AI Doesn't Fix Bad Data, It Weaponizes It with Elio Narciso

Elio Narciso (0:00)

I do think that RevOps is uniquely qualified, because they are like, you know, they've been some of the first teams to really use AI for practical applications that have like real life output and consequences and outcomes.

Danielle Parker (0:27)

Welcome to Human-First, the GTM Hiring Show. I'm your host, Danielle Parker, Head of Marketing at Captivate Talent. Everyone is deploying AI across our organizations right now. We've got reps vibe coding their own CRMs and workflow agents. We got ops building different enrichment workflows, and leaders are under pressure from above and below to move fast and make AI make them more efficient. Very few people are asking the question my guest today thinks is the most important one.

What is the data that all of this AI is sitting on top of, and what happens when it's wrong?

Because the answer in his experience is not that it gives you bad results. It's that the AI confidently scales that data across the entire go-to-market organization. Elio Narciso is co-founder and CEO of Scalestack, the GTM data infrastructure platform used by companies like MongoDB and Typeform to make sure the data their agents are working with is actually trustworthy. Welcome, Elio. Thank you so much for joining us today. Obviously, we've had a number of conversations across the last few weeks about AI and RevOps. I kind of want to dig right in with this line you have. I want to start with because I think it frames the whole AI conversation in a really interesting way, that AI doesn't fix bad data, but it weaponizes it.

Where does that conviction come from? First of all, I'm so curious. Do you have an example of what that actually looks like when it goes wrong at scale?

Elio Narciso (2:06)

Yeah, I think that, well, experience, for sure, either individual or as managing teams and customers, what that means is really that, well, first of all, if you ask AI, Cloud or ChudGPT, anything, they will always give you a confident answer, even without much content or inputs, because they are trained on other data. And so they take what's probably a best guess on the question that you asked them.

If you then plug AI to existing data, let's say the data in your Sierra or like in a spreadsheet, AI will assume that that's an input that it needs to consider to do whatever it's asked to do, whether that's putting together an email or recommending like an outbound strategy, sequence or preparing you for like a call that you have to do with a customer or a prospect based on that data. And so if the original data is not correct, if the original data is full of holes or mistakes or like things that are not up to date.

Danielle Parker (3:37)

So like classic early stage CRM.

Elio Narciso (3:40)

Or even later stage, I mean, like we only work primarily with like mid-market type of customers and the situation is probably worse because you have like a legacy amount of data that is huge, that is valuable for the most part, but that it's also been stratified over the years over the go to market motion, over like, you know, changed like ICPs and strategies. And so you easily find that like the biggest problems are a larger orgs where like, you know, you have bigger systems, you know, let's say like people use Salesforce as their main system of record that like most enterprises. And so they, you know, they will have a lot of bad data, a lot of duplicates, a lot of like information that is not up to date and it's difficult to maintain up to date.

And so that's what the problem of like AI weaponizes bad data, because to begin with, it will start with anything and just like run with it and run confidently. And so confidently run like a sequence, an email, a research, et cetera. And so if you plug bad data, it just weaponizes because the output may look credible at first, but it's based on the wrong inputs.

Danielle Parker (5:09)

Yeah.

As an aside, I remember seeing a tweet, I think it was from the Secret CFO on Twitter about finance data. And it was like, usually when you were looking at like a spreadsheet or something, there were hallmarks of like, okay, this person does not know what they're doing. And it was a little easier to suss out. But now AI, like you said, makes everything look so confident. So it's so hard to find that those hallmarks of like, okay, like this isn't right versus AI being like, yes, this is correct. And then you dig a little deep and you're like, wait, you made up like 70 people.