Building vs. Buying Enterprise AI Tools: Advice from the Boardroom

Bill McNabb — Former CEO and Chairman, Vanguard. Bill led Vanguard from 2008 through 2017. He currently serves on the boards of IBM and UnitedHealth Group, giving him a front-row view of AI strategy at two of the largest enterprises in the world. He has been a Valence board member for three years.

Parker Mitchell — Co-Founder and CEO, Valence. Parker leads Valence, the company behind Nadia, an AI coach deployed across dozens of Fortune 500 organizations to support leadership development at scale.

Should your company build its own AI tools or buy from focused external vendors — and how do you navigate the tension between your CTO and your business leads? Bill McNabb sits on the boards of IBM and UnitedHealth Group, and advises several startups and smaller-cap companies. In this clip, he shares the framework he uses across those boardrooms for thinking about AI investment allocation — and explains why even Vanguard's 35% software engineer workforce couldn't out-build a company purpose-built for AI coaching.

Key Takeaways

• The board-level tension is speed and bet-placement — not whether to act. Across large public companies and startups alike, boards are pushing leadership teams to stop deliberating and go try something. The question is no longer whether to invest in AI, but how fast to move and where to place the bets.
• Even 35% engineering teams can't out-build purpose-built AI vendors. Vanguard — where a third of all employees are software engineers — recognized that its internal team could not match the speed and focus of a company designed specifically to build AI coaching tools. Scale and talent alone don't close the gap against specialist vendors with a narrower mission.
• A portfolio approach balances internal development and external partners. The right AI investment strategy is not all-internal or all-external. It is a deliberate balance: encourage the CTO to build where internal development makes sense, while simultaneously giving business teams the freedom to go deep with a small number of focused external vendors. Both tracks run in parallel.
• Going deep with a few external partners is not that expensive — and produces insights you can't get internally. Bill's practical advice: whatever capital allocation you have for AI experimentation, ring-fence a portion specifically for going deep with two or three external firms. The cost is manageable, and the quality of insight from purpose-built partners consistently outpaces what internal teams can generate on the same topic.

Questions This Clip Answers

Should large enterprises build their own AI tools or buy from vendors?

Bill McNabb's answer is both — but with an important caveat. Unless a company is a deep tech firm by design, its internal engineering team will not be as nimble or focused as a vendor built specifically for a given AI application. Even Vanguard, with 35% of its workforce composed of software engineers, found it could not match the speed and capability of purpose-built AI partners in areas like coaching. The right approach is a portfolio: give the CTO a track to build internally while giving business leads the ability to experiment with focused external vendors simultaneously.

How should leadership teams navigate the tension between CTOs and business leads on AI?

The tension Bill McNabb observes most consistently in boardrooms is between CTOs who want to build everything internally and business leads who want to move fast with external tools. His recommended resolution is not to choose one side but to run both tracks in parallel — explicitly authorizing CTOs to develop while equally authorizing business teams to experiment with external vendors. Boards, in his experience, are increasingly pushing for this parallel approach rather than letting internal build cycles slow down the business from gaining real AI experience.

What does a smart AI investment portfolio look like for a mid-sized enterprise?

Bill McNabb advises leadership teams not to overthink AI portfolio construction. Every company has a capital envelope for these experiments — some large, some tightly controlled. The key discipline is ensuring a portion of that envelope is deliberately allocated to going deep with a small number of external partners, not spreading it thin across many tools. Two or three focused vendor relationships, pursued with genuine depth, will generate more useful organizational insight than either a broad internal build program or a wide assortment of shallow vendor trials.

Full Clip Transcript

What are the differences in the tenor of AI conversations at the board level from 12 months ago to now?

Parker: Can you share maybe what are the differences in the tenor of conversations from 12 months ago to now at the board level?

Bill: What's really interesting is I have the privilege of serving on two very large public company boards — IBM and UnitedHealth. As you would expect, AI is very topical inside the IBM boardroom: Watsonx, and all the work being done there, especially inside IBM itself. At UnitedHealth, with the vast scale we have, we think about AI and what it could do not only from a productivity standpoint but down the road from a clinical standpoint. There's a lot of work going on.

But what's really interesting is I also sit on boards of several startups and smaller-cap companies, and we're having the same discussions. The big tensions are how fast to go and where to put your bets. In most of the discussions I'm involved in, what we as board members are doing is really encouraging companies to not talk about it forever, but actually go do something. Find use cases that really make sense for their particular business and go try something.

In the larger companies in particular, there's becoming a tension between business leads who want to go try things and chief technology officers who are saying, "Let us build it for you." What we're doing in the boardrooms I'm in is saying to the CTOs: great, go develop. But we're also encouraging business teams to go experiment with people who are maybe a little deeper on particular topics. At Vanguard, as an example — and I'm not on the board there anymore, but talking to my former colleagues, they've been a very early adopter of Valence. Love it. It's deployed through probably about 80% of the company at this point.

Parker: We checked — 16,000 users.

Bill: Yeah. 16,000 out of 20,000 employees. Pretty remarkable adoption. We also have a company called Writer, which is a startup doing content creation. And the CTO has four or five big projects he's driving development on. For companies with those kinds of resources, I love that approach. For smaller companies, I think it's finding people like Valence — who can really solve a specific problem for you, really quickly, and give you real experience with AI — that makes the most sense.

How would you advise a leadership team to navigate the build vs. buy tension over the next few years?

Parker: I've heard other people talk about a portfolio approach — some internal, some external, some existing vendors, some new vendors. If you were giving advice to a leadership team on how to navigate that over the next couple of years, with all the internal tensions that come with it, what would you say?

Bill: I actually wouldn't overthink it. Every company has a certain amount of capital to deploy — in some it's a large amount, in some it's really tightly controlled. The thing is to make sure you actually do have a balance. The one thing I'm pretty convinced of, unless you're a deep tech company yourself: no matter how good your engineering team is — and let me step back. At Vanguard, 35% of our employees are software engineers. Most people don't think that. They think you're an investment firm. We are an investment firm, but 35% of our population are engineers. And our engineering team is good. Really good. They think they're even better than really good.

The truth of the matter is, we can't be as nimble and agile on things like AI coach development as a company that's designed to do it. As business leaders, the advice I would have is: make sure that whatever the capital allocation you have for these experiments, you've got a piece where you can pick a couple of firms and go really deep with them. It's not that expensive. And you're going to get insights that you won't get from your own teams.