say their software are used for medical diagnostics. Instead, they simply give information that physicians could use to diagnose a patient.
This is because if a doctor diagnoses a patient with cancer based on what a machine learning algorithm tells them exists in their medical scan, the doctor would be unable to explain to the patient that they have cancer because “the machine said so.”
In this example, the hospital needs to ask themselves both where AI can be used in their business (answer: medical diagnostics), what the scope of what AI can do to solve that business problem is (answer: detecting only tumors in specific parts of the body in specific kinds of medical scans), and if it’s even practical to implement AI to solve the business problem at all (answer: probably not at this point in time).
If you don’t have anyone at your company who understands where AI could be used in your industry and can speak what we call “the language of data,” speak to data scientists on their own terms, you’re going to make poor decisions when it comes to AI.
Determine Your Integration Needs
You also need to understand the kind of data you’ll need to implement AI. For example, in order to implement a fraud detection system, you might need certain kinds of payment and customer information. Also, you’ll need to ask if the data you do have is uniform and in a format that you could feed to a machine learning algorithm. In addition, you’ll need to know if it’s accessible. Is it available in real-time or does it exist somewhere you’re not even tracking or collecting from?
Now, of course, all vendors are going to try to say you don’t need to do any kind of upkeep with their solution, that they’ll figure out any issues with your data for you.
But the fact of the matter is these are often much more complicated questions than people presume.
Read our Executive Guides to Learn the Nuances of AI Adoption in the Enterprise
Our Executive Guide articles contain everything we’ve learned from speaking with vendors, buyers, AI PhDs, and executives in all different industries about how to successfully (and unsuccessfully) apply AI to business problems and adopt AI in the enterprise. In those guides, we talk about separating AI hype from reality, vetting AI companies, and about making the right executive decisions on AI.
One of these guides is called How to Apply Machine Learning to Business Problems, and I think it’s one of our best. We also did an interview with Madhu Sekhar at Amazon about the best places to apply AI first in business and getting started with AI at one’s company.
Our AI Strategy Work
We provide the kind of insights you’ll read in our Executive Guides to businesses at a high level. These companies bring us on for AI Strategy. It’s a two and a half day process that we normally do with enterprises or mid-sized companies to give them a framework to make vendor and adoption decisions themselves.
But the insights in this article and our Executive Guides alone hopefully should be actionable and useful, things to keep an eye on. In the next two installments of this series, we’re going to get into the meat of understanding integration needs and underestimating time to value. What are the harsh realities here that we need to understand if we’re headed for AI adoption? Those are the topics for the next two pieces in this series.
Header Image Credit: Irish Times