Most machine learning leaders focus more on the technology than its deployment, so most new machine learning initiatives fail. But if we instead focus on concrete operational improvements, we bypass the hype and pave a drivable route toward realized value.
Make no mistake, operational change is a tough sell, especially in comparison to hot tech, which sells so effortlessly that we actually call it sexy. It’s less glamorous to propose a process overhaul. Folks respond like you’re suggesting a root canal. But that’s life: Great gains come only by imposing great change.
Yet, by realistically and concretely communicating what ML offers — and, ideally, by calling it ML rather than AI — we can differentiate it from the often misinformed hype that defines the AI brand and save ML from being a victim of its own hotness.
3 Tips for Your Elevator Pitch
- Lead with the value proposition.
- Estimate the value with KPIs.
- Keep it short and listen to feedback.
How To Express the Value of Machine Learning
We need only reorient the focus. Don’t propose a technology project. Instead, pitch and, ultimately, lead a project that will improve operations, with no more than a side note mentioning ML as part of the solution. Usually, ML projects are framed in this way: “AI will improve operations.”
In addition to dispensing with the term AI, which usually compromises clarity, we must reframe the project in this way: “We will improve operations (using ML).”
By way of illustration, let’s apply this reframing to a couple specific examples put in the briefest of terms:
- “We will run a new marketing process to retain customers (using ML to target those most at risk of canceling).”
- “We will improve the efficiency of UPS’s package deliveries (using ML to predict delivery destinations).”
Reframing ML projects in this way puts the business objective first, rather than the technology, and, likewise, it shifts the agency from the technology to the business. The first word of the sentence is “we” not AI, humans not machines.
Moreover, this adjustment puts change management squarely on the agenda. Change management is a well-established discipline designed to facilitate operational shifts, but it can only do so if it’s employed. Many ML projects don’t recognize that the notion of change management applies, but model deployment means changing the very way the business operates and that change must be proactively managed like any other.
By viewing the endeavor as an enterprise project rather than a technology project, folks will recognize an often-overlooked truth: ML deployment presents a challenge that only a change management process can meet.
Preparing the Pitch
When you’re selling machine learning deployment, swimming upstream against resistance and inertia, it sometimes feels like you’re hustling. But in actuality, you’re recruiting. You’re enlisting collaborators and orchestrating a vision. Don’t get me wrong; as you advocate for the project, there are times you might need to aggressively cajole, enduring a battle of wills against a universe of naysayers.
Resist the temptation to ride the wave of AI hype. It oversells.
When you first pitch, bending over backward is part of the deal. You may feel like this stuff should basically just sell itself. After all, the value proposition can seem totally obvious when you’re already invested in it. The potential operational improvement is a no-brainer.
But, to get the green light, you must get the people in charge not only interested but enthusiastic. This means taking a step back from the excitement and telling a simple, non-technical story that is dispassionate rather than fervent, one that could just as well come from the lips of a truly impartial third party. In the art of sales, evenhandedness is more rousing.
To sell ML convincingly, sell it succinctly. Strengthen your pitch by distilling it down to the fundamentals: the precise operational change, the value of that change and how ML will achieve the change, in that order. It’s time to perfect your elevator pitch.
The Elevator Pitch
The premise to this practice is simple: Reframe “ML projects” as “operations-improvement projects that use ML.” Leading with the scientific virtues and quantitative capabilities of the technology — such as modeling algorithms, the idea of learning from data, or the notion of probabilities — is putting the cart before the horse. Instead, lead with the business value proposition, a simple story about how processes will improve.
Here’s an example elevator pitch: “Currently, 99.5 percent of our direct mail is ineffective. Only half a percent respond. If we could increase that to 1.5 percent, that would mean a projected $500,000 increase in revenue in return for our current marketing spend, tripling the return on investment of marketing campaigns. I can show you the arithmetic in detail. ML can hone down the population to whom we’re marketing by targeting the customers more likely to respond. This should deliver the gains and ROI I just mentioned. What do you think? Would you support this project or would you have objections? What questions do you have?”
When you pitch, get straight to the point, the business value and the bottom line, and then gauge the person to whom you’re speaking. They will be interested in the business value, but they’re not necessarily excited about ML. ML is only the technical solution, the means to the end, so, in this early stage, its details can easily distract, confuse or bore.
Your narrowly focused pitch must accomplish three things.
1. Lead with the value proposition
Express the value proposition in business terms, without details about ML, models or data. For now, share nothing about how ML works, only the actionable value it delivers, the operational improvement gained by model deployment. This usually means avoiding the words “model” and “deployment.”
2. Estimate the value
Think of the value of performance improvement in terms of one or two key performance indicators, such as response rate, profit, ROI, cost reduction or labor reduction. You must include a potential KPI win, even if only from scratch calculations. Convey this potential in simple terms, such as a bar graph that has only two bars to illustrate the potential improvement. It’s not yet time to mention predictive performance metrics such as lift. Make the case that the KPI win will more than justify the expense of the ML project.
3. Stop and listen
Keep the pitch short and then open the conversation. Realize your pitch isn’t the conclusion but rather a catalyst to begin a dialogue. By laying out the fundamental proposition and asking them to go next, you get to find out which aspects are of concern and which are of interest, and you get a read on their comfort level with ML or with analytics in general.
After the pitch, you’ve got to interactively gauge when to get into details about how ML will be applied, and at what depth and speed. It’s more common than you may realize for the business professional to whom you’re speaking to feel nervous about their own ability to understand analytical methods. People are skilled at covering this nervousness.
Keep it simple. As with many technologies, convolution and the appearance of arcane complexity threaten to extinguish a newcomer’s excitement about the potential value. This might leave them feeling compelled only by the pressure that comes from all the “Everyone’s doing it!” hype. Nip that in the bud with a straightforward, concrete explanation. Cover just enough of the inside mechanics to demystify ML.
Stick to the Facts
Resist the temptation to ride the wave of AI hype. It oversells. The propaganda’s sheer excitement does successfully broadcast that there’s value to be had, but it only distracts from the concrete value proposition by idealizing the core technology.
Don’t passively affirm starry-eyed decision makers who appear to be bowing at the altar of an all-capable AI. If you do, here’s the risk that you face: When the hype fades and the overselling is debunked, much of ML’s true value proposition will inevitably be disposed of along with the myths.
This article is adapted from the book, The AI Playbook: Mastering the Rare Art of Machine Learning Deployment, with permission from the publisher, MIT Press. It is a product of the author’s work while he held a one-year position as the Bodily Bicentennial Professor in Analytics at the UVA Darden School of Business.