Do Managers Matter When It Comes to AI Adoption?
Absolutely! According to Gartner an astounding rate of AI projects fail. This is not fate and there is much that you as a manager can do to beat these statistics. For the last 12 years I have had the honor to work in various leadership roles within the Artificial Intelligence Group of Intel IT. This team created and productized thousands of AI algorithms and implemented some solid methods to increase the likelihood of AI success that I would like to share with you.
One of the initiatives is the Intel “AI Everywhere” program, founded last year with the objective of increasing success rate and value when applying AI across the company to solve various business challenges. We do so by providing a rich portfolio of consultation and training sessions, self-service AI tools, creation of end-to-end AI capabilities on-demand as well as fostering a large internal community of AI practitioners and enthusiasts.
Throughout this year, I’ve had the privilege to talk with and consult with many individuals, teams, and leaders across the company, all on a journey to utilize AI. Some are just beginning, while others are well on their way. Even though some of the challenges and solutions are unique to specific domains, there are clear commonalities among the different teams. One clear point that I’ve seen is that the level and style by which managers and leaders of an organization are involved in their organization’s AI adoption process makes a real difference. It’s by no means the only reason for AI adoption pace, but as managers we can help to speed things up. Significantly. Or at least create a real and feasible plan to adopt AI “right.” In the rest of this piece, I will try to outline the role a manager or a leader needs to play in AI adoption and the main activities, I believe they should be involved in to facilitate a faster, more successful AI adoption.
AI Sponsor, Driver, and Enabler – the Roles Managers and Leaders Need to Take
At Intel, I have seen two ways by which an organization has started its AI journey ─ “bottom-up”: an employee or a small group of employees takes an AI initiative and then goes to prove its value to get management buy-in; or a manager identifies the potential and takes decisions (and hopefully actions) to foster AI adoption, a.k.a. “top-down.” From my experience, it’s not how the AI initiative was started that is the best predictor of the pace or likelihood for its success. What is highly corelated is whether the managers involved (regardless of if they were there from the get-go or joined the party later) played the right roles: that is, the roles of AI sponsor, driver, and enabler. As such they need to make sure the right resources, goals, actions, and behaviors are devoted to the process. Below I will call out the 7 main activities I believe are most critical for any manager who wishes to help speed up the journey of their organization in applying AI techniques:
1. Set Clear Strategy and Goals for the Scope and Method of AI Usage in the Organization
When a team is just getting started with AI, and especially if the first initiatives were born bottom-up, there is still no immediate need to define an “AI strategy.” It is more about getting to tangible business results. However, as soon as managers in an organization would like to increase investment and speed AI adoption pace, they should start defining an AI strategy and goals.
And yes, things will most likely change along the way, and strategy adjustment will most likely be required, but it is still better than not having one to begin with.
Some questions that should be answered when defining the strategy and goals:
- Will AI be done by individuals from the team, or will it be outsourced to other teams or vendors?
- Is there an aim to create an AI center of excellence or distribute AI work across the organization? (Even if outsourcing some or all the work)
- Planned initial scope: Start small and simple (e.g., use out of the box tools and go for low hanging fruits) or will you go after the big stuff and create tailored capabilities?
- How will success in AI adoption be measured? ROI? Number of productized capabilities? Something else?
- Adoption pace – are you after taking a leap of faith and going big or gradually growing investment?
One direct implication of defining an AI strategy is that you will need to have a clear plan around growing AI knowledge in your organization:
2. Define the “Pyramid” of AI Skills in Your Organization and an AI Knowledge Growth Plan
AI is well on its way to being a core technology in all aspects of our home lives, and eventually it will be so also in our work’s. With that assertion in mind, I strongly believe all employees need to grow their AI knowledge and skills. I’m not aiming to make any employee an AI practitioner. I view it as a pyramid – where at its base is the AI knowledge all employees should have, and going up to the top, each organization can and should define the different “AI personas,” according to their AI strategy and goals. An organization’s “AI knowledge pyramid” should reflect how aggressive is its plan for AI adoption and how much of the AI competency it would like to grow internally vs. working with external partners.
In the example below (figure 1) – 3 main AI personas exist: “everyone,” “AI champions,” and “AI experts.”
The assumption here (which applies to many organizations at Intel) is that each employee needs to understand basic AI concepts. By that we can eradicate fear and objections and help identify opportunities. Even more so, when it gets to it, and it will, they will be more open to making changes to how they do their work to accommodate AI. To some extent at least.
The specific definitions of the next 2 layers in the pyramid below vary between teams, some break it further down, while others break it differently.
Bottom line here: make sure you have clarity about who should learn which skills and how they will get there. Some organizations will form ambitious “big-bang” AI upskilling plans while others opt for a slower, more organic pace of upskilling. All AI upskill methods are valid, as long as they are aligned with the organization’s AI strategy and goals.
3. Capitalize on Your Existing Data and Invest Smartly in Improving It
Data is arguably the most important enabler of AI. That doesn’t mean that until you have perfected all the data in your organization, you can’t apply AI. In fact, from my experience when organizations went for long and expensive “data foundation projects” without working simultaneously on deriving clear business value through their data, they typically discontinued the data foundation work, leaving all involved frustrated. I suggest starting with the data you have, or that you can get relatively easily, creating the highest value AI capabilities you can with it, and then, building on your success you can gradually grow the investment in mining more data. My strong recommendation, at all stages of your “data evolution” is not to be a “data hoarder.” Instead, collect data you have line of sight of how to collect and what can be done with it, and ensure constant business value growth enabled by your growing data sets. The more proof points and confidence you will have in the value of your data, the bolder goals you can go for, including collecting data currently not available in any system, making profound changes to tools, and working methods to collect data and even creating an altogether new data platform.
4. Manage the Change: Top Down and Bottom Up
Whether it’s to create better data or encourage AI adoption, often you will face objections. Some objections can be attributed to fear or conservativeness, but many objections can be removed if the process of AI adoption is managed properly. From my experience it is not enough to just manage the change “top-down” – i.e., defining clear strategy and goals, downloading them to the employees and assuming all will go smoothly with the implementation of each new AI capability. The most valuable AI capabilities I’ve seen require some level of collaboration between the human expert and the AI. It can be that the expert is expected to utilize the AI recommendations to improve their output, or they just need to provide feedback or improved data for the AI to succeed, or it’s simply required of them not to turn the AI capability off the first chance they have. Either way, if the end-user is not involved in the process and educated/empowered/assured to utilize AI for their best interest – implementation will likely fail. Therefore, end-users and experts need to be looped in early, at the specific capability level as well as the overall direction the organization is headed. Not only will the objections be significantly reduced, and thereby the likelihood of success will grow, they would most likely bring amazing ideas and speed the process significantly.
5. Select Appropriate Use Cases with Clear ROI and Business Goals
In my experience, many if not most of the AI failures could have been prevented if enough energy and scrutiny would have been applied at the use case selection phase. What I expect of managers, specifically, is to make sure the team as well as any decision makers fully understand the feasibility (see figure 2), risk, and potential value an AI idea holds. I see managers as the gatekeepers, of avoiding bias towards with a specific AI idea or the overall technology, to the point of missing potential problems. It’s not that by being diligent about which AI ideas you move forward with, you can prevent every obstacle and difficulty your future AI project will encounter, but at least you will be aware of the risks. Even more importantly you will ask yourself and the team difficult questions to ensure there is enough business value to be achieved to endure and overcome the setbacks that might come.
6. Assign the Right People Based on the Task’s Criticality and Complexity
By now, you’ve probably realized I strongly believe not all AI ideas were born equal. Thereby there isn’t a one-size-fits-all in terms of the skillset required. However, a typical scenario I see is that managers assign only data scientists or data scientists in the making to work on an idea. And while data scientists are often the right people to have when you want the algorithm side of things to be handled, usually they are not enough. Especially if going after more complex, integrative, and transformational AI projects. To increase the likelihood of a more complex AI idea getting all the way to production and yielding high business impact, these are the main personas recommended to be involved:
- Data scientists: accountable primarily for creating the best algorithm for solving the problem considering defined scope and goals.
- Subject matter experts: hold deep business acumen of the problem being solved and can stir decisions to maximize business outcome (e.g., how and what data to handle, how to properly define problem to be solved, how to go about integrating and implementing the AI solution as part of the business processes, etc.).
- ML engineers/AI Platforms engineers: accountable for the architecture and execution of an end-to-end AI solution that is optimized for the business problem and algorithms the data scientists created. They need to be proficient in modern AI SW and MLOPs practices.
- AI Product/Project Managers: people with deep proficiency in Product/Project management as well as in AI technologies. They need to be experienced in leading a multi-disciplinary team through the definition and execution of an AI idea all the way to productization and maximizing a sustainable business impact over time.
- Integrators: when applicable, it is highly beneficial to work closely with individuals and teams that can integrate the AI capability back to the existing processes and are able to make needed changes to make AI integration as smooth as possible.
- AI sponsor: a person in a leadership position that can help influence and remove roadblocks in the process of working on an AI idea. Including and not limited to allocating resources, making POR decisions, and shaping the right levels of decisions to maximize success. Often this is the most critical role you as a manager can play to speed things up in your overall AI adoption journey.
7. Set Expectations and Patience: Transformation Takes Time and Investment
This last one is probably the most important piece of advice I have to offer specifically to managers and leaders. If you’ve gone this far reading it means you have already made or are seriously contemplating making a substantial investment in making AI work for your organization. With that, you would like to ensure your ROI is large enough. And quick enough. However, from my experience, it is always a longer and more complex process to get to the point of “enough ROI” than most people anticipate. There are multiple reasons: AI is still a new and often intrusive technology; it holds perceived and real risks to your business if not done properly. In addition, the path from a successful POC to productization is always longer than anticipated. Finally, if taken to its full potential, AI is geared towards truly transforming how you run your business. And transformation takes time, investment, and patience. Figure 3 below shows a typical AI maturity graph with the different phases of AI adoption, their characteristics and typical duration. It is not aimed to discourage you but rather to set more realistic expectations about results and how soon to expect them.
Where Do You Go from Here?
A question that might linger in your head right now: “Having read and followed all 7 steps – can I guarantee to speed things up?” Well, the answer is: “maybe.” As I mentioned, even if done well, adopting AI is often a lengthier and more resource intensive process than initially anticipated. However, if done right, and regardless of if you opted for being your org’s AI sponsor, driver, enabler, or all, you will significantly increase your likelihood of success as well as the expected business impact to be derived. And these are some great motivators.