Punctuation matters when working with data: BI is periods. AI is question marks.
The biggest mistake I see companies make in ML/AI, is to not have a plan. Diving in headfirst, without first developing a data strategy, means relying on luck. That really decreases your chances of success.
The next-biggest mistake, even in companies that develop a plan, is to assume certainty in their ML/AI outcomes. This is hardly the place to make bold declarations.
How many times have you heard the following?
We will use AI to predict [whatever].
This sounds very positive. Very definitive. Very much in-tune with Western business culture’s obsession with staying upbeat.
It’s also a great way to get in trouble. As I’ve shown in the posts on Total Cost of Model (TCM) and Setting Expectations, the ML/AI world has its share of uncertainty, wrong turns, and bad luck. All of these conspire to derail your project.
That’s is why I tell people: when planning an ML/AI effort, use question marks, not periods.
That means building your project around a question. We can rephrase that earlier statement as:
We will use AI in an attempt to determine: can we predict [whatever]?
This wording opens the door to accepting the inherent uncertainty of ML/AI work. You can now explore “what are some ways we can try to get there?” followed by ranking those approaches in terms of costs, potential payoffs, and risks.
Some readers may say that I’m language-lawyering here. I would agree with them. When company leaders speak, their words set expectations. Those words must therefore be clear and align with the reality that every ML/AI exercise is an experiment.
Once you accept that it’s an experiment, you’ll plan accordingly: you’ll be far more mindful of how you invest time, effort, and money, because you’ll understand that you may not achieve your preferred outcome. You’ll prevent the buildup of unrealistic expectations. Most of all, you’ll spare your reputation and your team’s morale by not having to walk back a bold declaration.
There’s still some certainty in the world of data, if you look to business intelligence (BI).
BI is all about looking at your data as it already is. Those roll-ups, sums, and cubes lack the probabilistic nature of ML/AI models, which is precisely why they’re so useful: as long as the data is good, anything BI tells you is the truth. “How many widgets have we sold? Broken down by year? And then by store location?”
In the hands of someone who knows the business well, insights drawn from BI analyses tend to be very reliable. Compare this to ML/AI, where you won’t know it’s working (or not working) till after the fact.
For a concrete example of BI’s power, consider Cameo. This company makes money by connecting celebrities with fans. In an interview with Kara Swisher, Cameo CEO Steven Galanis explained that they use the site’s search data to determine which celebrities to invite to the platform.
This isn’t very fancy data – it certainly doesn’t involve neural networks – yet it is extremely valuable to the sales team. It narrows the scope of whom they contact, which saves them time and reduces guesswork. It also gives them hard numbers to show that celebrity how much demand already exists for them in the platform based on how many times people have searched for them.
Imagine a Cameo salesperson explaining to you: “Searches for your name have trended dramatically upward over the past few days. Roughly fifteen thousand searches for you, each day. Fans clearly want to hear from you on Cameo. Don’t you want to capitalize on this momentum?”
All this, from simple counting.
Now you see why I say: “BI is periods. AI is question marks.”
Knowing when to rely on BI as facts, and how to treat AI as a maybe, will help you make the most of your company’s data efforts.
(Update, 2021/10/04: I’ve added ellipses to the list of punctuation. )
This post is based on my workshop for executives, Unlocking Your Data Potential. Contact me to run this workshop in your company.