'Just Do Something With AI': Bridging the Business Communication Gap for ML Practitioners

August 23, 2024

It’s an increasingly common experience among data scientists - you’re minding your own business, getting your job done, and your boss’s boss walks in to your office or drops in to your Slack DMs. “Do we have AI in the product? What’s your plan for making our product AI powered?” (That’s if you don’t just get ambushed with this question in a big meeting in front of a bunch of executives.)

You have choices. You can see where this could lead, and there are hazards on all sides.

Do you:

accept this at face value, pivot away from your carefully curated roadmap, and throw all your resources at finding some way to make your product “AI enabled”, whatever that means? I know your team is going to be thrilled to have all their hard work to this point in the quarter delayed yet again for executive priority changes.
internally roll your eyes and say something to immediately shut down this nonsense, giving ammo to your rapidly developing reputation as being “not a team player” among executive leadership?
nod and agree, mentally making a note to try and slow-walk this whole thing until the asker gets discouraged and lets it go, hoping they’ll forget that you said anything?

Of course, none of these is the best approach if your goal is to benefit your own career and also to effectively support the business. (If your goal is something else, then this article is not for you.)

Instead, here’s the smart guide to handling this scenario.

1. Ask lots of questions.

Who is asking you for this, and what is their motivation? What are their priorities for the business, and how does AI fit in? As a data scientist, I’ve spent a career thinking about why people are asking for a specific machine learning solution and determining whether that will actually serve their goals or not. It’s not reasonable for those of us in the field to assume that everyone else in the business understands the deep inner workings of machine learning. It’s normal that they may have some misconceptions about what the right ML tool is for their problem, through no fault of their own.

I’m a big believer in educating your organization about the essentials of machine learning and data science, even people who don’t work on tech teams or rarely collaborate with you. This is because, first, I think everyone should have a foundational level of science literacy in order to better understand our society and economy, and I have the ability to help with that. But second, selfishly, having colleagues who understand ML and AI better will make my job easier in just such situations as this.

2. Decide if there is a real problem to be solved

I’d actually recommend your first question be “what results are we hoping to achieve with AI?” Asking this may lead the requester to reconsider the wisdom of the idea, if it isn’t well thought out. If they do have a business reason for this, however, you’ll start to get some insight into that. There may very well be a business problem occurring that really needs to be solved. However, if they just say “everyone has AI now” and look at you blankly, then you have my permission to just say “No” and go back to work.

3. Decide if AI is the right tool

Now that you know what the business goal is, that should help you a lot with deciding whether AI should be involved. However, one of your next questions ought to be “What does AI mean to you?” because you can get wildly differing answers to that one from different people. The term “AI” has a lot of baggage, and some people will think it means ChatGPT, and others will think it means basic automations, and there are lots of others in between. Once you know what you both mean when you say AI, you’re able to start a real conversation about what they’re really proposing, and whether it has the potential to help with your business task. Does it? Is there some other way you could solve this problem that is easier and less laborious than building something with AI? Or, is there an AI/ML approach that your asker doesn’t know about that might actually really help? Give this some thought, and if you find an AI-adjacent solution that seems like it might potentially be worthwhile, proceed.

3b. Decide if the risk is worth the reward

This is part of deciding if AI is the right tool, but it’s so important that I think it deserves its own section. AI is not a free lunch - you already know that adding AI to your product or business processes takes work and may be challenging, but also consider that there are data privacy risks, security challenges, and potentially risks to your business in other forms that can come with some types of AI. Will this require large volumes of customer data, to train or fine tune a model? Is that data sensitive or potentially dangerous if it gets into the wrong hands? What data privacy regulations is your business subject to, and what do those laws say about use of AI or machine learning? Ask your questioner whether they have consulted legal or your security department about this, and if they haven’t, you should. You’ll need to determine if the risks are outweighed by the benefit to the business that would come from solving this problem, and also whether you’re legally empowered to do the project at all.

We’ve reached a point where you know what the business goal is, and you know what the questioner means when they say AI, so we’re equipped with an understanding of more or less what the question was to start with. In addition, we’ve passed by some trip wires- we’re not going to get halfway into planning something only to discover when the executive said “AI” they meant an “if-then” statement, and we’re not going to spend weeks on this before realizing there is no market for AI features in our product at all. We know what risks we might encounter and what is safe to try. However, we still have more work to do.

4. Scope out what the work would really entail

There’s a good chance the person who asked you for this project doesn’t realize how much hard work goes into building anything with machine learning. As the practitioner, you should have much better insight into this, and taking some time to really think about how much work this project would entail is important for deciding if it’s worth doing. You should already have a good sense of whether AI could do anything to help with the business problem at hand, theoretically. Now you need to determine if it’s practical, given the circumstances and resources at your disposal. Consider what areas of expertise will need to be involved, and which of your team will need to assist, or if you are a team of one, which other teams’ support you might need to call upon. Are those other teams willing and able to support you on this? This may call for discussions with the leadership of those teams, because you don’t want to spend your valuable time on part of the project, only to discover you can’t get it across the finish line because support isn’t available.

5. Know your own priorities

Even if you’ve reached a point where you believe that developing something using AI might work, you are capable of doing it, and it would be valuable for the business, that still doesn’t mean you should agree to do it, at least not yet.

Your department has plans for where you are prioritizing your time and resources (or if not, you should), and these plans are aligned with larger business priorities. There are reasons you have chosen where you want to spend your time, and that means you should have at least some of the necessary information to weigh the value of changing track to some kind of AI related solution versus other things you planned to do. Maybe this AI opportunity is worth it! Maybe, however, it is not. Maybe it’s worth doing eventually, but not now, and it needs to go to the back of the queue and be prioritized for a future sprint. Take a clear eyed look and decide which is the case, and prepare to articulate the argument if asked. That argument should include noting things you can’t do now if you decide to take on the AI project. Getting good at explaining what has to give if you switch directions is a valuable skill for business success.

Once you’ve determined what the ask is, what the business goal is, whether AI is going to be any good for this, and whether it seems to you like it’s worth the tradeoffs, you should have developed an opinion about the whole thing. Now you need to go out and communicate that decision. It may or may not be entirely up to you whether you take on this AI project or not, but you can greatly improve the chances that your preference is the direction things go by having a thoughtful argument for your point of view.

6. How to say yes or no

Odds are, no matter whether you agree or disagree with the idea of adding AI to your product, someone’s going to be at least a little bit irked. Either it’s the exec who asked in the first place or the sales team who really wants to be able to tell prospects about our AI pizzazz, if you say no, or it’s your team who’s feeling some whiplash or the other department who’s waiting for you to complete something for them that was already on the roadmap and got pushed, if you say yes.

It also doesn’t matter whether you are in the decision-making seat or not, you need to be able to effectively argue your case.

If you are in a position to make the decision yourself, then you might not think it’s important to convince others of your choice. However, let me argue that your life will be tremendously easier and work will go much more smoothly if you are able to be at least something of a consensus builder instead of an autocrat. People will be more likely to help you and support you if they understand why you are doing what you do, and are at least aware of the value proposition.
If you are not in the decision-making seat, you still have an opportunity to push your organization’s decisionmaking in the most productive direction, if you make this case well. Ideally, leadership in any organization will be open to hearing thoughtful arguments that prioritize what’s best for the business, and using that to drive decisionmaking. Furthermore, you can enhance your reputation among those leaders by making a wise argument and showing your competence. (If you don’t work in such an organization, I’m very sorry and I hope you find a better workplace soon.)

Regardless, it’s best if you can present a solid argument and empathize with whoever is less than thrilled about the situation. Recap the components that you developed during this process that got you convinced to proceed.

What’s the business problem? If it’s an important problem, explain why, or if the problem is not really worth addressing, point that out too.
Does AI help, or not? If it does, how? Connect back to the business problem and be clear about what this contributes to a solution.
What are the risks? Don’t say “none” because that’s never true. There’s always some risk, some opportunity cost, something that you should acknowledge honestly and forthrightly. Even if you want to say yes to the project, show that you’ve thought about it with clear eyes first. If you don’t want to do it, then clearly identifying the risks and explaining how they influence your opinion is very important too.
If you want to do it, when? Why is that the right time to do it? If you don’t want to do it, explain all the other things of higher value you’re going to be doing instead - things you’ll have to skip or delay if you decide to do something with AI. Don’t just tear down the project idea itself, but talk about the value that will occur instead.

I hope this is helpful in giving you a road map for handling the pleas for AI that are a growing part of our work lives. Approaching things reasonably, without being either a doormat or a brick wall, will lead to better results at work and will have positive impacts on your quality of life. Good luck!

Photo by Unseen Studio on Unsplash

This article also appears at the ODSC blog, and I will be speaking at ODSC West 2024 in October.