- ABOUT
- VIDEOS
- TOPICS
- TESTIMONIALS
- COMMUNITY
- EVENTS
- NOMINATIONS & LISTS
- SPONSORSHIP
- ABOUT
- VIDEOS
- TOPICS
- TESTIMONIALS
- COMMUNITY
- EVENTS
- NOMINATIONS & LISTS
- SPONSORSHIP
One thing contributing to the short, average tenure of chief data officers or CDOs (2.4 years1) is the challenge of clearly demonstrating the value they deliver. AI can help immensely. The hype cycle has set a strong expectation that high-impact AI is a natural outcome of improving data practices and that incremental, targeted investments in AI can provide reasonably quick, quantifiable business results.
Yet, a 2022 survey by Accenture2 reports that only 24% of companies feel they can fully realize the value of AI in operations. As a CDO, what should you expect as you work toward using AI to help show the value of your data efforts?
Be Ready to Influence Everyone – Especially Operations
Like many other C-level roles, chief data officers cannot own every part of the company necessary to fulfill their missions — CDOs must achieve results by influencing others. A 2022 NewVantage Partners survey3 highlights the fact that a majority of CDOs are now concentrating their efforts on growth and innovation initiatives and are expected to serve as change agents.
This is especially critical for your success in delivering value through AI. People from across the company — including those in operations, IT, data science and analytics, and governance — must play their roles for AI to deliver value and most will be outside the CDO’s span of control. Some may feel skeptical or even threatened by the work.
Why “Especially Operations?”
Operations is critical because AI solutions can only generate value when actively, consistently and appropriately used by their intended end-users. There is another consideration, though, that warrants an additional focus on operations.
The same people will consistently handle many aspects of the AI lifecycle. They have the chance to practice, learn and mature over time because they go through their tasks many times as new AI solutions are developed. Operational adoption is usually a different story. Each new AI solution is likely to focus on a different team or process, which means that the hands-on operations folks may never go through the adoption of a new AI solution twice. Essentially, you almost always have a group that is critical to success but also inexperienced in their tasks.
To make this even more complicated, companies often target their AI solutions at simplifying burdensome, labor-intensive, expert tasks. This often raises fears in the minds of the experts putting in the labor for those tasks. They may be concerned about losing their jobs or their competitive edge — or wonder who will take the blame if they embrace the change and it doesn’t work. No one wants to make themselves a target.
Actively investing the time, effort and expert resources to help end-users and their leaders manage fears and uncertainties while supporting and ensuring change is critical to the success of AI solutions. As a CDO supporting AI solutions, it would not be surprising if, in the long run, you focus most of your effort around AI here.
Bring a Consultative Skill Set
We’ve said it before, and we’ll say it again — we realize value from AI through its active use in operations. Having a mismatch with operational needs is one of the easiest ways to fail with AI. This can take several forms:
The most critical way to avoid this kind of mismatch is to bring a consultative skill set to the party. It is imperative to have someone who can serve as a bridge between the uber-technical folks who are building the AI model and the operational users of the model. This should be someone who can understand the needs of operations, understand the model development process, and help negotiate changes and requirements on both sides to achieve the most useful business outcomes.
Treat AI Projects Like an R&D Investment Portfolio
AI projects are more like pharmaceutical research and development than construction projects. If you commission the construction of a house, you know you will get a house. With pharma, you have a list of chemicals that have some potential, but you know some of them will not pan out in the end.
AI projects depend on consistent patterns in data that are strong enough to advise business decisions. It is entirely possible to have an AI modeling project where the available data has no consistent, strong patterns. In those cases, you want to “fail fast” and learn what you can. This allows you to get more or better data and try again, revising the goals based on what you’ve learned so far or abandoning the effort.
For this reason, it is critical to actively manage expectations with requesters so they are less likely to be surprised if an AI model does not pan out. You should also have several incremental screening or go/no-go gates in the intake process for AI model requests to help foster the fail-fast approach.
Be Intentional, Incremental and Opportunistic about AI Lifecycle Maturity
As you can see, there is a lot of ground to cover regarding the maturity of the AI lifecycle. You have a lot of overlap with software development, data development and process change management. Beyond that, you also have the additional skills, tools and infrastructure for AI model development and monitoring. As a CDO, this is likely more than you should be managing day-to-day, so it may be more practical to designate a champion or trusted partner to help plan and manage the priorities in detail.
In addition to having a champion, it is also incredibly useful to have a framework or checklist of capabilities to use as a touchstone when navigating through the complexities of how it all needs to tie together from end to end. This can help you piece together data points about your organization and properly discern where the gaps are between where you are now and where you want to be. Once you’ve identified the gaps, you can develop a plan to close them.
A Caution About AI Lifecycle Maturity Planning
It is extremely easy to drift into big-bang or waterfall mode when trying to build a roadmap to maturity. But neither of those approaches is likely to be practical for several reasons.
First, it is typically infeasible to introduce and adopt changes impacting many groups outside your scope of control. A good example is achieving AI lifecycle literacy among all potential AI consumers across the enterprise, which requires coaching and experience participating in the AI lifecycle. Even if you could provide enough coaches to cover every organization, it is extremely unlikely you would be able to engage actively in AI development projects with every organization at once.
Second, different partners in the AI lifecycle will each have their own constraints, including changes scheduled that do not apply to AI. It is unlikely they will be able to follow a single integrated AI lifecycle change schedule.
Many areas, especially data development, can respond opportunistically to change. For example, if triage of a high-profile AI project in the intake process indicates you need more data to make modeling feasible, you can change data development plans to prioritize the missing data.
What Does Incremental Look Like?
AI requires end-to-end thinking. You need ideas, enthusiasm for said ideas, people and infrastructure for getting and serving the data, people who can develop models, people who design how others will use those models in practice, and people to put them into practice. However, like baklava, this can be done one thin layer at a time. How it scales can look different for each component.
For example, you don’t need to have that end-to-end process in place all at once for every business in your enterprise. You can start by going one process or proposed change at a time and keeping it confined to the groups or departments involved before expanding your footprint in other areas.
You can work on every component incrementally. Take, for example, a common scenario for AI projects: We don’t have the data. If this concern surfaces often enough, you can still make incremental progress on this front by making it a priority and fitting it into your data acquisition plan. Similarly, you can scale staffing for the modeling itself based on demand, i.e., the number of businesses or processes they support, independent of data availability.
Moving incrementally also means bringing in business processes and focus areas individually. You may start with the low-hanging fruit or high-value areas to highlight early successes. You might also opt for a lower ROI area that is ready to start sooner rather than later. This can help you showcase your capabilities to help build buy-in for solutions with higher ROI. In any case, you can use this intake process to help inform and advise your data acquisition strategy and plans. This allows you to get into the details about how you would incrementally move the needle for each element of the AI lifecycle.
Conclusion
By following these steps — especially consulting closely with operations — you can fully realize the value of AI in your business and begin to deliver quantifiable results. And as a successful change agent, you may even lengthen that average CDO tenure.
About the Author
Tony Lung leads the Center of Excellence for Machine Learning and Data Science at Centric Consulting. His passion is using his 30 years of experience in data science, information technology, process improvement and organizational change to help people discover sustainable value and capture it. Lung earned bachelor’s degrees in math and computer science from Northern Kentucky University and an MBA from The Ohio State University. He is a proud United States Marine Corps veteran and has served on two school boards in Central Ohio for more than 10 years.