Building a product also means consciously deciding what stays and what needs to be let gone when it has completed its purpose.
At Encatch, we recently made the decision to sunset AI Studio, a set of AI-powered features that allowed users to upload raw feedback data and ask an AI model to analyse it, answer questions, and generate insights or charts.
It was a useful feature when we built it. More importantly, it was built to solve a genuine problem.
But the way people use AI has changed considerably since then. And as we looked at where Encatch was heading, we realised there was a better way to solve the same problem — one that gives users more flexibility, better context, and significantly lower costs.
So, here's why we built it, why we're retiring it, and what we're doing instead.
What was AI Studio?
AI Studio was our way of allowing users to interact with their feedback data using natural language.
Instead of relying only on the summaries and dashboards available within Encatch, users could provide their raw data and ask questions such as:
- What are the most common complaints?
- What themes are emerging from the responses?
- How does feedback differ between different user segments?
- Can you identify the biggest reasons for dissatisfaction?
- Can you turn these findings into a chart?
AI Studio had two ways of doing this: External Insights and Internal Insights.
External Insights
With External Insights, Encatch acted as the bridge between the user and an external LLM.
When a user submitted a query, Encatch processed and simplified the request where possible, with the intention of reducing unnecessary token usage. The resulting query was then sent to an external AI service, and the generated response was displayed back within AI Studio.
This made AI-powered analysis available directly inside Encatch without requiring users to set up their own AI tools.
Internal Insights
Internal Insights followed the same basic workflow, but with a different approach to data privacy.
Instead of sending the feedback data to an external AI service, the analysis was performed using a local LLM tool.
The local LLM tool was a prerequisite for using Internal Insights. Once configured, the processing happened locally, so the feedback data did not need to be uploaded to an external server for the AI analysis.
The problem we solved back then
Encatch already provides a dashboard where teams can see summaries of the feedback they collect.
But summaries only answer some of the questions a product team might have.
A product manager may want to ask:
- What are the biggest complaints from users on our mobile app?
- What themes are emerging among customers who rated us poorly?
- How does feedback differ between new and returning users?
- Can you turn these responses into a chart?
- What should I investigate further?
We wanted users to be able to go beyond predefined summaries and ask their own questions of their feedback data.
That was the problem AI Studio was built to solve.
So, why are we sunsetting it?
The problem itself hasn't gone away. If anything, it has become more important.
What has changed is where the AI capability should live.
When we first built AI Studio, using an AI model directly inside the product made sense. Today, users already have access to powerful AI tools and, importantly, they often have their own preferred models, contexts, instructions, and workflows.
Continuing to provide the AI layer ourselves meant that Encatch would effectively sit between the user and an LLM provider.
That comes with a cost.
As model capabilities have increased, so has the cost of processing large amounts of feedback through frontier models. For users working with substantial datasets, the cost of running these analyses through Encatch could become significant — potentially making the AI experience considerably more expensive than we wanted it to be.
We didn't want to solve the problem of:
"How do I analyse my feedback?"
by creating another problem:
"Why is analysing my feedback so expensive?"
So instead of continuing to build and maintain our own AI execution layer, we decided to change the approach.
From providing the AI to connecting you to the AI you already use
Rather than asking users to use an AI model through Encatch, we're moving towards enabling them to use their existing AI tools to work with their Encatch data.
This has a few important advantages.
More choice
You aren't locked into the AI model that Encatch happens to use.
You can use the LLM tool you already work with and are comfortable using.
More context
Your existing AI environment may already contain context about your product, customers, roadmap, terminology, or internal processes.
That context can make the resulting analysis more useful than a standalone query sent through an AI interface that knows nothing about your product.
Lower costs
Instead of Encatch absorbing the cost of running AI models on your behalf, you can use the AI resources available through your existing tool.
For many use cases, this can reduce the incremental cost substantially, potentially from hundreds of dollars to just a few dollars, depending on the AI tool, model, and volume of data being analysed.
Less infrastructure for us to maintain
AI models and APIs are changing rapidly.
By separating Encatch's feedback infrastructure from the AI execution layer, we can spend less time maintaining model integrations and more time improving what we believe is Encatch's core job:
helping you collect better feedback and making that feedback accessible when you need it.
What happens now?
Instead of making Encatch the AI itself, we're focusing on making Encatch work with the AI tools you already use.
This is the direction we're taking with two capabilities:
Analyse with AI
With Analyse with AI, Encatch prepares your feedback data and generates a ready-to-use prompt.
You can take that prompt and data to the LLM tool of your choice.
This means you can:
- Use the model you're already familiar with
- Provide additional context about your product
- Ask follow-up questions
- Change the analysis
- Request different visualisations
- Continue exploring the data beyond the initial question
In other words, instead of Encatch deciding which AI you should use, you get to decide.
And because you're using your existing AI setup, the incremental cost can also be considerably lower. Depending on the model and usage, what could otherwise become a $50 AI expense can potentially be handled within a much smaller existing AI allowance or usage cost.
Encatch MCP Server
We've also built support for MCP, allowing compatible AI tools to interact with Encatch directly.
With this approach, you upload your feedback data directly to your LLM tool and use the available MCP tools to retrieve the insights you need from Encatch.
This takes us closer to the experience we originally wanted from AI Studio: asking questions about your feedback and getting meaningful answers — but without requiring Encatch to own the entire AI layer.
Why we think this is the right direction
Sunsetting a feature isn't something we take lightly.
AI Studio was built because we genuinely believed there was value in letting users ask AI questions about their feedback.
We still believe that.
What we've changed is our approach to delivering that capability.
We don't think Encatch needs to be the place where every part of the AI workflow happens.
Our job is to make feedback available, structured, and useful.
Your AI tool can then help you explore it in whatever way makes sense for you.
That means fewer constraints for users, lower infrastructure and maintenance overhead for us, and an AI experience that can evolve alongside the rapidly changing LLM ecosystem.
Sometimes, the best product decision isn't adding another capability but moving it to where it can work better.
And that's what we're doing with AI Studio.