Artificial intelligence and machine learning are two of the most tantalizing ideas to business executives around the world as they consider AI projects for their company. What better way to address complex business challenges than to teach computers how to solve them, and then let the machines churn through problem after problem, as revenues surge, operating costs wither, and efficiencies reign supreme?
If only it were that simple.
As many have come to discover, AI/ML projects are hard. Engaging in a machine learning project for the purpose of generating sustainable positive business outcomes at scale presents quite a few risks. Around every corner, issues with budget, compliance, and resources – to name a few – loom large, waiting to derail even the most momentous of AI projects.
At Encora, our goal is to help business leaders who want to tackle big problems with AI to be aware of – and avoid – the risks that can undermine their efforts. We have worked tirelessly with businesses around the globe to do just that.
In this blog, we want to share a common, yet highly destructive, area of risk that has affected many of our clients’ projects: allocating personnel.
When many project leaders approach company stakeholders to get buy-in for their AI project, they often have a one-track mind: “Get my budget approved, and then get to work!”
For obvious reasons, having enough funding to advance your machine learning goals is important, but we’ll address that issue later. Too narrowly focused on securing financial resources, most AI project managers don’t consider the personnel resources required to get the job done. This issue, which often flies under the radar, can spell doom for an AI project looking to scale.
You Will Need Two Types of Personnel Resources For Every AI Project
Even when resource allocation for a machine learning project is not overlooked, oftentimes the of personnel needed to get the job done are. Put simply, there are two critical types of people you will need so that you can create, train, and scale your model: Implementation experts and subject matter experts (SMEs).
Implementation Experts
Most project managers understand they will need people to actually do the project, but perhaps they overlook all the various elements of the “doing”, so to speak.
To implement an AI/ML project, you need to think end-to-end. You’re going to need to create a new machine learning model, train it, integrate it with existing digital products, retrain it, and scale it. Each step requires unique implementation experts. To achieve your project goals, you’ll need data engineers, machine learning engineers, data scientists, business intelligence analysts, project managers, in addition to traditional software development roles like DevOps and QA. That’s a lot of implementers, and they probably don’t all report to the same person.
Subject Matter Experts
When you set out to build an AI model, the goal – in simple terms – is to teach a “computer” to do things that an experienced human implicitly knows how to do. Whether that be to detect fraudulent banking activity or optimize service distribution routes, AI is usually applied to complex tasks understood only by a handful of experts.
Listen up, because you need these experts to buy into your project upfront if you want to be successful.
Subject matter experts are just as critical as implementation experts, though they are often much harder to recruit. It’s the job of developers and engineers to build AI models, however subject matter experts might work in operations, sales, product management, or finance. They might not even work for your company. Regardless, they usually have jobs to do, and devoting time to your AI project might not be high on their list.
Most people think about the implementation experts at the beginning. Few think about the subject matter experts in the same way. However, in order to get your AI model right, you’ll need the SMEs on board. You’ll need to carefully explain to them the long-term business goals of the project, ideally in language that communicates a direct benefit to them. If they don’t agree that giving their time to you is a worthwhile endeavor, you’ll be waiting on them to approve your model and validate its results.
How To Eliminate Stakeholder Ambivalence
To get your data science project off the ground, you must have the right personnel in place. And that means making sure that your implementers and subject matter experts (and their bosses), fully appreciate the scope and goals of the project, and agree that these goals are worth pursuing.
We’ve worked on a number of machine learning projects with clients, both large and small, and we understand that assigning and aligning resources can feel like herding cats.
In our experience, we’ve had success implementing Machine Learning Canvas to improve collaboration amongst stakeholders to keep the project goal-oriented, on time, and on budget. If you are trying to get your project off the ground and need to get your teams in sync, we strongly recommend ML Canvas.
Louis Dorard is the author of the Machine Learning Canvas, founder and chairman of PAPIs.io, and holds a PhD in Machine Learning from University College London. www.louisdorard.com
