Why Plugging Claude Into Your Data Won't Work — And What You Need Instead

Someone said to me yesterday that their team are trying to plug Claude into Shopify. And it was just like a weird echo of like, maybe not almost 10 years ago, when people were just starting to get this new tool, which seemed like an easy way to answer the questions you had. But you can't shortcut the bit in the middle, which is really important. Getting your data structured in a way that's going to power those sorts of tools.

The tool was Power BI back then. Now it's Claude. And I keep seeing the same mistake play out. To understand why it doesn't work, you need to understand how the modern data stack actually developed and what that middle bit actually is.

‍

Stage 1: Getting data out of its silos

The original bit was like, how do I connect my sources together? How do I even get into that source? Whether it's like Salesforce or NetSuite or Google Ads, whatever the source is, how do I actually go in and get that data out? Because often these sources will silo themselves off. So if I want to report on Salesforce, I can go and do that in Salesforce. But if I want to understand what happens to a customer once they come in through Salesforce and then start becoming a transactional customer, that's in NetSuite. All of a sudden I've got these two different systems that say two different things about the same customer.

So the first stage is like get a team of data engineers, send them in, build all these pipelines using Python, AWS, that kind of stack to go in, connect to the data, bring it into a central place. Whether that was like an on premise SQL Server, something like that was like the very first phase.

And like what we say now, anyone that wants to do that is like, think about the value, don't think about the technical components of that — because just bringing them in to SQL Server, there's no value for that. You're just having them in one place, doesn't really matter.

‍

Stage 2: Cloud warehouses change everything

The second phase of that is doing it in a more kind of cloud based, less engineering heavy way. So using ELT tools like Kleene to go in, connect to the resources and then putting them into a cloud data warehouse like Snowflake, BigQuery, Redshift and using the power that they have that is much more powerful than what was available before, it made it a lot easier to store massive volumes of data that became really cheap.

So all of a sudden it's like, right, instead of cherry picking like a few components from each source, we'll just bring everything in, put it all in one place, try and build this kind of data mesh, data lake type architecture. And at that point it started to become a bit more stakeholder facing. So you'd have more dashboarding, you'd have Power BI, Tableau, things like that, probably all still managed, built by your technical data team. So engineers, analytics engineers, BI analysts.

‍

Stage 3: The end user finally gets access

We're now in probably phase three, which is basically phase two. But you're having a much more easy to use interface that stakeholders — C suite, CFO, CEO — can interact with data in a much easier way. Whether that's more modern BI tools like Sigma or whether it's with something like Claude or KAI Assistant in Kleene, where the very end user can actually finally access the data they want to access.

Which was like the thing that started phase one — your finance leader wants to understand what's happened with that customer once you acquire them and they've become transactional, to understand their revenue performance, their churn potential, their growth potential, everything like that. So like 10 years ago when they asked for this project to start, they can actually now get what they want, which is just a really easy platform to interact with and understand your performance of the business, which is what it's all about.

‍

Why you still can't skip the middle

This is where the Claude-into-Shopify problem comes back. The exact same thing is happening now as happened a decade ago. Someone sees a powerful new tool and thinks it's the shortcut they've been waiting for. But the data in a transactional system like Shopify is structured for Shopify's operational needs, not for analysis. Plug Claude straight into it and you're not getting intelligence out of your data. You're getting Claude making its best guess from data that was never designed to be analyzed.

The bit in the middle — getting your data structured in a way that's going to power those sorts of tools — is not optional. It never has been. Every generation of better tooling at the top requires better foundations underneath it, and that has been true since stage one.

‍

What stage 4 looks like

I think stage four is where you kind of abstract away, like the need for someone to have a question to ask, where you just get told the things you need to know, where the system — you know, it just automatically looks at all the platforms you've got and goes, okay, we're going to do all of this for you and we're then going to go and tell you what your best customers are, the markets to go and investigate, even what tools you could replace your HubSpot with, if that's like a potential issue in your stack.

It kind of takes away even needing to have that person in the C suite or wherever they are in the business to go, oh, I want to understand the growth of this product line. It would just all be happening automatically and then the right person would get told at the right time — to the point where they wouldn't even need to know some of this stuff. It would just happen, just automated analytics going on with ideally a human somewhere checking it.

You don't even need to know what to ask. It tells you what you need to know without needing to ask it.

That is where this is all going. And the businesses that get there fastest will be the ones that didn't try to skip stages one, two, and three to get there.

‍

Stay in the loop

Subscribe to the Kleene.ai newsletter and be the first to hear about new guides, data trends, and product updates.