Conversation types
Conversation types is the part of ArcGlass Lens that sorts your calls by kind, such as discovery, demo, sales call or customer call, and reads every call of a kind against that kind's playbook: the steps a good conversation of that kind covers. For each step, ArcGlass records whether the call covered it, with the quote that shows it. Lens then shows which steps your team holds, which it skips, which calls covered the least, and what the calls that moved forward did differently. It measures coverage, not a score.
Key facts
- What it is: playbook coverage for every recorded call, grouped by kind of conversation.
- Built from: the meetings and calls ArcGlass analyzes, whether captured with ArcGlass Listen or imported from a connected source such as Zoom, Google Meet, Microsoft Teams, Gong, Fireflies, Fathom or Granola. Emails and chat messages don't get a type or a playbook.
- Types come from your calls. ArcGlass labels each call with its kind, using your pipeline's list of meeting types when you have one. Your team can change any call's type.
- Playbooks start themselves. For common sales and customer conversations, ArcGlass sets up a playbook from a built-in starter when Lens is opened or overnight, including for types your pipelines list before their first call.
- Evidence, not opinion: a step the AI reads counts as covered only when its quote is found in the transcript. The rule-based steps are decided from the call's agreed next step and its turns.
- Coverage, not a grade: there's no score per call, per rep or per team.
- Where: in the console's left navigation, under What you hear → Lens, then Conversation types.
Questions it answers
- What does a good discovery call, demo or renewal cover, and how often do our calls cover it?
- Which step does our team skip most?
- Which calls covered the least, so we know which ones to open?
- After a customer asks a question, do we ask back or go back to presenting?
- What did the calls that moved forward do differently from the ones that stalled?
How it works
- Every call gets a type. When ArcGlass analyzes a meeting or recorded call, it labels the kind of conversation. If your pipeline lists its meeting types, the label comes from that list. Calls it can't place stay Unconfirmed. Analyzing a call again fills in a missing type but never replaces an existing one.
- Each type can have one playbook: an ordered list of steps, each written as what a good call covers, such as "What happens next, who owns it, and when". A step can be required, and a step can apply only in some calls, for example only when the customer raises a topic.
- Each call is checked against its type's playbook. One AI read per call checks every step that needs reading. For each step the result is covered, not seen, or (for a step that applies only in some calls) did not apply.
- Covered needs a quote. The quote the AI gives must appear in the transcript, ignoring case and punctuation. A near match, with at least 80% of the quote's words (three or more) found in one turn, also counts and is labeled (near match). Anything else is recorded as not seen. A step the AI doesn't answer stays unmeasured rather than counting as missed.
- Some steps are checked by rules, not AI. "Next step has a date" is read from the call's agreed next step. Three conversation-flow steps are counted from the turns themselves: the rep asks back after a customer's question, no stretch of your team talking runs past 400 words, and no run of more than 15 turns from your side goes without a customer reply.
- Results add up per type. A step's coverage is the share of calls where it was covered, out of the calls where it applied: covered ÷ (covered + not seen).
Playbooks
Starters
ArcGlass has built-in starter playbooks for common conversations. Every workspace gets the Discovery and Demo playbooks the first time Lens is opened.
| Starter | What it's for | Steps (required) |
|---|---|---|
| Discovery | Early call to understand what the customer wants to change and qualify fit. | 10 (2) |
| Demo | Product walkthrough built around the priorities the customer named. | 13 (0) |
| Sales call | A conversation with a prospect about a possible deal, from a first call to negotiation: understand the problem, qualify how they'll decide, show value in their words, and leave with a booked next step. | 17 (2) |
| Customer call | A conversation with an existing customer: confirm they're getting value, catch risk early, find what's next for the account, and leave with owners on both sides. | 15 (2) |
| Investor meeting | A pitch, follow-up, or update with a current or prospective investor: a clear story, honest proof, a specific ask, and a known path to a decision. | 18 (1) |
| Technical evaluation | Technical deep dive to confirm the solution fits the customer's environment. | 6 (5) |
| Business review | Periodic check-in on progress, value delivered, and what comes next. | 7 (6) |
| Renewal | Confirm renewal terms and surface any risk before the contract lapses. | 6 (5) |
The Discovery starter, for example, includes "What it costs in money, time, risk, or effort", "What must be proven before the customer proceeds", "What is the budget or approval process for a purchase like this?", "What is the timeline to have something in place?" and "What happens next, who owns it, and when".
Which types get a playbook automatically
- Started for you: types whose names match demos or walkthroughs, investor or fundraising meetings, discovery, sales, prospect calls, proposals, pricing or negotiation, and customer, client or account reviews. These playbooks are marked as set up by ArcGlass, except Discovery and Demo, which every workspace starts with.
- Suggested, not started: renewals, business reviews and QBRs, technical evaluations, pre-sales, POCs and pilots, and intro or kickoff calls. Your team chooses whether to add one.
- Never: internal meetings such as stand-ups, one-on-ones, prep, debriefs, training, interviews, planning, pipeline and forecast reviews, and hiring.
Analyzing a call never creates a playbook. ArcGlass starts them when Lens is opened, and overnight after that.
Steps every playbook includes
Every type with its own playbook is also read against seven steps that apply to any customer conversation:
- Next step has a date
- Objection closed in the room (only when the customer raises an objection)
- Prior commitment confirmed (only when an earlier call left an open commitment)
- Committed to confirm by a stated time rather than guessing (only when the rep didn't know the answer)
- Tested for whether the gap is decision-blocking rather than absorbing it (only when a genuine product gap is named)
- Unsupported cloud or capability scoped as blocker, later-phase, or non-critical rather than deflected (only when the customer asks about one)
- Moved to backup material with a named follow-up owner rather than live troubleshooting (only when a demonstration failed)
Calls of a type without its own playbook, and calls with no type yet, are read for Next step has a date only.
Starting and changing a playbook
Each playbook is versioned. When a step is added or reworded, the new version starts measuring that step again from scratch, so old results never count for new wording.
- Start a playbook: on a call's Playbook panel, Create a playbook starts one from a matching starter and checks that call against it. Adding a type from the Unconfirmed card starts its playbook too, when a starter fits.
- Add a suggested step: on a type's What we're learning page, Add as a step adds a step that calls which moved forward had more often.
- The editor: "What a great <type> covers" lists the steps with their phase (Opening, Middle, Relevance or Close, for display only) and an optional prep note. Preview shows how new or reworded steps would have scored on the 10 most recent calls of that type, such as "Would have been covered in 4 of 10". Saving edits from the editor isn't available yet.
The Conversation types page
Open Lens, then Conversation types. The time window in the Lens bar offers 7, 30, 60 or 90 days, or All time (the default). It applies to the map and to a type's steps and calls; the type cards, the flow and the learnings use all of your calls. Calls without a date appear only under All time.
- The stage map shows your companies grouped into regions by conversation type, each region sized by its number of calls. Hover a company to see how many conversations it has, or drag it onto another region to move its calls of that type there. The dialog confirms "Their other conversations are not affected." Moved calls are checked again in the background.
- Type chips and Search accounts… narrow the map.
- Type cards list your conversation types, those with a playbook first. Each card names the least-covered step, for example "Pricing is the least covered: 2 of 9 where it applied", and says how many steps aren't measured yet. A type without a playbook says so; its Create a playbook link opens the editor, which can preview but not save yet. Open a type from its card.
- Unconfirmed counts calls that have no type yet and suggests what to do with them, such as "Move 4 to Discovery" or "Add Onboarding as a type".
One type's page
The page opens with the type's own map, where clicking a company narrows the summary, the steps and the grid of calls to that company. A short summary of how these calls run follows: the habit that holds (covered in 70% or more, but not all, of the calls where it applied), the thinnest step (under 50%), and the steps always met. Each needs at least three calls where the step applied. Below it, four cards open four views:
| Card | Opens | What you see |
|---|---|---|
| How we fared in each step | View the steps | Every step in order, with its coverage, "x of y applicable", the quotes behind it, and whether it's required. |
| How we fared overall | View each conversation | A grid of calls by steps, the calls that covered the least first, so you know which ones to open. |
| How the conversation flows | View the flow | How the turns went in these calls, counted from the transcript. |
| What we're learning | View suggestions | What the calls that moved forward did differently, and what to watch for. |
The flow
Five measures, each counted from who spoke when, with no AI involved:
- Customer asked, rep answered and resumed presenting
- Customer asked, rep answered and asked back
- Uninterrupted vendor stretch beyond the band
- Longest run of vendor turns with no customer reply
- Customer spoke as an owner
Flow is counted for types that have a playbook. Calls where ArcGlass can't tell your side from the customer's are left out, with the reason shown.
What we're learning
ArcGlass sorts calls into those that moved forward, stalled or went backwards, and asks the AI what the ones that moved forward did differently. Each finding must hold up against the calls it cites. The comparison needs at least 15 calls in a pipeline, with at least 5 that moved forward and 5 that stalled or went backwards. The findings come from your whole workspace, not just this type, and the page says so.
- Worth making a step: seen more often in conversations that moved forward. Add as a step writes it into the playbook as a new version.
- Worth avoiding: seen more often in conversations that stalled. Advisory only, because a transcript can't show that something was avoided.
- Worth watching for: signs that show up early when a conversation is going sideways.
- What we say we do: claims your team makes repeatedly across these conversations.
- What customers struggle with: recurring problems customers raise that map to what you sell.
Results on each meeting
Every analyzed call whose type has a playbook shows a Playbook panel with each step's result:
| Result | Meaning |
|---|---|
| Covered | The call covered the step. For a step the AI reads, its quote is in the transcript. |
| Answer incomplete | The step came up, but the answer didn't fully settle it. |
| Not seen | Nothing in the transcript shows the step. It may have happened elsewhere, such as over email. Shown in red when the step is required. |
| Did not apply | The step applies only in some calls, and this wasn't one of them. |
| Not checked yet | The step hasn't been read for this call yet. |
| Checking is paused | Checking is switched off for now. |
| No transcript to check | The call has no transcript. |
Lens uses shorter names for the same results: met, not seen, not applicable and not checked yet.
From the panel you can:
- Change the call's type with the type chip, or type a new one under Another type. The chip shows when the call has no type yet or its type has a playbook. The call is checked against its new playbook straight away.
- Check this call now when some steps haven't been read yet.
- Give a type a playbook when it doesn't have one: when a starter fits, Create a playbook starts one from it and checks this call against it.
Outside Lens
- Follow-ups: when Tasks is on, required steps a call didn't cover become lines on one "Ask on the next call" task per account.
- Account page: the Qualification card shows each AI-read step from the account's earliest scored call: answered, partial or not covered. Steps that didn't apply are left out.
- Deals: the same results feed the ideal-customer-profile and executive-engagement panels.
- Weekly brief: the brief to organization owners names the busiest type with at least 10 calls in the last 30 days and no playbook.
Admin center → Conversation types
A single table of every conversation type, including the types your pipelines list:
- Calls of that type, all time.
- Playbook: "None", or its steps, required steps and version, for example "17 steps (2 required) v3".
- Last edited, which shows "ArcGlass" for playbooks ArcGlass started.
- Pipelines: whether the type is in your pipelines' meeting-type lists, with Add to pipelines so the analysis can choose it for future calls.
- An action to open the playbook in the Lens editor: Edit playbook, Start from <starter> or Create a playbook. The editor previews but can't save yet, so start a new playbook from a call's Playbook panel.
The tab is visible to Members, Managers and Organization Owners.
How it stays current
- As each call is analyzed: if its type has a playbook and the call has a transcript, ArcGlass checks it straight away. A changed type is checked again at once.
- Every night: ArcGlass starts playbooks for new types that qualify, counts the conversation flow, checks "Next step has a date" on every call, checks up to 200 calls still waiting, and rebuilds the type cards and summaries.
- After a change: moving calls, adding a type or starting a playbook queues a background check of calls still waiting.
Who can do what
| Action | Who |
|---|---|
| See Lens, the type pages and every meeting's Playbook panel | Anyone who can read conversations |
| Change a call's type, move a company's calls, add a follow-up | Member, Manager, Organization Owner |
| Start or create a playbook, add a type, add a suggested step, preview, check a call now | Member, Manager, Organization Owner |
| Add a type to your pipelines' meeting-type lists | Manager, Organization Owner |
Conversation types is a workspace-wide view: its titles, quotes and companies cover every analyzed call in the workspace.
What it doesn't do
- It reads calls only. Emails and chat messages don't get types or playbooks.
- It doesn't grade people. There's no score per call, per rep or per team, and nothing groups results by person.
- Types can't be renamed, merged or deleted yet, and a playbook can't be deleted.
- The editor previews but doesn't save yet. Required flags, conditions and the seven every-conversation steps can't be changed in the app.
- "What we're learning" is workspace-wide, not specific to one type.
- A follow-up line clears only when the same call is checked again, not when a later call covers the step.
Frequently asked questions
What is a conversation type in ArcGlass?+
How does ArcGlass decide a call's type, and can I change it?+
What is a playbook?+
Which playbooks does ArcGlass set up automatically?+
How does ArcGlass decide a step was covered?+
Does Conversation types score my reps?+
What's the difference between "not seen" and "did not apply"?+
What does "Answer incomplete" mean?+
What happens to calls marked Unconfirmed?+
Do emails get conversation types?+
How often is Conversation types updated?+
Can I edit a playbook?+
Glossary
- Conversation type: the kind of call, such as discovery or demo.
- Playbook: the ordered steps a good call of one type covers.
- Step: one thing a good call covers. It can be required, or apply only in some calls.
- Starter: a built-in playbook ArcGlass uses to set up a new type's playbook.
- Covered (met): the call covered the step: with a quote found in the transcript for a step the AI reads, or by rule for the next-step and conversation-flow steps.
- Not seen: nothing in the transcript shows a step that applied.
- Did not apply (not applicable): the step's situation didn't happen in this call.
- Answer incomplete: the step came up, but the answer left part of it open.
- Coverage: covered ÷ (covered + not seen), across the calls of a type.
- Unconfirmed: a call with no type yet.
- Lens: the ArcGlass page that holds Conversation types and Landscape.
Related
- Landscape: the other half of Lens, mapping every company your customers mention.
- Product Signals: what customers ask for and what lands, from the same calls.
- Tasks: where follow-ups for missed required steps go.
- Admin center: workspace settings, including the Conversation types tab.