Stamped is an e-commerce review collection and embedding service. We have built out a robust connector to fully integrate Stamped reviews into the Artifact connector library.

In order to support specialized metadata like topic tags, we had to make some minor adjustments to the schema in order to support our customer's ability to filter and analyze by their taxonomies and tags they have created inside of Gladly. You can now add any Gladly topic tag into a filter operator.

We have built out a robust connector to fully integrate Slack channel messages into the Artifact connector library. It's not perfect yet! But we're excited for the direction it's headed. This integration works best with automated bot messages (e.g. Zapier moves survey results from Typeform to #product-feedback every day) and with customer shared slack channels. The connector still has a hard time extracting customer feedback from internal employee cited anecdotes.

This one is kind of embarrassing… we were missing an entire error page for 503 errors. :grimace: But not to worry, we've gone through and enhanced several error states, and made some of our error handling clearer. One of my favorite additions with this release is that if there is a critical problem in the AI pipeline in our Inspect feature, rather than show no new insights to you, we default to the previous day, and give you a heads up that we're running your new insights.

If you're like us, you use Grain every single day to record conversations with customers and team members! Well now you can include customer research, QBRs — whatever you use Grain for at your business, inside of Artifact. We'll automatically detect voice of customer and include in the daily analysis.

Topic modeling is a tough concept. There's no "perfect" boundary for what should or should not be in a topic because it's a spectrum! Each cluster has cluster candidates that are more or less a part of a particular cluster. Well, many of you gave us feedback that we should include that confidence or relevance in the sorting of verbatims, rather than defaulting over to recency.

We agree that it's a much better way to scroll through the verbatims pane. But you can also quickly toggle to recency if that's important to you.

Remember our document serialization machine? It chunks, queues, and passes documents into our inference pipeline, for every new document everyday. Well it wasn't very performant after we put the first version into production so we had to crank it up a notch. We made some massive improvements to its distributive qualities by introducing Dask to the document chunking.

This is a big day! We're releasing V1 of Reports. Reports is a self-service reporting tool for you team. You'll be able to build powerful reports using any of our templates, in just a few clicks. You can also save any of the analyses you build in Explore straight to a report that can be shared with other stakeholders.

Sometimes you don't want to just filter by people-related metadata fields. Sometimes there are fields that are attached to a conversation, but don't belong as an attribute to a particular customer. Well now you can filter by those fields and power your analysis in Explore with document and user metadata.

We've added Kustomer as a data connector. Now you'll be able to bring in all of your Kustomer support data.

S3 is a powerful tool. It can be a temporary holding place between transformations, or it can be long term storage for complex objects. We think they're pretty neat, and we know they have a very key place in your datalake strategies. So we built an S3 connector so you can easily pull, transform, and collate data from your datalake, and directly integrate it with Artifact.

BIG change to our topic model today. We already had a one of a kind topic model, but we made some significant enhancements to the topic model that you're going to love. Here's some of the highlights:

• More coherent topics

• More cohesive topics, less overlap— or near duplication

• Better insight detection over longer time series

Our insights pipeline changed in a pretty significant way today. We enhanced our anomaly detection to understand detect seasonality and cyclical patterns. We no longer pick up on annoying cyclical changes as anomalies. We also finally added trends into production. Trends will take the place of emerging themes, but you'll still be able to pick up on emerging topics that come from zero or low values.

One of the things that drove us nuts internally is that the grammatical representation of L1 topics sometimes needed to be pluralized to even make sense. And on top of that, you'd see duplicate topics. So we made some changes! You'll now see cohesive pluralization for relevant topics. (E.g. "Change Tire(s) instead of "Change Tire" and "Change Tires")

In order to get more context about what's happening in each customer interaction, we built a pretty snazzy document serialization machine. It chunks, queues, and passes documents into our inference pipeline, for every new document everyday.