buzzabout docsAnalyse pain points
Turn a collected dataset into a ranked map of what your buyers actually complain about, using the built-in Analyze pain points skill.
If you run an e-bike DTC brand, you already half-know the complaints — range anxiety, brake feel, delivery delays, patchy after-sales support — but you can't tell which one is actually costing you sales. The Analyze pain points skill reads a dataset you've already collected and returns a ranked map: a severity hero chart, pain-point clusters weighted by reach + intensity + recurrence (not raw mention count), and three handles for what to do next — a messaging fix, a product fix, and a content angle. You point the assistant at your mentions, apply the skill, and read the result.
Prerequisites
- A dataset with a completed collection run, so there are posts and comments to analyse. See Set up your project and Create a mention collection.
- Enough volume for the clusters to mean something. A few dozen mentions will surface anecdotes; aim for ~100+ mentions before you trust the severity ranking.
Walkthrough
You can run this two ways: apply the built-in skill (so the assistant follows the full pain-point pipeline every time), or just ask the assistant in plain language. Both end at the same Asset.
Open a chat
Open a chat against the dataset that holds your e-bike mentions — either from /chat or from your project's chat. The assistant works over the mentions that dataset's runs have already collected, so make sure the collection has finished first.
Open skills
In the chat composer, click the Open skills book icon (or type / at the start of an empty message) to pop the skills picker. It shows your Favourite skills with a Browse all skills footer.
Browse all skills
Click Browse all skills to open the full library. Analyze pain points lives under both Content & positioning and Audience research, and it's tagged as a suggested skill, so it often surfaces near the top. You can also type its name into Search skills….
Use this skill
Open the Analyze pain points row to read its instructions, then click Use this skill to attach it to the current chat draft. The modal closes and the skill appears as a chip in the composer. Attach your e-bike dataset as context if it isn't already, type a short prompt like Analyse the pain points in this dataset, and send.
If you'd rather skip the library, you can just ask in plain language — What pain points do e-bike buyers complain about most? — and the assistant will run the same pipeline. There is no separate "Pain Points" button; the skill is a saved instruction block, not a one-click report.
Let the assistant run the pattern pipeline
Once you send, the assistant runs the pattern-matching pipeline to discover and cluster the pain-point categories, weighing each cluster by reach, intensity, and recurrence rather than how often it's mentioned. This runs asynchronously and can take a few minutes. If the dataset already has pattern assignments from an earlier run, the assistant reuses them instead of re-detecting, so the answer comes back faster.
Open in side panel
The assistant wraps its deliverable in an Asset. Inline you'll see a clickable reference card with a short summary; click it to Open in side panel and read the full breakdown. You'll get a severity hero chart, a cluster detail table (each pain point with its weight and the mentions behind it), and three handles — a messaging angle, a product angle, and a content angle — telling you what to do with the top clusters.
(Optional) Narrow the mentions by Sentiment
To read the raw evidence behind a cluster, open the mentions view and use the filter builder. Click Add filter, set the Where field to Sentiment, leave the operator as is, and pick Negative. Combine rows with and / or, and use Reset to clear. This narrows which mentions you're reading — the pain points themselves are still derived by the assistant, not by the filter.
Refining the underlying query. If your dataset is pulling in the wrong conversations, fix it upstream in the collection's keyword query rather than in the filter. Operators are AND, OR, NOT plus parentheses for grouping and "double quotes" for exact phrases. The default join between two keywords is OR (broadens reach) — add AND to require co-occurrence, NOT to exclude. Operators are case-insensitive (shown uppercase). Examples: "electric bikes" NOT kids; nike AND (running OR marathon); rivian polestar lucid (implicit OR). Sanity-check complex queries against the live mention-count estimate.
Gotchas
- There is no "Pain Points" button or asset type. Pain points are auto-extracted by the Analyze pain points skill (or by asking in plain language) via the pattern pipeline. The result is shown in the generic Asset wrapper —
Pain Pointsis a label the assistant uses internally, not a dedicated control. - There is no "emotion" mention filter. The filter builder supports only Keyword, Sentiment (Positive / Negative / Neutral), Views, Likes, Comments, Engagement rate, and custom parameters. A finer-grained emotion breakdown exists per mention as a display field only — it is never filterable.
- The pipeline is asynchronous and may charge credits. Pattern/post-processing runs cost roughly 0.5 credit per mention and poll for several minutes. By default the skill reuses any existing pattern on the dataset to avoid re-running; it only re-detects when you explicitly ask.
- Clusters are weighted, not counted. Ranking blends reach, intensity, and recurrence, so a loud complaint from a few high-reach posts can outrank a quiet one mentioned more often. Read the weight column, not just the row order.
- You need collected mentions first. The pipeline needs a dataset that already has posts and comments — run a collection before applying the skill.
- The chat needs a user session. Over MCP,
buzzabout__askrequires OAuth/JWT auth; an API key alone returns403and cannot reach the assistant.
Now analyse it
The same dataset answers more than just "what's wrong." Apply another built-in skill to extract a different slice:
Run audience analysis
Build an audience dataset from a collected dataset, then read demographics, interests, brand affinities, creator tier, and an inferred Personality (OCEAN) fingerprint for the people behind the posts.
Analyse feature requests
Apply the Feature requests skill to a collected dataset to turn social chatter into a severity-ranked, triaged product backlog.