Leading a Workforce Empowered by New AI Tools

ALISON BEARD: Welcome to the HBR IdeaCast from Harvard Business Review. I’m Alison Beard.

When any new technology comes into a workplace, you’ll usually see the IT department and a few other early adopters experimenting with it first. Eventually though, as the tech becomes more user-friendly, there’s a tipping point where almost everyone could find a way to use it in their jobs, from the C-suite to the back office. Right now, artificial intelligence and specifically generative AI is having that kind of moment. Once the purview of mathematicians, engineers, and coders, AI tools can now be effectively designed and deployed by anyone who has some basic knowledge and training.

But how do we as individuals figure out how to do it most effectively? How do managers both encourage and corral these efforts? And what strategies should organizations put in place to harness the power of what our guest today calls citizen development?

Tom Davenport is a professor of Information Technology and Management at Babson College, a visiting scholar at the MIT initiative on the Digital Economy and co-author of the HBR article, We’re All Programmers Now, as well as the book All-in On AI: How Smart Companies Win Big With Artificial Intelligence. Tom, welcome.

TOM DAVENPORT: Hi, Alison. Happy to be here with you.

ALISON BEARD: Could you start just by illustrating really clearly this idea that everyone, me included, is or should be a programmer, a citizen developer now? How do you see that happening on the ground at organizations with people in non-tech roles?

TOM DAVENPORT: Well, this is a trend that has been developing for a while. I mean, I think even my traditional area of analytics and big data and AI, we’ve seen these tools getting easier and easier to use, and it’s just point and click now for many interfaces and so on. So you didn’t really need to have a lot of technical background.

And then a couple, maybe five years ago, you started to see these, quote, “Low-code, no-code” tools emerging that would let you build small systems without really having a lot of technical background or needing to produce programming code. And then there were automation versions of that. The robotic process automation vendors introduced tools that were very easy to use and people could create their own kind of workflow automations and citizen data science.

We’ve seen great advances with automated machine learning tools that are quite accessible. And now, generative AI really takes it to the ultimate level where if you can write an English sentence or whatever language about what you want, it can produce code or do a data analysis, or do almost anything technical that you would like. And so I think the last barriers have been erased to non-technical individuals being able to do almost anything that they would like to do with information technology.

ALISON BEARD: So give me some specific examples of people who I wouldn’t think of as being technologically-focused in their roles, putting these tools to work.

TOM DAVENPORT: Well, one of my co-authors on the HBR article and his name is Ian Barkin and I are working on a book on this topic. There are a number of people we found in large corporations who are starting to do this sort of work and having a big impact.

So I haven’t gotten permission to use his name, but there was a gentleman who was employed at Home Depot who had been doing a lot of spreadsheet work. He was in the area of forecasting and assessing consumer demand and looking at what that means for what kind of inventory they should have around the different Home Depot stores. And really, a very time-consuming job, basically lived on spreadsheets.

And he discovered a tool, in this case, it’s from a vendor called Alteryx, that lets you automate a lot of the data inputs and outputs and some of the analysis as well. And so he was able to produce a sales forecast in a few hours at most rather than the weeks that it took him previously. And Home Depot made a huge amount of money. They cited well into the nine figures of extra income as a result, and he got employee of the year.

Now as often happens, he’s no longer doing that. He’s now a consultant to help other people use these tools. So I suppose he’s moved from being an amateur to a professional, but that’s quite common. We’ve seen it at a number of other companies. I’m talking to somebody in Lego recently who’s done a similar kind of thing. I was talking recently to a fellow at BMW who, I wasn’t talking to him about citizen development, but it turns out he heads that area as well. And he said, “Well, we’re training 80,000 people in citizen development approaches.” So it’s really going viral.

ALISON BEARD: Yeah. My worry when you were telling that story about the fellow at Home Depot is that he was putting himself out of a job. Now, obviously he moved on to a much better one, but is that a danger associated with citizen development?

TOM DAVENPORT: Well, there seems to be enough development around that anybody can do it, and there’s still plenty more to do. Almost every company these days wants to digitize. And we have historically relied on IT departments to do that kind of work. But I think almost everybody has a story of asking for an application from IT and being told, “I’m sorry, we don’t have time for that.” Or, “Sure, we’ll get to it within a year or two,” and nobody wants to wait that long.

So I think this just radically accelerates the timeframe in which we can get the technology capabilities that we need. And there seems to be no limit to them.

ALISON BEARD: So is it your sense that sort of the average non-tech employee is eager to jump in and learn how to do this? Or do you think that most people need some incentivizing?

TOM DAVENPORT: I think it depends. I mean, if they are sort of technical heat seekers, then they’re probably anxious to do it. They also, I think, typically aren’t quite sure what this will mean for their career path in many cases. And you’ve typically either been a professional IT person or you’ve been a non-IT business person, and never the twain shell meet. But now I think there are more and more hybrid kinds of roles possible.

In many cases, I think you do need a little bit of instruction in how to do this well. But I think there’s a large population that is at least open to it with some instruction.

ALISON BEARD: On the organizational level, do you see interest in an applicability of citizen AI development differ across companies, industries, geography? Does it have to be a company that’s already focused on using AI or not?

TOM DAVENPORT: I think it’s bubbling up pretty much all over the place. As I mentioned, consumer products, companies like Lego, PepsiCo is another one that I’ve talked to, not historically known for their AI activity.

So I think it is bubbling up in places that have not historically been heavily AI-oriented. And maybe one could argue that the places where it’s least common, the industries where it’s least common are those that are very sort of transactional in nature with their IT systems and where there’s a fair amount of regulation.

So you don’t see as much of it in banking, for example. One of my co-authors on the HBR article was head of data science at a big European bank, and he’s since become a consultant himself, but he said there was some opposition, and there probably should be. You don’t want your basic demand deposit accounting system that’s keep track of how much money you have in the bank to be generated all over the company in different ways. But I think there are lots of possibilities in almost every industry.

Well, I did recently talk to a very prominent technology company, and they’ve recently introduced a number of tools for doing this kind of work, both on the data analysis side and on the automation side, and also quite big into generative AI. And I think they’re right on the edge of chaos. They’re not over the edge yet, but thousands and thousands of different dashboards being created of small automations, small departmental level programs, and they’re doing their best to try to maintain some control over all of this, but it is certainly challenging.

ALISON BEARD: Yeah. So let’s get into some of those challenges or pain points. What are the big ones that arise when people start either doing this on their own or with encouragement from their managers?

TOM DAVENPORT: Well, I think most companies are not that far along yet, but certainly you worry about someone developing an important application on which the organization comes to rely, and maybe it’s not well-built. Maybe some of its calculations are not accurate. Maybe that person leaves and hasn’t documented the system very well.

I mean, one of the good things about generative AI is it’s also quite good at documenting systems, not just at creating code. So maybe we’ll see some help there. But if we become highly reliant on systems that are really quite critical to the success of the business, that I think requires some controls, it requires sort of registration by some central group, maybe IT, maybe a review of the functionality of the program to make sure it’s up to standard and so on.

And some companies have done that. We talk in the article about PWC’s efforts at citizen development, and they have a group of people who were amateurs, for the most part technically, who have been charged with taking them into the digital age and digital accelerators, they call them. And they do, they’re encouraged to submit all of the applications they develop to some central hub. And in that situation, they’re evaluated. And if they work well, and they appear to be a value to people around PWC beyond their own little groups or selves, they will make them broadly available. And they even pay a little bit to the person who contributes it, depending on how much usage it gets.

ALISON BEARD: So that’s the role that the IT department plays in this new world, oversight, quality control, that type of thing?

TOM DAVENPORT: That is certainly one important role. And I think broad-minded IT departments can play some other roles as well. They can provide a lot of the necessary training, if that is deemed appropriate. And really, there’s a kind of community development aspect of this where you not only encourage people to use these kinds of tools, but also to share what they’ve learned and to get together in regular meetings and hear updates about the technology or hear what great things other members of the community have done. So I don’t know who else is going to do that in general, if it’s not IT, or if we’re talking an automation or an analytics-oriented group, they may not be officially in IT, but they can still sponsor some of those community development kinds of activities.

ALISON BEARD: Yeah. And I imagine for those groups who have over the past decade been flooded with demand from colleagues to help them do all of this work, this will unburden them in some way if they can figure out how to do it right.

TOM DAVENPORT: There is that appeal. On the other hand, I think many IT organizations have been resistant to this for quite a while. They don’t think amateurs will be able to create high quality code. They are worried about a citizen-developed application being dumped on them and say, “Hey, fix this up. It has a problem or update it,” or something like that. They’d rather develop their own code. So I think only slowly, in some cases with reluctance, is the IT organization moving toward this view of the world. In some cases, we’ve talked to a couple of companies where they got a new chief information officer and he or she realized this is the wave of the future, the way things are going. And so they started to open the floodgates a little bit, but after many years of resistance.

ALISON BEARD: What role should team leaders be playing to ensure that all of these efforts are happening effectively?

TOM DAVENPORT: Well, I think that’s probably the area where you’re most likely to see the motivation coming from. If that Home Depot forecaster hadn’t had a boss who was tolerant of that sort of technical innovation, then it probably wouldn’t have been successful. At Lego, we talked to someone who’s a demand forecaster, and I hadn’t really thought about this previously, but it appears… Demand forecasting typically involves a lot of different types of information that you’re trying to pull together from across multiple systems, internal and external data and so on. And so it can be very labor-intensive to pull all that information together unless you automate the process. And so it worked quite well at Lego as well-to-do that. Initially, the IT department was not supportive, but the head of demand planning was supportive of it. And that kind of team leadership role, I think, is arguably the most important one in making this happen, both in terms of providing the encouragement and also ensuring that people are following the necessary control processes.

ALISON BEARD: And at the higher level, maybe C-suite leaders, for example, how do they think about maintaining control over all of this AI development, if control is the right word?

TOM DAVENPORT: I think if we’re talking about generative AI, there’s so many different things that an organization can do with it. And I was just looking at a survey that I did with the MIT Chief Data Officer group symposium for the second annual survey sponsored by Amazon Web Services and data just came in. We have strong focus on generative AI. I think that software engineering was the third most desired use case or most focused on use case in there.

But in another question, 16% said they were banning. There was no authorized use of generative AI in their companies. I think in general, that’s a bad idea, but it at least shows that an organization is paying attention to it. I think almost every organization today should have high level meetings about A, are there some ways that we can use this in our organization? B, what sort of risks and concerns do we have about its use? C, what kinds of policies do we need to put in place to make it more effective? And that’s whether you’re talking about generating code or generating marketing blogs or using it in customer support. Any of those things, I think, requires some deliberation about what’s the organization’s strategy and what controls does it need to put in place.

ALISON BEARD: And are there any examples you could point to from organizations that are finding a good balance between keeping things open and nimble but then also protecting against some of the dangers we’ve talked about?

TOM DAVENPORT: Funny, yesterday I got an email from the chief data and technology officer of a large marketing services company. They identified, I don’t know, five or six key areas in which generative AI could be used within the organization. They said, “In general, our philosophy is to make this a kind of decentralized activity. So we don’t want to apply to heavy a hand, but we want to sort of see what people are doing.”

So they looked at a whole variety of use cases and said, “What’s the status of this particular use case? Is it in production?” And there were very few of those that were. I think in the AWS survey, I was just looking at only 8% of organizations said they had systems like this in production.

“Is it a proof of concept where we can see whether it works or not or is it just an idea?” So I think a kind of inventory of things that are happening around the organization is a very good idea. And they, like many other organizations that I’ve worked with, have decided that the risk of using a public generative model is too high. So we need to do a deal with OpenAI or with Google or whatever provider they are using for these models and say, “We don’t want our prompts to make it into your model. That solves a lot of the problems relative to intellectual property ownership that a lot of companies seem to be worried about.

ALISON BEARD: Yeah. And you also mentioned training. So what types of training do you recommend or what are some examples of good programs you’ve seen?

TOM DAVENPORT: Well, again, it kind of depends on what sort of citizen you want to empower. If for example, it’s generative AI, then you need to tell people that… what types of prompts are likely to be effective for producing code and how to interface with the needed data and other transactional systems that you may be working with. So that probably would be run by an IT organization or an AI organization, or if your company has one. If it’s automation, typically we find that’s being done more by… Some companies have specific automation-oriented groups, but a number of them are primarily focused with process improvement. And being a former process re-engineering person, I’m very interested in the idea that we can produce code that will automate workflows within organizations. But in those cases, you not only need to know something about the tools that are used, but also you need to know something about process improvement. But you can save some real money that way.

If it’s data science that you’re trying to empower the citizenry for, there is some sort of probably statistical training that you’re going to make sure that people already have or you can provide it. It’s relatively easy to find a lot of that stuff in online courses these days, but that, I think, the software is now quite easy to use. And by the way, there’s a generative AI version of that called Code Interpreter. It’s a beta offering from OpenAI now that lets you do basically machine learning analysis with a very simple prompt. You’re basically saying, “Here, use this dataset and here’s the particular variable or feature that I’m trying to predict. And tell me which variables in my data are likely to be good predictors and how good a job can they do.” I did this the other day, a two line prompt for a little data set, a two line prompt gave me three pages of data analysis with a machine learning model applied to it. This is quite astounding.

ALISON BEARD: Wow, that is astounding. How are you seeing companies identify the people who will become their citizen developers? Is it something that they offer this training to their entire workforce, a certain subset of the workforce, or is it mainly a volunteer army?

TOM DAVENPORT: I think it’s mainly volunteers. There are some companies… I mean J&J has a series of criteria that they apply to you once you volunteer. Do you have any experience with data? Are you technically adept, et cetera? Some other companies will take anybody who comes along and figures, even if they drop out, it’ll advance the cause of citizen development if they get at least a little bit of training.

There are some organizations that have certifications. I do a lot of work with Deloitte, and they have certification of citizen developer, citizen data scientists and so on. You have to go through certain levels of training to get to each level. So I think it depends in part on what you’re trying to accomplish. Are you open to a broad kind of democratization of these capabilities or do you really want to be a lot more targeted about it and a little bit more careful about who’s doing the work?

ALISON BEARD: You have talked about sort of hours saved, money saved, and you talk in the article about how important it is to track these measures of value creation. So what are some of the effective ways that you see companies do that? When you say that this company has saved X hours of productivity, or this company has saved X amount of dollars, how did they track that? How did they measure it?

TOM DAVENPORT: Well, yeah. And by the way, I think there’s a big difference between the minutes and hours saved. And I think it was AT&T that said, “Oh, we’ve saved 13 million minutes from our automations.” And my immediate question is, well, what are people doing with all that freed up time?

ALISON BEARD: I think about that Home Depot guy too, like when he automated his job. What happened then?

TOM DAVENPORT: Yeah. I think in his case, he moved on to other aspects of production forecasting and got more detailed about the different product categories they would use it for. But I think in general, it’s not a good idea to just look at hours or minutes saved without a sense of what people are doing instead.

And so I think the smart organizations… I was talking to an individual at PepsiCo lately who is in the finance organization and sponsoring a lot of citizen automation efforts in particular, and he said, “Every one, we try to figure out what’s the economic value of it. Is it costs avoided somehow? Is it a hire that was prevented as a result?” Not too many people seem to get laid off as a result of these things yet, but that could be another source of savings. “Is it making a better decision that’s kind of yielded more sales in a particular domain? So has it increased the bottom line, or the top line, or reduce cost?” The various categories of economic benefit, you can say.

And I think that’s the only way you can really justify a lot of these activities. And typically, you want somebody to certify that and not just have the IT or automation or data science people do it to have the finance organization be behind it as well and kind of certifying the outcome.

ALISON BEARD: Yeah. And so if you’re the person who’s, say, working for a small or medium-sized company that’s very far away from the tech sector and you want to try this, do you talk to your manager about it? Do you experiment with it first and then show the results? As we talked about earlier, is there a danger there that you then just put yourself out of a job?

TOM DAVENPORT: Yeah. I mean, I think employers don’t own 24 hours of our time, so if you want to explore these capabilities on your own time, I don’t see that there’s anything wrong with it, but I wouldn’t take it very far. After you demonstrate to yourself that it’s a workable solution for improving your own productivity, I would talk to my boss about it. And I think I’d say that with generative AI. If you’re going to use it, you should inform your teacher, your boss, your whatever the stakeholder is that you’re going to use it or it’s probably not going to be good for your career in the long run. I don’t really know. I haven’t done enough code generation with generative AI to know how obvious it is, but I’ve read a fair number of term papers that have been created with it, and I can identify them pretty easily now.

ALISON BEARD: Yeah. Okay. So last question, big picture, does this totally change digital transformation as companies have previously thought about it?

TOM DAVENPORT: I think it does in the sense that, okay, you can have a small group of IT professionals doing your digital transformation, or you can have, I don’t know, 50, 60, 80% of your people working at it. Obviously, you’re going to get a much faster transformation if you have that democratized approach. And we don’t have too many examples yet of really that broad scale adoption, but it is happening at that vendor that I mentioned previously. I think it’s going to be happening at BMW when they finish training their 80,000 people in it. So I think the new digital transformation is going to have to be a lot more democratized than the previous version was.

ALISON BEARD: Well, terrific, Tom. Thank you. I came into this conversation a little bit scared about having to become a programmer, but it sounds like really exciting development for all organizations. So I’ll get on board. Thanks so much for talking to me today.

TOM DAVENPORT: Thank you.

ALISON BEARD: That’s Tom Davenport, professor at Babson and co-author of the HBR article We’re All Programmers Now, and the book All-in On AI: How Smart Companies Win Big with Artificial Intelligence.

We have more episodes and more podcasts to help you manage your team, your organization, and your career. Find them at hbr.org/podcasts, or search HBR in Apple Podcast, Spotify, or wherever you listen.

This episode was produced by Mary Dooe. We get technical help from Rob Eckhardt. Our audio product manager is Ian Fox. And Hannah Bates is our audio production assistant. Thanks for listening to the HBR IdeaCast. We’ll be back with a new episode on Tuesday. I’m Alison Beard.

The original article can be found at: Harvard Business Review for Managed IT