Signal Classification

What it does

Signal Classification is the AI layer that automatically processes every incoming signal — Slack messages, Jira tickets, Gong call excerpts, GitHub issues — and categorizes it along three dimensions:

Type — what kind of signal it is: bug report, feature request, complaint, praise, question, or general feedback

Sentiment — the emotional tone: positive, neutral, negative, or mixed

Feature area — which part of your product it relates to, mapped to your product taxonomy

This classification happens automatically in the background as data syncs from your integrations. You never have to manually tag or sort signals — Specky does it for you.

The classification model is trained on product management data and understands the difference between "this is broken" (bug) and "it would be great if" (feature request), even when the language is informal or ambiguous. Each classification includes a confidence score so you can see how certain the model is.

Beyond basic classification, Signal Classification also:

Clusters related signals — groups signals about the same topic even if they use different words

Detects emerging themes — surfaces new patterns that weren't present in previous periods

Scores signal importance — weights signals by the seniority of the source, the recency, and the volume of similar signals

When to use it

During discovery — filter the Discovery Hub to show only feature requests in a specific area to understand what customers want

For prioritization — use signal volume and sentiment by feature area to inform what to work on next

For monitoring — set up alerts for high-severity bug signals in critical feature areas

For trend analysis — compare signal volume and sentiment across time periods to see if things are improving or getting worse

For stakeholder reporting — show the volume and sentiment of customer feedback by feature area in a quarterly review

After a launch — monitor the sentiment of signals in the launched feature area to catch early negative reactions

How to use it

Signal classification runs automatically — there's nothing to set up. To use the classifications:

In the Discovery Hub:

Open Discovery Hub from the sidebar

Use the Filter panel on the left to filter by:

- Type: Bug / Feature Request / Complaint / Praise / Question - Sentiment: Positive / Neutral / Negative - Feature area: Select from your product taxonomy - Confidence: Filter out low-confidence classifications

Click any signal to see its full classification details, confidence score, and related signals

For trend analysis:

Open Discovery Hub and click Trends in the top nav

Select a feature area and time range

See a chart of signal volume and sentiment over time

For alerts:

Go to Settings → Alerts

Click New Alert

Set conditions: e.g. "Alert me when there are more than 5 bug signals in the checkout area in a single day"

Choose notification method: email, Slack, or in-app

Correcting a classification: If a signal is misclassified, click the classification badge and select the correct type. Your correction is used to improve the model for your workspace.

Example

Your team ships a new checkout flow on Monday. By Wednesday, you notice an alert: 12 bug signals in the "checkout" area in 48 hours, with predominantly negative sentiment — compared to a baseline of 2 per week.

You open the Discovery Hub, filter for bug signals in checkout with negative sentiment, and see a cluster of 9 signals all describing the same issue: the promo code field disappears on mobile after entering a code.

You create a Jira ticket directly from the signal cluster, link it to the checkout PRD, and add it to the Decision Log as a post-launch issue. The engineering team has a fix deployed by Thursday — before the issue reaches your support queue.