Can’t I do what Orita does using Claude? We get this question a lot, and we understand. AI tools are changing and getting better quickly. If an LLM can write your email copy, generate images, analyze data and learn to know what you want over time, why introduce a specialized vendor to handle segmentation?
Short answer: Claude has read every public article ever written about email marketing. It may have even read your customers' engagement history. It has not built a global model with thousands of brands inputs - and billions of customer journeys through email, SMS, etc. -, dedicated solely to answer: “Who wants to hear from you today?”
What Claude is great at
Creative suggestions. Copy variants. General best-practice advice. Analysis of your data. If you want to brainstorm campaign themes or rewrite a subject line ten ways, an LLM is a genuinely useful tool. Our own engineering and GTM teams use Claude and Cursor daily and get real leverage from them.
That kind of work is what LLMs are built for: pattern-matching against the public corpus of marketing advice, or handling a specific and incredibly well-specified task.
What Claude can't see
Email engagement data is proprietary. You have to give Claude access to your Klaviyo data via MCP. This expands its view from ‘all publicly available marketing advice’ to include your performance data.
It doesn’t get to see, or analyze, what’s worked or not at thousands of other brands. So it is limited to only the data you feed it.
A model gets good at CRM marketing the same way it gets good at anything else: by training on enough variation to learn the underlying dynamics rather than memorizing one company's playbook. Claude doesn’t have the breadth of specialized information to do this.
But Claude can use *my* data, and why a single-company models overfit
Claude can’t analyze the (typically) hundreds of millions of data points that stream through your Klaviyo account inside the context window. This means while it has access to your data, it doesn’t have the capacity to use the data the same way Orita’s models do. This means you can likely easily recreate rule of thumb segments like “90-day engaged” (Klaviyo’s native AI tool already lets you do this). But you can’t use Claude today to create an ML model that processes all your customer data and then tells you who to message.
With Klaviyo's new MCP, Claude can see your CRM data. It can see what your brand has tried, and what's worked vs not. But it is only working off your data, and is limited to only understanding what your team has done.
Our earliest single-company pilots produced overfit models. Too little variability in strategy. No ability to generalize. A single brand runs at most a few different playbooks in a year, so the model learns the playbook instead of the underlying dynamics. This meant the model wasn’t helpful.
Your Orita model leverages data from thousands of brands, tens-of-thousands of strategies, and billions of customer journeys. When we build your model, we’re actually fine-tuning a global model with all of these weights. These “priors” are the difference. They allow you to get the absolute best segments; this is the only way to truly optimize your audience by sending to the right customers at the right time.
Claude doesn’t have access or enough context to be able to do this.
We only do one thing at Orita: AI Customer Segmentation
Half of our team are ML engineers, data engineers, and analysts who own the domain expertise behind the product. They monitor drift, rebuild pipelines as data volumes grow, and troubleshoot edge cases continuously. They are experts that go deep in this one area.
In order to get Claude to build you a model, even with reduced data, you’d need to have that depth of expertise to know what to ask it to do. But, again, Claude doesn’t have the proprietary data that we do.
Where this matters most: bigger customer list = more impact from better segmentation
Larger brands have more data, more strategic variability, and more revenue at stake on every decision. The performance gap between an expert system and a mediocre one becomes material quickly, and it compounds.
The real work
The real magic in lifecycle marketing is combining two questions:
Does she want to hear from you right now? This is the question we answer. It's a proprietary-data problem, and it's the one we've been working on every day for years. Klaviyo named us their Premier Partner for Segmentation for a reason.
What do you want to tell her? This is your team's expertise. Your brand, your offer, your creative. Claude can absolutely accelerate this side of the work. And so can working with top-tier agencies (we know them if you want an intro to one of the 70+ that work with us)
Orita answers the first. Claude is a great collaborator for the second. Use each for what it's built for.
