Human Intelligence is an occasional series to help PBJ readers better understand how they might use AI to improve their businesses. We'll highlight takeaways and best practices shared by people who are adding AI to their products or who study the technology.
Dominik Facher has been working in the field of artificial intelligence for well more than a decade, starting in his days in university working on natural language processing models. Now he is chief product officer for go-to-market software maker ZoomInfo Technologies (Nasdaq: ZI).
He joined ZoomInfo when the Vancouver company bought the AI startup where he was vice president of product and engineering. It was a $575 million deal that brought Chorus.ai into ZoomInfo and created the basis for some new AI-based tools by ZI with capabilities such as meeting summaries or proposed emails for follow-ups or action items.
For Facher, the lure of AI has always been building tools that can help humans work better. In the world of B2B sales maybe this means software that can automatically fill in the growing number of fields in Salesforce or other tools. Or it’s taking all the data known on a prospect and connecting dots to show when it makes sense to start sales talks or whom to skip. He doesn't see the technology replacing salespeople.
We talked with Facher as part of Human Intelligence. This interview has been edited for length and clarity.
You’ve spent your career in AI. What is your take on the current hype around this technology and the influx of products?
We’ve been in AI a long time, and I think it all goes through cycles. And what’s interesting about generative AI is its mainstream awareness that some problems can be solved through AI that people thought were not possible before. It’s kind of the next evolution, and that’s really exciting. You see all those products that got released by so many different companies that I don’t think would’ve existed without this pressure. It really was lighting a fire under an entire industry, and I think that’s to everyone’s benefit.
When you talk to people, especially in the research side that have done this for years, like people at Google, they are not surprised. There’s probably a little too much hype out there than the reality of what happened. But, we've got the next evolution. Now what I expect the hardest problem that (the AI industry) will have to solve now that the tech got to the next level is nobody has figured out the user interface yet. How does the technology connect with the human? In enterprise software you need 99.999% precision and accuracy, and GPT gives me 80% and you don’t even know when it’s wrong. So those problems, it’s going to take us five years or more before we fully have that incorporated in.
You are putting AI products out now. How do you balance the limitations of the current tech? Or the privacy and ethics issues?
The first part is we have models and AI that need to be incredibly accurate. We’re not using external models for that, we keep that in-house and we control every aspect of it. And that is what our customers demand from us. They want to understand how it works and they want predictable outcomes. And then we (also) use generative AI for either our own or external models in areas we want to make it easier for a human that would have to (otherwise) do the work, like summarize a meeting. If that is 80% right it’s better than zero percent. But, before you send that out to your customers you are probably going to read it over. You’re probably going to make some changes (based on what you remember from the meeting). The same thing for generating email. (Next) we have a focus group that is specifically on (ethics and privacy issues). We have a legal team and a chief compliance officer to drive a lot of this type of discussion. It’s about the actual data that feeds into the models and understanding and training the output, understanding any changes of this. If the model changes, maybe the output changes, having basic controls to test against and always validating the answers. And then we have a forum where we go and look at new types of use cases.
When we think about where to invest AI, (we ask) can we actually put it into the context of is it possible? In many cases the answer is yes, or directionally yes. Then (we ask) can we actually produce the data in a compliant way? Can we predict the outcomes? And we go through that (decision-making) flow. In traditional software development those things are easy. In AI those questions are a lot harder. So we have a higher barrier to where and when we invest compared to traditional software.
How should business leaders be thinking about this tech?
What we see from most, especially generative AI, is it can get to 60%, maybe 70% version of the output done decently well. So, it’s really kickstarting that excitement. What we did at ZoomInfo, and I would recommend it to any business, is you have a set of beliefs of how the world works and how easy it is to do something and what business problem you can solve for your customers. And AI is a really good catalyst for questioning all of those assumptions.
I’ll give you an example. We thought there’s a couple of really hard data problems that are going to be insanely costly to solve with previous types of models or doing it manually. It turns out we can in two or three days solve those problems with generative AI to a 60% version and once you are at 60% getting it to 80% is not that difficult anymore. So it’s really about seeing it as a catalyst. I think these days everyone should have a GPT account and try it out, even just for five minutes a day. I’ve seen people journal in a GPT like application. I personally have all my notes fit into a GPT and then I can ask it how a two years ago version of me would have answered a question. I think even just getting used to the power and what these things can do for even five minutes a day, it might not be in the first week but soon (it will help) generate ideas that will help every business owner, every professional come up with new ways of thinking about things.
