You Ask, I Answer: Siloed Data Impact on AI?

In today's episode, Heidi asks, are companies challenged in their adoption of AI by silo data. I would say that it is certainly a contributing problem, but I think siloed data is a problem, period, right? Regardless of whether we're talking about AI or not. If we think about artificial intelligence really as just like a fancy kitchen appliance, um and it does really cool stuff, think about what silo data would look like. Suppose um you were trying to cook in a kitchen, and uh there were some ingredients you had, and there were some ingredients that were in another part of the kitchen, and you weren't allowed to go in that part of the kitchen, right?

Like, oh yeah, the uh the stuff in the refrigerator, you're not allowed to go in there. You you can't go in the refrigerator. How the heck am I supposed to make bread if I can't get it, you know, the eggs or the milk or the yeast or things like that? Um, I've got flour here, and then you start getting protective. Well, if I can't go in the refrigerator, you can't come in my pantry and you you know you can't have flour, sugar, or salt.

Obviously, it's gonna be a really rough day in your in your kitchen if you start having these silos, these organizational divisions that prohibit people from sharing ingredients. Data is an ingredient, and again, AI is just a fancy appliance. Which means that if you have those silos, you've got problems that um AI won't solve, right? If you have it, doesn't matter how fancy your new blender is, if you've got if you're not allowed to go in the refrigerator, there's a lot of things you can't cook, period. Blender or no.

And so from a structural perspective, companies need to get rid of silos as much as possible within you know practical and regulatory limits, if they want to be able to make use of the data for anything, not just for use with AI, but for anything at all. If you're building a data warehouse or a data lake, if you're doing even basic, you know, pre-machine learning stuff, if you're just doing basic statistics, uh, exploratory data analysis, you need to have access to the data to do that. And if there are silos, uh, it's gonna be a it's gonna be a rough time. There are there will be things like you know, if you're not allowed in the in the in the freezer section of your of your house's kitchen, there's gonna be some things that you're not allowed to do that and that you will need uh you will need access to. So here's where artificial intelligence can help.

Sometimes if if we make an AI project fancy enough and uh flashy enough, uh you know, it it's the shiny new object in the room. Sometimes that can help break down organizational resistance. Uh if AI is a strategic priority at your company, you can go to somebody and say, like, oh, yeah, yeah, yeah, I know we're you know, normally we're not gonna access your your sales data or whatever. But for this project, uh, we'd like to make an exception. And depending on the benefit to that division of the company, depending on the visibility at an executive or a stakeholder level, sometimes you can use AI as an excuse to dig into um those other silos of data and get stuff out of them.

Uh this happens a lot. Uh we've had this happen a lot with um uh analytics projects, big analytics projects, where uh ironically, uh as a consulting firm, Trust Insights would have access to more of the company's data than than any individual one department did because we were an outside neutral third party. And so we're just like, oh yeah, we're just gonna use this data for this project. Meanwhile, uh we had better visibility into the entirety of what was happening at a company and be able to share back with those divisions. Hey, here's what else is going on at the company.

It's kind of like kind of like being the data and AI equivalent of a bartender, right? Everybody comes up to the bartender and confesses their their woes individually, and you, as the bartender, you have you know, hear everybody's stories and go, oh yeah, yeah. And and you think, you know, Ralph here has got the exact same problems as Bob over there, and you know, Shiva's over there, she's got her problems and stuff, and and you all don't know that you each have the solutions to each other's problems because you don't talk to each other, you just talk to the bartender. So AI can be used as an excuse to get into other silos, and then ideally what you do is you show benefit to sharing data that goes above and beyond the scope of the AI project itself, so that it persuades those uh those silos, those departments, like, hey, if you share your data, things will be a lot easier for both groups, both groups will benefit. The worst case I've ever seen of this uh was just blew my mind.

Um, we had a uh B2B tech company uh a few years back as a client, and we were called in to build a model of their marketing uh data, uh combining marketing and sales data to help them uh essentially figure out which channels mattered the most. When we got in there, we were told here's the marketing data. For regulatory reasons, there's some data we can't get about our own company that um we can't like the marketing department saying we can't get it's not that we won't share it with you we can't get a hold of it can you see if you can get it from the outside and we were able to do that and then we asked for the sales data so that we could collect uh calibrate the uh marketing data with the sales data say okay if you have all these things and this is the outcome you're after and the VPSL is like nope can't have that data we're like why not because marketing's not allowed to see sales data like how how do you get anything done then like how do you communicate to marketing hey you know these programs are driving leads or not driving leads and they're like we don't I'm like so you just wing it and accept whatever leads marketing sends you and it's like no no no we don't we don't do that. Whatever marketing sends is incremental our our sales guys all just cold call everybody all day I'm like I feel like I feel like that might not be the best way to do things. It turns out this came out um after our engagement that sales was so poor at its job their closing rates were so bad that uh they didn't want anybody to know just how bad things were internally there's their sales closing rate for you know good sized commercial and enterprise deals was something like around one percent of every out of every 100 opportunities that were teed up there's the sales department closed one of them and so there was a lot of obviously house cleaning and personnel changes and things we didn't have anything to do with that because we were long gone to that point but I remember reading in the news about this company because it's pretty well known company that uh they they had run into some revenue issues um and the uh at a few quarters, and I'm like, huh, I wonder if that's because y'all are flying blind and have no idea what you know, the left hand has no idea what the right hand is doing.

So there are definitely challenges posed by siloed data. AI is no different than any other function or any other technique used to turn data into insights. It is hampered more by missing data. Uh, but if your company's got siloed data and and rigorous boundaries between departments, it's got problems already, right? And AI will not solve those problems, it will just make those problems bigger and faster, right?

That's what AI does, makes things go faster and bigger. And you know, if you have solutions, it will make your solutions faster and bigger. If it makes you have problems, it will highlight and make your problems faster and bigger, too. So really good question. Thanks for asking.