How Much Do AI Projects Really Cost?

iStock-1185373154-COST-Feb23-2021-825x340

It’s no surprise that enterprises have fully embraced the role of artificial intelligence in achieving better business outcomes, as AI projects continue to unearth new possibilities around productivity, risk avoidance, and sustainability.

According to a recent study by PwC, artificial intelligence has the potential to add over $15 trillion of value to the global GDP by 2030 – that’s more than a 10% increase directly related to advances in machine learning and its applications.

And when business leaders engage in AI projects, this is exactly what they are thinking about: value. How can we leverage growing computational power – and our ability to apply it to intelligent purposes – to gain a firmer hold on revenue growth and stability, or keep rising costs at bay?

At Encora, we help businesses find the hidden value in AI, and conversely, avoid making investments in machine learning products where positive value can’t be realized. After all, an AI project shouldn’t cost more than the business outcomes it enables. In today’s blog, we offer some advice to make sure that doesn’t happen to you.

Understand The True Cost Of An AI Project

Every AI project has a cost. To build and scale a data science product you’ll need a variety of personnel, plus the technological infrastructure to make it happen. The prudent AI project leader will estimate the resources and infrastructure needed to achieve the project goals, outlining a proposed budget to be submitted for approval.

But here’s the catch: AI projects don’t just have one budget. In fact, they have two. Let us explain.

The Allure Of Starting An AI Project

Most AI projects start with a question. The question often sets a lofty technical goal, enticing project sponsors to go all-in on development. For the creators of the first autonomous vehicles, it might have been posed like this: “What if we could teach a car to drive itself?”

To teach a car to drive itself, engineers might have begun by testing different types of geospatial sensors and building an algorithm around the data they create. With some effort, they might have taught the car to accelerate when the road was clear and brake when it encountered an obvious obstruction. As they further gathered data and tweaked the model, the car learned where to turn, when to use its blinker, and how to shift gears to maximize fuel efficiency. With considerable cost and effort, they did it. The car could drive itself…

… in an open parking lot. With constant supervision.

This is the phase of the project Gartner likes to call the “Peak of Inflated Expectation.” We see an initial technical goal, get excited about the possibilities it could create, secure the budget to accomplish that goal, and get to work.

Achieving the technical goal is exciting; it does indeed open up a world of possibilities, and visionary leaders get fired up over how this new intellectual property can differentiate their business.

But here’s the catch: patents aren’t the building blocks of business, products are. Most AI projects misfire because they fail to account for the risks of protracted experimentation especially with new technology applications such as AI, let alone the need for ongoing budget after achieving the initial technical goals. In other words, they fail to consider what it’s going to take to convert that differentiating AI algorithm(s) into a revenue-generating product.

Productization Is Much More Expensive Than Proving A Technology Goal

To understand the ongoing budget, let’s go back to our autonomous vehicle example. Our technological pioneers didn’t set out to develop a self-driving car just to say they could. They did it so they could sell hundreds of thousands of them.

And herein lies the problem. Letting one autonomous vehicle loose in an empty parking lot is trivial compared to merging millions of them onto the world’s highways. At that point, we don’t just need an algorithm that controls the gas pedal and the steering wheel, we need an entirely new transportation infrastructure!

This example illuminates what many AI project leaders fail to perceive: that the cost of transforming an AI project into digital products that result in tangible business value far outweighs the cost of developing the algorithms in the first place.

Developing an initial AI model can be as simple as putting a handful of data engineers and data scientists in a room with a CSV file full of training data, a Jupyter notebook, and two pizzas, and letting them pull an all-nighter.

But taking their model and integrating it with your products and revenue streams involves many more resources, both financial and personnel. What’s more, you’ll need quite a bit more technological infrastructure.

How To Ensure Your AI Project Has Enough Budget

When seeking stakeholder buy-in for your AI project, make sure to consider both AI project budgets: the one for starting up to prove a technological goal, and the one for productization. Both are required to achieve the ultimate business goal, and the latter is usually much more costly. You’ll need to set clear expectations for both, and those approvals will likely come from different stakeholders based on achieving milestones over time. The initial budget might come from Directors or VPs. Productization budgets almost always come from the C-Suite.

But remember, it’s okay to begin work with only your startup budget secured in order to achieve technical proof of concept. As you move ahead, keep expectations realistic and communicate frequently to all stakeholders in order to pave a smooth path for the business to invest in productization.

Share this post

Table of Contents