Call summaries & quality

Know what was said on every call, and whether it was said well.

TL;DR

• What it is — AI summaries of calls and of each customer, plus quality checks that score a call against your scorecard and the call script the agent was meant to follow.
• Who it's for — everyone reads summaries; admins build scorecards and score calls; agents see the scores on their own calls.
• Top outcome — a supervisor opens a call and sees, in seconds, that the agent skipped the cancellation policy, recorded the wrong budget in the dialer, and scored 74 out of 100 — with the exact words quoted.

At a glance

How to find it

Breadcrumb: Settings → AI & Automation → Call summaries & quality.

Direct URL: https://lodgestory.com/crm/settings/call-insights

The summaries and scores themselves appear where you already work: a call's detail sheet in Calls, the contact, the chat side panel in Home, and the lead.

What it does

Call summaries

Open a call with a transcript and click Summarise call. A few seconds later the AI summary card shows a one-line headline, two to four sentences, what the customer wanted, the key points, the next steps (who acts, and when), how the call ended, the customer's mood and the languages spoken. It is written in English whatever language the call was in, and saved, so the next person who opens the call sees it at once.

Customer summaries

One summary per customer, from everything said across their calls and chats (WhatsApp, Instagram, email, web chat): who they are, where things stand, what is still open, their preferences and lasting facts. Update folds in only what is new since the last update.

Quality checks

Admins click Score call on a call's detail sheet. AI judges each item on your scorecard and each step of the call script from the transcript, quoting the words it relied on, and the score out of 100 is worked out from your weights. The Quality check card shows:

• the score, and when a must-pass item failed, why the call scores 0 (and what it would have scored);
• the Script section: how much of the script was followed (Villa pitch · 63% followed · 1 step missed · 1 required), each step as Followed, Partly, Missed or Didn't come up, steps the agent clicked through in the Agent Dialer but never said, and answers that don't match the call — where the answer recorded in the dialer differs from what the customer actually said;
• every scorecard item with its result, points and, on click, the reason and quotes (a quote that isn't found word for word in the transcript says so);
• up to three coaching tips.

The scorecard is saved with the score, so editing or deleting a scorecard never changes a score already given.

Core concepts

Quick start — your first scorecard

1. Open Settings → Call summaries & quality and click New scorecard.
2. Click Start from template for 19 items across Opening, Communication, Problem solving, Compliance and Closing (100 points), or Add item to write your own.
3. Under Which calls, tick Outbound (or leave everything empty to match every call) and set the minimum talk time.
4. Turn on Call script check. Leave The call's own script selected, choose How closely steps must match, pick a script and tick Required next to the steps that must never be skipped.
5. Click Create scorecard, then open any transcribed call in Calls and click Score call.

AI model

Settings → Call summaries & quality → AI model picks the model that writes this workspace's call summaries and customer summaries and runs its quality checks. The default is Claude Sonnet 5.5, called directly at Anthropic. An admin can switch to one of these models through OpenRouter:

• Every listed model returns the same structured result; Lodgestory only uses OpenRouter providers that enforce it.
• An OpenRouter model needs an OpenRouter key: Lodgestory's, or your own added by the Account Owner under Settings → API Token & AI Keys. Without one, the OpenRouter models can't be picked, and if the key is removed later, summaries and checks say so instead of failing silently.
• The change applies to the next summary or check. Summaries and scores already given keep the model that wrote them.

Run automatically

Settings → Call summaries & quality → Run automatically has three switches, each off until an admin turns it on:

• Automatic summaries and scores are the same as clicked ones: they appear on the call, contact, chat and lead, go to webhook subscribers (with trigger auto), and are recorded in AI usage. A call sheet or chat panel that is open when one lands refreshes by itself.
• At most 2,000 automatic summaries and checks a day per workspace. Suspended workspaces run none.
• Clicking still works whatever the switches say.

How a check picks its scorecard and script

Scorecard. The first active scorecard, by order, whose rules all match the call — direction, calling account, department, lead track (Agent Dialer calls and calls to leads on that track) and minimum talk time. On a call's sheet an admin can also Choose scorecard; a chosen scorecard skips the talk-time rule. When no scorecard applies, an Agent Dialer call is still checked against its own script (score = how closely it was followed); any other call asks you to create a scorecard.

Script. With The call's own script: the script the agent ran in the Agent Dialer — with the steps they clicked through and the answers they recorded — or else the script of the lead's track. With Always one script: the script you pick, for calls that have no dialer script, such as inbound support calls. When the script was edited after the call, the check says so; it is checked against the current version.

Natural speech. Agents don't have to read the script out. At every How closely setting, a Hindi or Hinglish version of an English script line, steps in another order, a step spread over the call, filler words and transcription slips don't count against the agent, and an answer the customer gave in other words or another language is not reported as a mismatch with what was recorded in the dialer.

What isn't scored. Calls that weren't connected, conversations under 10 seconds or shorter than the scorecard's minimum, calls with too little said, transcripts with only one side, and connect calls (no agent). The card says why.

Roles and permissions

Webhooks and usage

Every summary and score is also sent to your webhook subscribers — call.summary_available, customer.summary_updated and call.quality_scored — with the results and short quotes, never the transcript or message text. Each AI request is recorded in your AI usage under call summary, customer summary or call score.