How to Start Generating Insights

Insights are the data points Harmony extracts from your conversations. They live inside Projects, which group Insight Definitions (the questions you want to ask) with the Teams whose meetings should be analysed and optionally Project Criteria that narrow which conversations enter the project.

Open or create a project

In the sidebar, open Data Lake and pick an existing project. If you don't have one yet, see How to set up a project.

A project's tabs follow this layout:

• Reports — generated insight cards and charts.
• Configuration — the list of Insight Definitions that run for the project.
• Criteria — the Project Criteria that gate which conversations enter the project.
• Teams — the teams whose conversations are eligible.
• Revisions — reruns of past conversations (only visible when enabled for your workspace).
• Basics — name, language, Project Knowledge, teams.

Decide which conversations are in scope

By default a project runs every active definition on every conversation produced by its assigned teams. To narrow that down, use the Criteria tab — not the Configuration tab.

Open the Criteria tab (?tab=criteria) and add one or more Project Criteria. Each criterion is one of three types:

There are three types of Project Criteria you can use to control which conversations are included in a project: Attribute, Boolean, and Category.

The Attribute criterion operates by filtering conversations based on structured data fields that are already recorded about each conversation. For example, you can filter by the conversation’s length, tags it has been given, values in custom fields, or the date when it occurred. Attribute criteria are similar to database queries, so they do not require any AI processing. Instead, they serve as straightforward database filters that only include conversations matching your specified conditions.

The Boolean criterion takes a different approach by allowing you to pose a Yes/No question that is evaluated against the conversation transcript using Harmony’s AI models. For example, you might ask “Did the customer ask about pricing?” For each conversation, the AI reviews the transcript and returns an outcome of “Yes,” “No,” or “N/A” (if there is not enough information to determine an answer). This type of criterion is ideal when you want to include (or exclude) conversations based on whether a specific topic or behavior is explicitly present in the conversation.

The Category criterion enables the AI to classify each conversation into one bucket from a predetermined set of options. You define the list of categories that are relevant to your project — for instance, these could be “Discovery,” “Demo,” “Negotiation,” or “Closed.” For each conversation, Harmony’s AI considers the provided transcript and assigns it to one of these named buckets. The project configuration determines which category or categories a conversation must fall into in order to be included.

In summary, Attribute criteria are based on existing structured data, Boolean criteria rely on AI-driven answers to yes/no questions about transcripts, and Category criteria use AI to classify conversations by type or stage. You can use any combination of these to create highly specific or flexible conversation coverage in your project.

Criteria are owned by teams and can be reused across projects of the same team. A project can attach any number of criteria; a conversation must satisfy all attached requirements to enter the project. See Setting up a project for where Project Criteria live and how they relate to the rest of a project.

Add an Insight Definition

Open the Configuration tab (?tab=configure). This is where you add and manage the questions Harmony will answer for every in-scope conversation.

1. Click the + button at the top right of the tabs row.
2. Step 1 — Pick a type. Harmony shows the catalogue grouped by category (Most Popular, Custom, Built-in) with descriptions and supported report-type badges.
3. Step 2 — Configure. The form is dynamic per type. The right-hand column always includes a Project Knowledge panel so you can add or edit project knowledge from this flow without leaving it.
4. Most types accept an optional Additional Context string. This is appended to the prompt for this single definition alongside Project Knowledge — useful for instructions that apply only to one definition.
5. Click Create to save. The button stays disabled until the type-specific minimum is filled.

Insight types you can create

Harmony ships with 12 definition types in 3 categories.

Most Popular

• Summary — a short, structured summary of the conversation with key points and action items. No configuration required.
• Competitors Analysis — list of competitors mentioned, with positioning and advantages. You can pre-define Known Competitors to look for and Excluded Names that should never be flagged.
• Discount — detects whether a discount was mentioned and, when relevant, captures the rate.
• Buyer Engagement — fixed 0–10 score with built-in interpretation per band (8–10 strong engagement, 6–7 shows interest, 0–5 limited engagement). No configuration required.

Custom

• Open Question — free-text question, free-text answer per conversation.
• Yes/No Question — a question with a Yes/No answer. You can choose which answer counts as the positive (green) outcome in reports — useful when "No" is the desired answer.
• Checklist — a list of items the AI checks for. Each item gets a Yes / No / Unsure answer with a reason.
• Markdown — a free-text custom prompt that runs against the conversation. The reply is rendered as Markdown.
• Group — a list of named buckets (with descriptions). The AI assigns each conversation to one bucket.
• Scale Rating — a configurable numeric scale (default 1–5, configurable min/max) with optional labels per value.

Built-in

• Funnel — tracks which stages of a sales funnel were reached during the conversation.
• SPIN Improvement — suggests how to improve the rep's use of SPIN selling.
• SPIN Distribution — distribution of Situation / Problem / Implication / Need-Payoff questions in the conversation.

Both produce a numeric score, but they are different. Buyer Engagement is fixed 0–10 with built-in band guidance — the score does not adapt by team type. Scale Rating is fully configurable — you choose the range and the labels.

Manage existing definitions

The Configuration tab lists every definition in the project. From there you can:

To manage existing insight definitions within your project, you have several options, each accessible from the Configuration tab.

If you wish to change the name of an insight definition, simply click on the definition’s current name. This action makes the name field editable, allowing you to type in a new name directly and save your changes immediately.

To change the order in which definitions are displayed, look for the handle icon next to each row. You can click and drag a definition to a new position in the list. The order you set here determines how insight cards are presented on the Reports tab, giving you control over which insights appear most prominently.

If you find that a particular definition is temporarily not relevant, or you want to stop collecting new data for it without losing the existing results, you can pause it. This is done by toggling the active state for that definition. Once paused, the definition will not run on any new conversations, but all previous results remain available. At any time, you can resume the definition, making it active again for future conversations.

When you need to permanently remove a definition from your project, you can delete it. Deleting a definition is irreversible and removes both the definition itself and all of its associated insight results across conversations, so proceed with caution if you choose this action.

Finally, if you want to dive deeper into how a definition is performing across multiple conversations, simply click on its row. This opens the detailed report view, letting you analyze aggregated results and trends for that particular insight definition.

In summary, these management actions are designed to give you flexibility and full control over your project’s definitions, letting you adapt your reporting and insights as your needs evolve.

Re-activating a paused definition does not automatically re-run it on past conversations. To back-fill, see Reports and exports for the Revisions / rerun flow.

Reprocess a single conversation

If you change definitions, criteria, or project knowledge after a conversation has already been processed, the existing results don't update automatically. You can reprocess that single conversation:

1. Open the conversation.
2. Use the ⋯ action menu next to the title.
3. Pick the option that restarts from extraction or insights.

For project-wide reruns of past conversations, see the Revisions flow in Reports and exports.

View the results

Once configured, results appear in two places:

• Meeting level: open any meeting and switch to the Data Lake tab to see specific answers for that conversation, grouped by project.
• Reports tab: each project's Reports tab shows aggregated charts and metrics. Use the Display dropdown to switch between Graphs and List.

Best practices

• Start simple. A handful of high-quality definitions beats dozens of vague ones.
• One question per definition. "Did the buyer ask about pricing AND mention competitors?" is two definitions.
• Use the right type. A Yes/No question gives a clean graph; an Open Question is more flexible but harder to chart.
• Use Project Knowledge for context that applies to every definition (product names, methodology). Use Additional Context for instructions that apply to one definition only.

Think of custom insights as asking the AI a question. Start with "What do I want to know?" and phrase it the way you would for a teammate.