What we've shipped

Every analysis cycle now runs a change-analysis pass and writes the results back as queryable insights against your dataset, categories, subcategories, and individual topics. You no longer have to spot shifts yourself by scrubbing through charts. The interesting moments come to you.

What it detects

Four families of event, each with a significance score so only the ones that actually matter surface in the UI:

• Step changes in volume. Step-ups and step-downs in how often a topic or category is being mentioned, with the date of the change marked on the trend chart.
• Emerging themes. Topics that are new or newly active in this analysis window compared to the previous one.
• Concentration shifts. When mentions of a topic cluster more tightly, or spread out, across the dataset.
• Distribution shifts. When the mix changes across sources, sentiment, or sub-issues for a given category.

Where you see it

The events show up wherever you’re already looking. On trend charts they appear as dated annotations on the line, with the full period range in the tooltip. On insight detail pages, the right rail lists the recent changes for that topic. At the dataset level, the hero of an overview now leads with sentiment and the topic trends that have moved the most.

Runs on the first batch too

Previously change analysis only ran on incremental refreshes, which meant the first time you loaded a dataset you saw none of it. It now spawns automatically on initial-batch ingestion, so newly imported datasets land with the full set of detected events already in place.

The insight-embed surface we shipped a fortnight ago now covers trends too. Volume and sentiment lines for any insight, category, or subcategory can be linked as their own page on embeds.sunbeam.cx, ready to drop into Slack, a Notion doc, or a Linear ticket.

The card captions itself

Each trend embed carries a subject block that names what you’re looking at: the topic or category, the dashboard it belongs to, and the time period covered. You don’t have to write your own caption around the link. Paste the URL, and the card explains itself.

Step changes appear on the chart

The same temporal-step events that change analysis detects in the dashboard show up here too, as dated markers on the line. If volume tripled on a Tuesday, the marker says so. The tooltip on each bucket shows the full period range, not just a single timestamp.

Sensible buckets at every zoom level

Trend buckets now pick from a human-aligned cadence ladder (day, week, month, and so on) based on the period you’re showing, rather than landing on awkward intervals like “every 11 days”. Charts read cleanly whether you’re looking at the last fortnight or the last year.

Share straight from the dashboard

The sentiment and volume trends in the dashboard now have a share-this-graph link, which copies the public embed URL for whatever you’re looking at. Public dashboard links deep-link to your current view too, so filters, time range, and selected category travel with the URL.

Not a product change, but worth flagging. We’ve been pointing Sunbeam at public App Store review piles and writing up what falls out, both as a stress test of the analysis pipeline on real-world data and as concrete worked examples of the kind of signal it surfaces.

Four investigations went up on the blog today:

• Why aggregate review scores hide the signal that matters. Three case studies showing why a single sentiment number is a bad starting point and what segmentation actually buys you.
• Reading Discord’s reviews beyond the privacy headlines. 1,854 reviews. Filter out the privacy noise and there’s a distinct monetization signal underneath worth a separate look.
• Trading 212 reviews split sharply by use case. 926 reviews. Beginners are happy, active traders aren’t, and 13 of them want one specific UI revert.
• 99 MyFitnessPal customers want a classic-look toggle back. 99 of 1,981 reviews asking for the same thing. The strongest single signal in a post-redesign pile.

Expect more of these. If there’s a product whose review pile you’d like us to take a look at, send it over.

When you hit Share on an insight in a published dashboard, you now get a clean public link from embeds.sunbeam.cx. Open it anywhere, no login required, and you see that one insight on its own page: category, headline, summary, and a representative customer quote, with your dashboard’s name on the footer.

Built for Slack, Notion, and Linear

The link unfurls into a proper preview card. We render an Open Graph image at request time using the dashboard’s design tokens, so when someone pastes the URL into Slack or Linear or a Notion page, the team sees the insight title, summary, quote, and source straight away, without having to click through.

A real comment, not a synthetic one

The representative quote is an actual customer comment from the underlying dataset, with the source it came from and the date it was posted. It refreshes alongside the analysis, so the quote stays current as new reviews land.

Public surface, scoped to one insight

The embed is anonymous and read-only, scoped to a single insight on a dashboard you’ve already published. Private dashboards still use the authenticated link, and the embed never leaks anything beyond the one insight you linked to.

You can now send a dashboard to anyone with a public link. No login required. Copy the link from the Share button, and whoever opens it sees a view-only version of the dashboard, with your organisation’s name and logo at the top.

What the recipient sees

The full category breakdown, impact scores, comment excerpts, and filters, all read-only. They can’t edit taxonomies, see other dashboards, or change what’s in the view. Each link is scoped to a single dashboard.

Sessions stay separate

A public-link session is fully isolated from any authenticated session on the same browser. If a colleague is already logged in to Sunbeam in another tab, opening your share link won’t inherit their permissions, and the share session won’t leak back into the app.

Every category and subcategory page in the dashboard now leads with a short summary of what customers are saying, and there’s a dataset-wide summary on the overview page. A few sentences on the main themes, any notable sub-issues, and anything that’s recently shifted. No need to scroll through hundreds of comments to get the gist.

How they’re generated

Summaries run as part of the analysis pipeline, against the actual comments for that slice of the data, not a generic template. Prompts are tuned to stay close to what’s in the reviews, with hard length limits, so the summary doesn’t drift into padding or invent trends that aren’t there.

Summaries refresh automatically whenever new reviews land.

You can now send survey email invitations from a sender address on your own domain, not just a shared Sunbeam address. Add your domain in settings, add the DNS records we give you, and Sunbeam will send invitations from surveys@yourcompany.com (or whatever you choose) with your branding end-to-end.

Verification catches the common mistakes

When you verify a domain, we check your DNS records and tell you what’s actually out there in the wild, not just pass/fail. If you’ve pasted a record into the wrong field or typed @ where your DNS provider wants the bare hostname, we’ll spot it and say so.

What else lands with it

Every survey email now carries your organisation’s theme colour and logo, a proper preview text, a compliant unsubscribe flow, and renders correctly on mobile and in dark mode. The “Powered by Sunbeam” footer is styled to sit alongside your brand rather than fight it.