Asif Mujahid:
I got a call one day from our chief privacy officer. At that point, I was thinking to myself, do I need to update my resume and start looking for jobs? Now it’s not a technology question anymore. The technology has exponentially surpassed our expectations, in my opinion. Now it’s a risk, governance, and adoption problem. These AI models are as good as a junior doctor in terms of their knowledge level. What about judgment? Are we going to introduce bias into the system that negatively impacts disadvantaged members of our population?
Saket Saurabh:
Hello everyone, and thanks for listening to another episode of Data Innovators and Builders. Today I’m speaking with Asif Mujahid, Chief Data Officer in the healthcare and health systems industry. Asif, thanks for chatting with me today.
Asif Mujahid:
Yeah, it’s my pleasure to be here, Saket. Thanks for having me on.
Saket Saurabh:
You’ve had a very wide set of experiences. Tell us a little bit about your background and how you got into data.
Asif Mujahid:
Definitely. You know, it’s one of those things where, when you’re young and growing up and trying to figure out what to do with your life, you figure out what your interests are and what your strengths are. It’s a great thing if those two align. And if they don’t, you figure out the Venn diagram of where they overlap and what you want to do with your life from there.
I was fortunate enough to figure out that my strengths lay in complicated subjects like mathematics and data, and I had a strong interest in those too. So I said to myself, why make life complex, let’s just build a career in this domain. I went through my education with a background in economics and econometric modeling, which led to further interest in developing a career in the quantitative sciences.
Once I graduated from college, I went through different industries: aviation, finance, and then the healthcare industry, here in the US, over the past few years. It’s been a fascinating journey for me because, at the end of the day, I think I can speak for everybody, we all like to be data-driven, we like facts to guide our decisions. Folks like you and me, analytics practitioners, have a service to meet that expectation: to make things clear for our stakeholders. So that’s been my main reason for the career I have right now, offering clarity, going through the data, offering clarity to folks so they can make more informed decisions.
Family-wise, we have a couple of young kids, one is about to go to college, so that’s a whole drama by itself, and then an eleven-year-old who keeps us on our toes all the time. My wife’s a physician, and as I always joke with folks who know me, I’m the one doing all the simple stuff, building algorithms, while she’s the one actually dealing with lives. She’s doing the more important work in the household. But again, it’s my pleasure to be here, Saket, and I’m looking forward to a good conversation.
Saket Saurabh:
That’s an incredible background, Asif. Right before we started, we were talking about AI a little bit, especially how the impact will be, and what impact we might assess when we look back on it a few years from now. So maybe let’s kick off with that AI question. Let’s talk about actually making a real-world impact with AI, maybe not as much as a physician, but what sort of impact can it create? Let’s chat a bit about building these AI programs. Can you double click a bit into that from your experience? Where are you seeing the challenges in building real AI use cases?
Asif Mujahid:
Certainly. I’ll start with a statement that I think encapsulates a lot of what I’m seeing happening in the industry right now. Over the course of the past two to three years, with OpenAI and so on, there’s no magic there, we’ve progressed a long way when it comes to AI, and I think everybody aligns on the notion that there’s a big value opportunity with AI. I think we’re now at a crossroads: how do we get from that opportunity to value realization? Essentially, if I were a CFO, how do I see it in my books? This value that we’ve all been thinking about, articulating, expecting over the past few years, can I see it in the financials within the next one to three years?
We’re at that inflection point right now, in terms of how you go from A to B. That’s one of the biggest things people are trying to figure out, because now it’s not a technology question anymore. The technology has exponentially surpassed our expectations, in my opinion. Now it’s a risk, governance, and adoption problem. Here’s what I mean by that: now you’re moving from the world of technology into the world of people and human behavior. How do I, as a non-AI individual, use this black box of AI model information that’s coming my way? We’ve got to tackle that issue.
Some of the tactical things a lot of us have been thinking about, myself included, is that you address the trust question. Keeping everything else aside, governance mechanisms and frameworks are extremely important, but it comes back to trust. At the end of the day, will I, as a non-AI professional, have enough trust in this model, or whatever it is, to help me make more informed decisions?
In order to build that trust with your non-AI stakeholders, there are some fundamentals. There’s an element around model transparency: if that’s part of the program, how can we offer transparency in very simple terms of what this model does, how it does it, to make life simpler for the people who are going to use it. The second is data quality. The principle I’d use when addressing a data quality issue behind a successful AI program is, can I defend this in front of a regulator? Does it meet the standard of a regulator? Can I put myself in the regulator’s shoes and convince them that the data quality is good enough for us to move forward?
The third piece, probably the most important, is accountability. That’s a big gray area right now, in the sense that you have a model, an AI program, who is it going to impact, what are they going to do differently, who owns that decision to do things differently? Equally important, if things go wrong, what’s the communication mechanism, what’s the rectification mechanism, so we can learn from it? There’s a whole body of work around accountability and responsibility that needs to be defined as part of an AI program.
So, to cut a long story short, to answer your really good question: there’s a technology component to an AI program, and I think that’s mostly been resolved. But then there’s a whole wrapper around trust, transparency, data quality, and accountability that needs to be built around it. It’s not anything new, people are thinking about it, but that’s what takes us from value identification to value realization, and that’s one of the challenges I see for all of us to overcome, to build that wrapper around the technology piece of a successful AI program.
Saket Saurabh:
I think this is absolutely right. You’re in an industry that’s regulated, an industry that has a real impact on people’s lives. So all the things we did, even pre-AI, in terms of what we’re building and what the impact and risk are, you have to defend in front of regulators as well. When you look at AI approaches, generative AI specifically, do you think it creates a unique challenge on that front, in terms of predictability not being the same?
Asif Mujahid:
Yes, it does. That’s also a really good question. There’s a predictability piece, absolutely, which gets into how we can eliminate the black-box piece of it for our stakeholders. But there’s also a big governance piece to this, because at the end of the day, there’s a saying that just because something can predict something doesn’t necessarily mean we need to act on it right away. So how do you get from prediction to action, and what do we need to do to get from point A to point B?
Let me share a real-life story. I won’t mention names, but this goes back ten or twelve years. My team and I were building predictive models, and I got a call one day from a chief privacy officer. At that point, I was thinking to myself, do I need to update my resume and start looking for jobs? So I went over to the chief privacy officer’s office, and their question, essentially, was: with all this predictive modeling, this was pre-AI, are we going to introduce bias into the system that negatively impacts disadvantaged members of our population? It’s a great question, the right question to ask, and it’s one of the pitfalls we need to consider.
At times, not always, but at times, we, myself included, get so enamored with the model, with the technology, that we ignore that aspect of it, which, in this case, the chief privacy officer was raising: the customer experience, the patient experience. I know we weren’t going to do this, but say there’s a scenario where you build a model, everybody’s excited, you throw it into production, and now it’s negatively impacting a segment of the population. How do you account for that moving forward?
That’s an example of a pitfall. It’s less about governance and more about the efficacy of the model, but it’s one of those things where, if I’m a non-AI professional thinking, “these folks are sitting here building models while I’m dealing with the patient,” will it eventually lead to a better patient experience? If I’m a physician, I’ll channel my wife, I don’t know the answer to that right now. Someone needs to convince me of that. That’s an example of the kind of thing we need to figure out, in my opinion.
Saket Saurabh:
That’s a great point, and it actually brings in a very important element that I think many AI leaders are looking at: the human in the loop in the whole process, and the diversity of opinion and perspective that matters. That’s why, in your story, the chief privacy officer brought a diversity of perspective, because there’s a mathematical optimization or solution, but that’s not necessarily a socially optimized or business-optimized outcome. This is an absolutely critical point. In fact, it reminds me of stories I’ve seen about people using AI as a therapist, and realizing in some cases it’s actually giving the wrong advice. It might look good on the surface but push people toward outcomes or actions that are actually negative. There are examples where people have left AI systems entirely on their own to observe them and have seen behaviors that are self-preserving, but not in the overall interest. It’s a great point you bring up.
So, let me switch to a bit of the work you’ve done within the healthcare industry. You’ve been at companies like UnitedHealth Group and regional players like Quads. First, let me ask, continuing on that AI question, we’ve seen coding become one of the success stories of enterprise use of AI. I’d say until coding came along, the big question was, “AI is cool, but what does an enterprise actually do with it?” and software engineering became the first answer. When you look at your experience in the insurance and healthcare sectors, what use cases do you see as having been successful, or worth looking at?
Asif Mujahid:
Good question. I’ll make this simple. I’ll focus on healthcare, since that’s where I’ve been for the past ten to twelve years, and I’ll put the whole healthcare model into three buckets, three places where value can be created through AI. There’s an administrative bucket, a supply chain bucket, and a care delivery, care optimization, care management bucket. Those are the three places where, theoretically, AI can create value.
The administrative one is, as the name suggests, a lot of administrative cost, because there’s a lot of redundancy from a process standpoint, very complex documentation in healthcare. I’m not saying that’s good or bad, there’s regulation, so you need policies and practices in place, but there’s a fair amount of inefficiency in the system as a result. That’s a good use case for AI, because you’re not really dealing with the patient experience, so that risk and liability goes away, you’re dealing with process-related things. That’s where I’ve seen a lot of effort concentrated, and there’s value there, because it relates to margin leakage, or reducing the cost to maintain a program. Even a small reduction in cost is a good thing, because then you can scale it.
The second is supply chain, the end-to-end model of healthcare, which we’d probably need a few hours to fully discuss, but to simplify it: you’ve got drug manufacturers, the dispensary, the health system, the health plan, and all the middlemen in between. From a supply chain perspective, can AI come in and create value here? There’s an inefficiency component in all of this. I called out supply chain specifically because it’s not purely process, there are entities involved with their own specific interests, which makes it not the easiest bucket. The easiest is the administrative one; supply chain is second, but there’s still an opportunity there.
The third piece, where everybody’s excited, is the care management bucket. This is where your earlier point comes in: when you think about care management, it’s about the patient. We all want to do the right thing, even though people have different incentives, but assuming everybody wants to do the right thing for the patient, if you leave an AI model ungoverned and unfettered, you’d have no idea what that patient experience would look like. In my conversations with folks across the spectrum, physicians, health system executives, people in private equity, there’s always a feeling that the biggest value is in this care management space. If we can figure out how to get to a better patient experience and a better physician experience through applying AI in its different forms, that’ll lead to something better.
I’ll give a quick, tactical example. The other day I went in for my annual checkup with my primary care physician, and he mentioned he wanted to get an opt-in from me for an AI-based note-taking system they’d just started using. So, obviously, I did a bit of market research and asked him what he thought about it. He actually thought it was really helpful, this AI scribe. It’s a very early-stage use case, but in his opinion, it helped him because, in the old world, he was talking to me and taking notes, talking and taking notes, whereas now his entire attention could be devoted to me. He felt it led to a better experience because he could listen to me more attentively without worrying about the note-taking, which the AI scribe was handling.
That’s an example where, on the face of it, in the care management space, you’ve created some value, a better patient experience, because I was talking to my doctor without him being distracted. But to come back to your question, the bulk of the use cases I see right now are in the administrative space, some in supply chain, and just a few in care management. But at least from what I understand, the market thinking is that care management is probably where we can create the greatest value with AI. That gets me excited, because we all have to contend with the fact that, despite our best efforts over the past ten to fifteen years, healthcare costs are still going up. Can AI come in and offer a better care and patient experience while also bringing down unnecessary costs in the health system? That’s the end of the rainbow we’re looking at, and hopefully we can get there as we learn and mature in this space.
Saket Saurabh:
I love the framework you’ve laid out here, the administrative side, the supply chain, and the care side. It’s a great way to think about the risk, the impact, and the value all in a clean way. And it’s great that you bring up the cost of healthcare, because that’s a national challenge for us in this country. From talking to physicians and healthcare professionals in other countries, I hear they have different systems, which have their own challenges too, but in many ways, when it comes to the quality of care, especially for complex health situations, the United States is still considered to have some of the best care you can get. The challenge is the cost of that care. I agree with you entirely that the administrative side is essentially overhead, it’s not leading to better outcomes, but there’s a lot of it, whether regulatory or structural, given how differently healthcare is managed and administered in this country. So it’s a great opportunity to be more efficient, and hopefully the ROI a patient gets, the care they receive relative to the cost they pay, improves significantly.
Good healthcare is great, but if it’s not affordable, or people can’t get it at all, then it’s pointless, at least for that individual. So that’s definitely a worthy cause to focus on. Now, one thing you mentioned in your example was your own experience with a healthcare provider, and how gathering data was in the way of that experience, and now, with technology and AI, gathering data is out of the way, which improves the care experience. It’s probably more time-efficient for the doctor too, since I hear from doctors that the amount of time they spend entering information is overhead for them, not where they get job satisfaction, but a necessary part of it. So, since we’ve touched on data, can you share your thoughts on data management in this space and how it’s evolving? You’ve seen it at different scales, and one important element I’m seeing is the ability, with generative AI, to take advantage of unstructured data, not just call recordings, but images or free-form notes, alongside structured data, to quickly assess what’s going on. Give us a picture of how data management, in your opinion, is evolving here.
Asif Mujahid:
Definitely. There are two things I’ll mention. When we think about data management, we use this term, “data is messy.” That’s always been true, we’ll never get to perfectly clean data, since data is always changing and evolving. That core truth, that data is messy, was always there, regardless of whether AI was in the mix. There are data challenges, data issues. What AI has done, directly or indirectly, good or bad, is thrown a spotlight on data. AI is coming back to us and saying, “I’m the shiny new thing, but you’ve really got to fix your data if you want me to do a good job.”
So I see a lot of organizations across healthcare investing more in data governance and modernizing their data platforms. In the old world, this was looked at as technical debt, and companies were unsure about investing in it because the ROI was always challenging to demonstrate for modernizing a data platform. But now, because of the opportunity for value with AI, there’s a fair amount of investment moving into the data middleware, which is a good thing, because we’ve been saying for years that predictive modeling and data science are great, but there’s a whole data layer that needs fixing too. So that’s the first thing, there’s a fair amount of investment happening, and companies are thinking deeply about modernizing data platforms and, especially, data governance, to support their AI programs.
The second thing is the example you brought up, which is actually one of the best use cases I see: at the point of conversation with the patient, how can you glean information that isn’t in your healthcare claims? There’s always a lag with claims data; if somebody’s looking at my claim information, it’s from an event that happened thirty, sixty, ninety days ago, which doesn’t speak to what I’m going through right now. So if, at the point of conversation, a physician or care provider can glean new information about my health, that can lead to a better patient experience. The question becomes, first, how do you tap into that new information I’m providing; second, how do you make sense of whether it’s meaningful; and third, if it is meaningful, how do you act on it, in real time?
I’ll go back to a quick example, this is dated, from a few years ago, pre-AI. At one of the organizations I worked with, we were trying to do something similar. We didn’t have the AI tools we have today, so the use case was: a patient calls into a call center and, while talking with a customer service agent or clinician, offers new information we didn’t already have. Could we use an API-based system to tap into that information, run it through a decisioning engine on the back end, and then surface updated information on the clinician’s screen while they’re talking to that patient, in a way that made the conversation more valuable? We were doing that a few years ago.
I think AI has a lot to offer in enhancing and evolving that use case further, because, going back to one of the biggest challenges in healthcare, the patient experience, in a lot of cases we’re talking to a patient about information that’s thirty or sixty days old, and we leave it up to the patient to volunteer new information. Without an AI solution, that information just gets lost, it’s not captured or saved anywhere. So there’s a big use case in building a model around that and improving the patient experience. It goes back to what we talked about a few minutes ago, the care management space, how you create value there, and that’s actually a very good example of creating value for both the patient and the physician delivering their care.
Saket Saurabh:
One thing I wonder about, especially as we hear that these AI models are as good as a junior doctor in terms of knowledge level, is what to think in terms of judgment. Many of the decisions doctors make are judgment calls from years of experience, not simply looking up symptoms or information. I was curious, now that you have data in all these different forms, notes, recordings, reports, images, radiology, and you bring that together, because AI can make this kind of data queryable in ways it wasn’t before, and part of the data challenge has always been that it was hard to tap into, most systems were more structured, in a database, cleanly organized, and if the data quality is high, great, you can do a lot with it. But there’s a new set of capabilities enabled by being able to tap into that unstructured information. I’m curious, when it comes to the role of the data team and the AI team in supporting the physician, you mentioned making an API call and surfacing information, what’s the vision for how the role of data will evolve on the caregiving side?
Asif Mujahid:
Definitely. I’ll go back to this old framework we all know, the V’s of data: volume, velocity, and variety. With the technology we had five or ten years ago, we were at a certain state of volume, velocity, and variety of data you could attach to a patient. To your point, AI has come and blown the top off that, especially generative AI, where now the sources of unstructured data you can tap into are much greater, which impacts all three V’s, the volume, variety, and velocity of data you can attach to a single patient.
If you think about it simply, say you’ve got a patient record with, I’m making this up, ten fields of data, coming in weekly. Now you introduce gen AI into the mix, and you’ve got twenty-five fields of data coming in daily for that patient. That’s the data piece of it, a much richer, deeper set of data sources you can attach to the patient, and that’s one of the big contributions of gen AI in this field.
The piece that’s still missing, and I think companies are working on this, some more advanced than others, is the decisioning piece, the judgment piece you touched on. Now that we have this data for me as a patient, what do we do with it? What algorithms do we need to run it through to come back with a different recommendation for how my treatment or care plan should look? A very simple example: based on the original ten pieces of information, there was a certain care plan for me. Now we have twenty-five pieces of information coming in more frequently, and we pass that through some sort of decisioning engine, does that change my care plan? And here’s the key statement: is it a different decision the physician makes about the care applied to me? That decisioning shouldn’t be based only on the collection of data, or only on the decisioning engine. It should be the combination of the new data, the decisioning engine, and the physician’s judgment. At the end of the day, the physician should be the judge, because they’re the one interacting with me as a patient, and you want their judgment, intuition, and experience complementing all these data and decisioning pieces.
That, to me, is the framework through which we can create some value, going back to the earlier discussion about moving from value identification to realization. If we can make that happen, and I know people are working on it, I haven’t really seen any standout solutions yet, but it’s good that people are thinking about it. That would be one pathway to substantial value through AI.
Saket Saurabh:
I feel like one of the key capabilities of people who are doing really well is being entrepreneurial in their thought process and execution, because you have the tools, the data, and the technology, and it’s about execution, and the pace at which new capabilities and products are coming out is very fast. So I want to double click on your own entrepreneurial experience. Tell us a bit about that, and how it’s shaped some of the work or thought process you’re bringing in.
Asif Mujahid:
Definitely. We all have an entrepreneurial bug in us, and I had that too, at one point, and it’s still ongoing in some ways. I’ve been fortunate enough to work for large organizations, smaller organizations, and to have had a startup of my own. The two things that have been really different across those experiences are speed and consequence. In a startup environment, you don’t have strong consequences, you’re focused on speed, on failure and learning from failure, so you can get a product into the marketplace that the market adopts. You’re not as concerned about consequence.
At the other end of the spectrum, if you’re a large organization servicing millions of customers, there are many more consequences to failure. Even though we’ll all say, honestly and transparently, that we love to test and learn, love to be a learning organization, love to fail fast, there are varying degrees of how much of that you can actually tolerate, depending on whether you’re in a startup or a larger organization. And there’s nothing wrong with that, because in a large organization there’s much more to lose from unfettered failure all the time. So, basically, the difference is the pace at which you learn, the speed, and the consequence. For me, the learning was: when you think about learning, failure, and being a learning organization, don’t ignore the context of the organization you’re part of, because there are different degrees and levels of failure an organization can tolerate, depending on whether you’re at a startup or a large organization.
Saket Saurabh:
I think this is very well put. Everybody in their work wants to create impact. In a startup, there’s a bit more freedom, because you haven’t yet found out what will create the impact, and your main job is to experiment and find that. In fact, not doing anything is the biggest failure, so it’s better to do something, even if you don’t know whether it’ll work, and iterate until you find something that does. There’s a reward to failing fast. But in an enterprise environment, with real consequences, failure can have significantly negative costs, healthcare being a good example. The mission still exists, people want to create impact, but you have to work within those boundary lines. Figuring out how to still experiment and find what has impact, but with more rigor before you can put it out there and deploy it and start impacting the lives of millions of people.
Asif Mujahid:
Yeah, that’s absolutely right, Saket, and one thing I’ll also say here is that one of the interesting models I’ve seen some large organizations use, for people listening to this podcast, is creating a company within a company. So, while the overall organization sustains its model and maintains its current processes, you have an incubation arm within that organization doing this, and hopefully you achieve the best of both worlds. In engineering, we call this a sandbox, you’ve created a shell where you can do things without the risk, you’re in a safe environment.
Saket Saurabh:
That’s a great segue into the next point I want to talk to you about: building and structuring teams in today’s organizations, as a CDO, across multiple functions, data engineering, analytics, AI, data science. What would you say is the approach you’d recommend there?
Asif Mujahid:
Definitely. The first statement I’ll make is that, when we think about these components, data engineering, analytics, AI, in my opinion, for an AI program to succeed, ownership needs to be centralized and consolidated in one part of the organization. It could be a chief data officer, a chief technology officer, a chief financial officer, I don’t care which, but it has to be centralized under one individual or team that owns it end to end.
Now, the reality is, and this is advice for new CDOs in the marketplace, when you’re centralizing, make sure there’s organizational momentum behind it and that people are aligned with that structure. Because the second piece, and if you don’t have it, it won’t work, is alignment and collaboration with your partners. It’s not only about centralizing, because the last thing you want to do is centralize, go off, shut everything down, build an AI model, and then, six months later, surprise everyone with “we have an AI program now.” You’ve got to centralize, and then have a collaboration model with your stakeholders to keep them informed of what’s going to happen, what’s going to change, and what partnership you need from them.
There are two other things I’d mention for a team to be successful. One is that, when you’re in the analytics and data world, you have a lot of information, the good, the bad, and the ugly of the organization, because you have all their data. What that means is there’s a service you owe your stakeholders, because they want to be data-driven, they want facts to inform their decisions, and I see a lot of frustration in organizations because they’re not getting data and information in time to make decisions. So you need that service and delivery model as part of your organizational structure. Not only centralize, not only have a collaboration model, but have a co-developed, trusted service and delivery model that lays out expectations with your stakeholders: what’s going to be delivered by when, what should they expect, what do they need to change in their workflow or decisioning as a result.
The key statement I’m getting to is that we need to move from a world of delivering reports to a world of assisting or supporting a stakeholder in an outcome. It’s not a new or novel statement, it’s just that we haven’t gotten to that second piece, and AI is now compelling all of us in the data and AI world to think differently, that it’s less about building and delivering a report on time, which is still a good thing to do, and more about being jointly responsible for the outcome that you and your non-analytics stakeholder are trying to achieve for the end customer.
Saket Saurabh:
I completely agree, and I think the chief data officer role has always been about business impact, because data is just the means to an end. How are decisions being made? The evolution has been that decisions used to be made by gut feel, and over the past fifteen to twenty years, there’s been enough evolution that so much data can be captured that you can make highly informed decisions. I want to bring that to the question of how the CDO role itself has evolved. I feel like data engineering was thought of as more of a technical function, writing code, so it went toward the engineering, CTO side of the org, but data science is rooted in statistical expertise, which is very different from writing code, and analytics is very close to business understanding, what’s happening in the business, what matters, because looking at thousands of charts won’t help you if you don’t have a good feel for the business, the seasonality, the customer behavior, the structure of the business. It brings together a lot of different know-how, and then you add AI into the mix. So, are CDOs supposed to be superheroes who know everything from engineering to AI to statistics to analytics, or how do you see that evolving?
Asif Mujahid:
The answer is yes. That’s a really good question. Here’s my thought on this. The first thing I’ll say is that the expectations of the CDO role have changed a lot compared to ten years ago, and it’s a bit of a pet peeve of mine that job descriptions aren’t changing to reflect that. There’s some confusion in the marketplace, part of which you touched on, about what exactly the chief data officer should be, the title itself boxes you into a certain space, so there’s something around semantics we need to figure out.
But to look at it more deeply, the chief data officer plays certain archetypes in an organization. There’s the technical archetype: technical capability, building a data platform. A second archetype is people leader: building a strong, business-focused organization. The third is stakeholder manager, because, at the end of the day, as you pointed out, data alone doesn’t change business decisions, decisions change the business. It’s not that the CDO plays only one archetype, they’re playing all of them at the same time. What’s changed is the relative importance of these three over time, and what I’m seeing is that the importance of owning joint outcomes with your business stakeholders has risen to the top of the mix.
Here’s what I mean. From my own experience, a lot of our stakeholders always anchor on the phrase “actionable insight,” not just data, which is important, but an insight that’s actionable. My pushback on that statement is: is that really the right statement, or is a better one, “I need something that tells me whether I need to make a different decision about the problem I’m facing”? That could be an actionable insight, or it could be something else. The point is, are we asking the right question? My advice to CDOs would be, with your stakeholders, one scenario is you’re sitting there, well-intentioned, trying to understand their problem, they say “actionable insight,” you nod your head and try to figure out what that actually means. Another way to interact is to not start with the phrase “actionable insight” at all. Start with the problem they’re trying to solve, what decision they feel they need to make differently, and work backwards from there to decide what you actually need to build.
That, to me, is the biggest change CDOs need to think about: sit down with your stakeholders, figure out their business problem, tactically narrow it from multiple use cases down to one important use case you can deliver in the next ninety days. Have that discussion, come back to your team, and then figure out what you need to build, a dashboard, a report, a new model. Otherwise you get caught in a loop of trying to build technical capability, which you obviously need to do, while also trying to decipher what your stakeholder wants, since they’re not speaking the same language you are on the technical side. So, to cut through that confusion, take some time, sit down with your stakeholders, get to that one use case you can deliver on in the next ninety days, deliver on it, and move forward from there. It’s easier said than done in organizations, but that’s one thing CDOs really need to focus on going forward.
Saket Saurabh:
I think it’s very well said. I feel like the concept of “actionable insight” is a great one, but it sounds like a fairly narrow definition, almost an antidote to the analytics manager who’s just throwing dashboards at people, where the question becomes, “okay, there are a lot of dashboards, but what’s the actionable insight?” But the role is much broader than that, in terms of impact on the business, from a strategy and impact perspective, whether it’s how you run finance, marketing, how you understand your customers, your own product, and when you bring that together with the delivery of AI capabilities, it’s much broader. But I can understand why that term became important at some point.
Asif Mujahid:
Definitely, and I’ve got nothing against the term itself, it’s just that, as a non-analytics stakeholder, you might be missing something, and that’s the term that’s supposed to help people understand what you mean. That’s where I’d advise newer CDOs to spend some time parsing down “actionable insight”: what do we actually mean by this action, can we tie it to a use case, what do we actually need to address that use case? Call it actionable insight or whatever, but that’s the level of detail you want to get to, to solve the problem your stakeholder is bringing to you.
Saket Saurabh:
Technically, all work is about actions, we get paid for what we do, not just what we discuss or think over, but that’s part of the process of getting to action. I think it’s very clear that fragmenting ownership of data engineering, analytics, and AI is not the solution, when different people own it, it’s chaos, you don’t get the deliverable or the business value. It definitely comes together and raises the bar on the role, but, as you said, align to the business and sit down on what outcomes they’re looking for. I want to ask one last thing about some of the research work you’ve done. Can you share a bit about the academic research side of things, and how that relates to the real-world work and impact you’ve been applying?
Asif Mujahid:
Definitely. I’m a big proponent of academia, because a lot of the ideas and solutions we’ve implemented in the real world, fast forward to today, have come from academia. Now, the challenge is that when you bring something from academia into the real world, there’s a strategic tension, because in academia you’re dealing with clean data, variables you can control, sample populations that don’t change over time. All of that is very different in the real world, where your populations are shifting, you have operational constraints, and so on.
The way we approached our research at one of the prior organizations I worked with was: look at academia, look at models, figure out what problem they address, is that a problem we’re facing as a business, and if so, how do we move from point A to point B, bringing that academic model into the real world? We used some academic models, brought them into the real world, operationalized them, and then shared that research with the world. At the end of the day, the phrase I’d use is that academic models offer us the right solution to a problem, and when you bring it to the real world, you’re figuring out the managerial solution, what you need to actually make it work in the real world.
The point I’m making is, I wouldn’t start from the real-world solution, because then you’ll be very constrained. You want to start with the academic solution, the right, broad solution, and figure out what you need to do to address real-life constraints and shifting populations to make it work in the real world. That’s some of the work we did in the past, and we published our research. It was a very successful program, and it gave me an even greater appreciation for academia, because there’s still a lot untapped in the academic world that we haven’t brought to life in the real world. There’s an opportunity for us to do more there.
Saket Saurabh:
I think it’s very powerful, and great work, being able to take the time and effort to work on the academic research side of things as well. There’s a lot of evolution in the thought process that has to come, and much of it can’t be applied directly in industry right away, but thinking about those frameworks and approaches is extremely important as we evolve and get there. Thank you so much, this has been an extremely engaging conversation, I’ve really enjoyed learning from you, and I think our audience will too. Before we wrap up, if people want to follow your work, what’s the best place for them to do that?
Asif Mujahid:
Just connect with me on LinkedIn, happy to have conversations. I’m always looking to talk to folks, I always joke that my work in the analytics world is boring, so I always look forward to conversations. I want to thank you, Saket, for the opportunity, I really enjoyed our discussion, some great questions. Hopefully this is helpful for folks listening in.
Saket Saurabh:
I’m sure it will be. I think one of the things people will want to follow is what you’re working on, on LinkedIn, but is there any other resource you’d point people to, for learning the way you’ve been learning and exploring, any books, podcasts, publications, or tools you’d recommend?
Asif Mujahid:
One that comes to mind immediately is Dan Ariely, behavioral economist at Duke University. I won’t say much, you just have to go and listen to him. It’ll change your life.
Saket Saurabh:
That’s great advice. Thank you so much, Asif. Great chatting with you.
