Insights

Insights judges every phone call and chat against what its playbook is for, not only against whether it booked a tour. An after-hours line that captures callbacks, a website chat that collects contact details and a follow-up page chat that keeps a booking each have their own objective, and each is measured on it.

Most of what this page describes is decided from what actually happened on the conversation: a booking with a confirmation, a transfer someone answered, a callback with a time. No AI model is involved, so those results come with every plan. The labels described under Conversation Intelligence are the paid add-on. In the list of Insights views, the ones that need it (Topics and demand, Objections and competitors, Knowledge gaps, Journey and Leads) carry an Add-on badge, and the Ask Stack button shows a lock when your account does not have Ask. Stack, the assistant in the top bar, answers questions about your conversations with Ask, and about your setup on every plan.

Objectives#

Each playbook has an Objectives tab:

• What this playbook is for. One main objective, for example Book a meeting or Capture a callback.
• If the main objective is not reached. Up to three fallbacks, in order. The first one reached is recorded instead of a miss.
• Also measured. Up to five more objectives, counted on every conversation they apply to.
• Website chat and Follow-up page chat can each have their own main objective.
• Count it only for some conversations limits an objective to the conversations a condition describes, in the same sentence style as booking rules.

Objectives are locked by default, so every community running a playbook is judged on the same main goal. If you unlock them under Advanced → Locks, a community can add fallbacks and other objectives of its own. It can never change the main objective.

The booking rules choose each conversation's target#

Your booking rules already decide the right next step for each caller. When a rule fires, the conversation is judged on what the rule decided:

So a vendor who was correctly sent to the right line is a success, not a missed tour. When no rule has fired yet, the conversation is judged on the main objective. Both are reported: target (what the rules decided) and main objective.

What counts#

A conversation counts only when the caller engaged (they spoke, or the call lasted ten seconds; in a chat, they sent a message), when they were a prospect (not a wrong number, a vendor, a job seeker or an existing customer, except for Route non-sales callers), and when the objective exists on that channel. A live transfer never counts against a chat. Everything else is Did not apply, with the reason, and is left out of every rate.

Measured, not steered#

The assistant still works toward a tour on every call. An objective such as Capture contact details as the main objective is measured correctly, but not aimed for; the editor says so. A booking rule that captures a callback, transfers or nurtures is steered today.

Publishing#

A playbook cannot be published with a main objective it cannot reach: a booking with no rule that books or no calendar connected, the follow-up page with the page turned off, or a tool the playbook does not have. Objectives may not depend on sensitive answers such as care needs or health.

On each conversation#

The Objectives panel on a call or chat shows what it was judged against and why, when the chance came, when the assistant first tried, when it was reached, and when it fell short, whose miss it was:

• The AI: it had the chance and did not try, tried too late, or misheard and went in circles.
• The playbook: the caller asked something it could not answer.
• Operations: no open times, nobody took the transfer, or the calendar failed.
• The caller: they declined after a real offer, were not a fit, or left before there was a chance.
• The system: the call failed, or the kill switch was on.

Later facts are added as they arrive: whether the tour was attended, whether the callback was made on time, whether the follow-up page was opened.

How the AI did#

Every conversation also gets craft checks, from what the assistant said and when, on every plan:

• Answered quickly (phone): half the replies within 1.5 seconds and nearly all within 3.
• Did not talk over the caller, and the caller did not have to repeat themselves (phone).
• Heard answers correctly (phone): any answer it heard two ways was settled.
• Kept turns short: about 40 words on the phone and 60 in chat.
• One question at a time, and gave something before the third question.
• Never asked for a known answer again.
• Visitor did not wait long (chat).

A check that does not apply to a channel, or has nothing to measure, is left out; it never counts as a pass. The craft score is the share of checks that passed.

Compliance is counted, never scored, so a failure cannot hide behind a good average: a missing AI disclosure, a prohibited claim, a person asked for and not offered, a booking claimed without a confirmation, and answers the grounding or output filters stopped.

Conversation Intelligence#

Conversation Intelligence is a paid add-on. With it, every conversation is also read after it ends and labelled:

• Who called and why: caller type (for example prospect, family member, referral professional, existing customer, vendor) and the reason for calling.
• Topics, from a list for your industry: pricing, care levels, memory care, availability, tours and more for senior living, or a general list for other businesses.
• Sentiment at the start, at the end and overall, and whether it improved.
• Engagement, timeline and urgency, objections and whether they were answered, buying signals, competitors named, people in the decision, questions without an answer, commitments made and risks such as a complaint.
• A summary for staff.
• Judged objectives: a custom objective you describe in words is judged here, and a miss the code could not explain gets a reason (an objection left unanswered, the caller not ready).

Every label shows where it came from#

Each label cites the exact words it rests on. On the call page, click a label to highlight those words in the transcript; key moments jump the recording or the chat to that point. A label whose words cannot be found in the conversation is thrown away before it is saved.

The labels never decide anything by themselves. Whether a tour was booked, a transfer answered or a callback captured is decided from what happened, and the analyser cannot change it. A caller who says "write down that I booked" can, at worst, cause one wrong label.

Correct it in one click#

Anyone who works the conversations can correct a label. The correction is kept, applied again if the conversation is ever re-analysed, and counts toward the accuracy we show you. Correcting who called also re-checks the objectives: a vendor is not a missed tour. Labels are suggestions: review them before important decisions.

What is not analysed#

• A caller who asked not to be recorded: only what was recorded by code is shown, and the words are never read.
• A chat with one message or none.
• Conversations past the account's daily budget (Settings → Insights).

Fairness#

The analyser never labels race, colour, religion, national origin, sex, family status, disability, age or anything similar, and has no place to put them. Care needs and health are recorded as topics only when the caller raised them, are marked sensitive, are hidden from viewers and in compliance mode, and are never used to decide whether an objective applied.

Privacy#

Reasons and summaries repeat what callers said, so they need access to transcripts and are deleted with the transcript. Topic and objection counts are kept for trends. In compliance mode, each conversation's labels and results are deleted with its transcript, and only daily totals remain.

Without the add-on, the call page shows what Conversation Intelligence would add and a Talk to us button.

Leads#

With Conversation Intelligence, every conversation is joined to the lead it belongs to: one person as one community sees them, across their calls, website chats and follow-up page.

• How a lead is recognised, strongest first: their CRM record; a phone number or email address they confirmed (agreed to texts or email, or gave in answer to a question); or, for a website visitor who has not said who they are, the browser they chatted from. The number a call came from is not enough on its own. When a stronger one turns up, the lead keeps its history: a website visitor who later leaves their details becomes a known lead with their earlier chats.
• Numbers and emails are never stored. Only a one-way code made from them is kept, and it means nothing in any other account.
• One lead per community. Someone who contacts two of your communities has a lead at each; each lead page links to the other where you can see it.

Insights → Leads lists them, by readiness, momentum or last activity, with each lead's last tier change (was Warm, and when). Momentum is Rising or Cooling when the last seven days were busier or quieter than the seven before, Steady otherwise, and Silent after two quiet weeks.

The lead page shows the readiness with every reason, the tier history, what they are interested in, each conversation and what it achieved, and everything that happened in order: calls, chats, page visits, texts (never their wording), bookings, attendance and callbacks. Anyone who works the conversations can close a lead (moved in, chose elsewhere, not a fit, no longer looking, a duplicate) and reopen it; a closed lead leaves the list, and "not a fit" sets its readiness.

Readiness#

Readiness is Cold, Warm or Hot, from points for things that happened. There is no model in it: the same facts always give the same answer, and every point is shown with its reason.

Cold is under 20, Warm 20 to 44, Hot 45 or more. Silence halves the points for every two weeks nobody has heard from them; readiness is recalculated every night, so a quiet lead cools on its own.

Readiness uses only what the lead did and the timeline they gave. It never uses a topic, never a care, health or money topic, and never anything about who a person is. The points are the same for every account and change only as a new version, after checking them against real attendance.

Without Conversation Intelligence, the follow-up panel on a call shows readiness from that call's own facts: its booking or callback, the timeline answer, what happened on the page and whether they came.

Journey and follow-up#

Insights → Journey follows leads whose first conversation falls in the period: how many reached their target, opened their page and attended, by readiness and by how far they went with the page (no page, not opened, opened, or watched, shared or answered). Attendance nobody recorded is its own column, never counted as a no-show. It also shows what leads are interested in (from their conversations and the page sections they read; never care, health or money), how often a shared page is opened by the family, and whether readiness predicts attendance: attendance by the tier each lead had when the visit was booked. If the tiers do not separate who comes, it says so.

Flows → Lead journey draws the same path: first contact, follow-up sent, page opened, what they did next, target, whether it was attended, and the CRM's status where there is one.

Today#

Today (Conversation Intelligence) is the list of tasks: what needs a person, such as a commitment coming due, an interested lead with no next step, or a visit tomorrow whose page was never opened. Every task comes from one of a fixed set of rules, decided in code from what happened. The AI never puts a task on the list, and nothing on it contacts a lead. A task in your CRM is always called a CRM task.

• For me, my communities, for editors. Tasks go to the lead's rep when the lead's owner in your CRM is a staff member with a login (their staff profile names their CRM user and email); otherwise to whoever the visit is booked with; otherwise to the community's backup contacts. Anyone with an active login counts, a Sales rep included: they work theirs on My tasks. A group manager counts only at their own communities. Editors' tasks go to the people who edit playbooks.
• Work a task: Done, Snooze (until tomorrow, for 3 days or a week), Dismiss with a note for the team, Assign it to someone who can see the community, or Reopen it. A task whose rule stops holding (the Hot lead booked) closes on its own and says so. Closing a callback's task or commitment closes the callback too, as done or as dismissed (see Callbacks).
• High-priority tasks email the person they belong to once, with the rule and the community only; in compliance mode the email is a link alone.
• Switch rules off for the account or for chosen communities in Settings → Insights, and

choose which rules also create a CRM task.

Commitments#

A commitment is something promised to a lead. It is a separate record from a task: a commitment coming due puts a Commitment due task on Today. Commitments are recorded from what happened first: a callback the caller asked for, a question the assistant could not answer (someone needs to confirm the answer), and the follow-up page the caller agreed to receive. The analysis adds any other commitment it finds, with the turns it was said in.

Each commitment has an owner (found the same way as Today's tasks), a due time and a status. It closes when someone marks it done, when its callback is done, when its CRM task is completed (where the CRM can tell us), when the page is sent, or when a later conversation with the same lead answers the question from knowledge you added. Commitments show on the conversation page and the lead page, and your own open ones at the top of Today. When the lead has a CRM record, each commitment also becomes a CRM task for their rep.

What a commitment says is treated like the transcript: only people who can read transcripts see it, and it is deleted with the transcript.

Visit briefs#

Before every booked visit, whoever runs it gets a brief on one screen: the situation, who is coming and who decides, their timeline and interests, the concerns they raised and how the assistant answered each, other places they mentioned, open commitments, how they engaged with their page and in chats, and talking points quoted from your playbook (its objection guidance and FAQ answers).

• It is made within 15 minutes of booking, and refreshed the day before and two hours before.
• It never includes care, health or money topics, and never anything about who a person is. Nothing in it is written by the AI except the conversation summary your team already sees.
• The day before, the person it is for gets an email with a link to it (never the brief itself), and a short version goes into your CRM as an activity: the tier, timeline, interests, concerns and open commitments, without the summary or anything the caller said.
• Open the brief from the booking on the conversation page. A rep opens their own from the tour in My tours, and the email takes them straight there. It follows the transcript's retention.

Simulate before you publish#

Simulate on past conversations runs a draft's objectives on the playbook's conversations from the last 30 days, next to the objectives they actually ran with. Nothing is saved.

The Insights area#

Insights in the main menu shows the results at any level. The breadcrumb at the top says what you are looking at: the whole account, one community, one playbook, one version of a playbook, or one playbook as one community runs it (with that community's changes). The period, the channel, where conversations came in (a phone number, the website widget or a follow-up page) and who called (prospects or everyone else) apply to every view; flows always include every kind of caller. The period is in each community's own time zone, and defaults to the last 30 complete days. The channel is Phone unless you choose another, or Every channel: phone calls and chats are not mixed without your asking.

Every number has See conversations, which lists the conversations behind it. Every chart can be shown as a table.

How comparisons stay honest#

• Enough conversations. A rate shows only once it has the minimum (30 unless you raised it). Below that it reads Not enough yet (n = 12).
• A range, not only a number. Each rate shows the range it is 95% likely to fall in, given how many conversations it is based on.
• Better or worse only with evidence. A difference is called Better or Worse only when a statistical test says it is unlikely to be chance. Otherwise it reads No clear difference. Rankings correct for comparing many at once, so a list of twenty communities does not produce false winners.
• Against the rest. A community is compared with the rest of the account, without itself.
• Like with like. Traffic differs: one community gets more website chats, another more existing residents' families. Compare shows each rate raw and adjusted to the same mix of channels, callers and sources, and the mix itself. When the mix is too different to adjust, it says so.
• Different objectives. Playbooks with different main objectives are not compared on main objective attainment. Compare them on objective value, the AI's misses, and the objectives they share.
• A period, not a cause. A trend after a publish is that version's period. It does not show the version caused the change. Only an experiment is labelled Tested.

Flows#

A flow draws many conversations as one diagram: the wider a band, the more conversations took that step. Step by step numbers each step, so a conversation that goes back to questions after answering them still reads left to right; First visits folds a return into the first visit. Steps taken by fewer than five conversations are never drawn, and thin ones join Other, so no one caller can be picked out. The lead journey counts each lead once, on the day of their first conversation. Grey steps are not known yet. Pick another community, playbook or version under Against to colour the steps that changed by more than chance.

Two flows are new:

• Objection flow (Conversation Intelligence): the first objection each conversation raised → whether the agent answered it → whether it came back to the target afterwards → the result. Sensitive objections read sensitive for viewers and are left out in compliance mode.
• Website chat funnel: launcher shown → opened → chat started → first message → engaged (two or more messages) → the stage it ended in → the result. The launcher and open counts are kept per community, so above a community the funnel starts at the chat.

Topics, objections and competitors#

These views use the labels Conversation Intelligence puts on each conversation, so without the add-on they say what they would show. Every share is of the conversations the analyser labelled in the period, with its range, and is called up or down against the period before only when the change is more than chance. Sensitive topics and objections read Sensitive for viewers and are left out in compliance mode.

In one community's view, topics the community added for itself sit beside the account's. At account, playbook and version level, and when comparing communities, they are counted together as one Community topics row, listed below by community.

Knowledge gaps#

Every question the agent could not answer is kept (with the add-on), and similar questions are grouped every hour, so "Do you allow cats?" and "Are cats allowed?" count as one gap. Each gap shows how often it was asked, in how many communities, and the questions themselves, with links to the conversations.

Close the loop from the gap:

• Draft an answer: write the answer yourself. It is added to your knowledge as a document, and the gap is marked fixed.
• Link a fix: point at a knowledge document, a playbook version, or say what was done.
• Dismiss a gap that needs no answer, or Open again one that came back.

A fixed gap shows how often it was asked in the same number of days before and after the fix, so you can see whether the question stopped coming. Owners, admins and editors fix gaps; drafting an answer also needs the right to manage knowledge. The questions are what callers said: they need access to transcripts and are deleted with them.

Chat#

The Chat view is built from what happened on your website, so it comes with every plan:

• The funnel, from the launcher shown to the target reached, with the share at each step.
• Where visitors stopped: chats that ended without a goodbye, by stage and by the last question the agent asked.
• Answer buttons: how often a visitor tapped one instead of typing.
• Return visitors, and identified visitors (a saved contact, a booking or a callback) against anonymous ones.
• Follow-up page chats and how they ended.

Traffic on the question map#

On a playbook's question map, Traffic shows how many calls took each arrow over the last 7, 30 or 90 days, and where calls stopped on a line before finishing it, with the question most of them stopped after. Fewer than five is never shown as a number.

Experiments#

Comparing a new version with the old one's period can only say what happened while each ran: the season, the callers and the staff change too. An experiment runs two (or three) published versions of the same playbook at the same time, on the same traffic, and lets chance decide which conversation gets which.

To plan one, open Insights → Experiments and choose New experiment:

• Where it runs. A phone number, one part of a number's schedule (such as Daytime), or a website widget key. New conversations there are split; conversations already under way keep the version they started with. Test numbers, test keys, test calls and follow-up page chats never take part.
• The arms. Arm A is the version you have now (the control); arm B is the one you want to try. Add a third arm if you need to. Each arm runs the community's own changes on top of its version, as usual.
• The split. 50/50 by default. No arm can have less than 10%.
• The target metric, chosen now, before any results: the one result the experiment is for.
• The minimum sample: how many conversations each arm needs on the target metric before the answer is read. At least your account's minimum.
• The end date. The latest it runs.

It is saved as a draft. Start checks that every version can run in that community and starts splitting conversations. Each conversation is placed by its own ID, so it always runs the same arm, even if it reconnects.

An experiment completes when every arm has reached the minimum sample, or at its end date, whichever comes first. It stops when you stop it, and on its own when something it depends on changes: a new version of the playbook is published, one of its versions or the playbook is archived, or the number's assignment, its schedule or the website key changes. Hold your next publish until an experiment completes. When it stops or completes, conversations go back to the usual version straight away, and the results so far are kept.

The results show every measure for each arm, with its range and how many conversations it is based on. Each arm is compared with arm A. The target metric's verdict is marked Tested once every arm has its sample; before that it reads Not enough to compare. The other measures are there to explore, corrected for comparing many at once, and are never marked tested.

Owners and admins plan, start and stop experiments. Everyone who can see Insights can see them and their results.

Ask#

Ask (part of Conversation Intelligence) answers questions about your conversations in plain English: "Which community booked the most tours last week?", "Is the new main-line playbook better than the old one?", "Why did Lakeview drop in August?". Open it with Ask at the top of any Insights view; it starts from what that view shows.

An answer is a few sentences with up to three parts built from the data: a table, a chart (with its table view), and a link to the view that shows the same thing, with the same communities, dates and filters.

Ask follows the same rules as the rest of Insights, and checks its own answers before you see them:

• Every number comes from Insights data. A sentence with a number the data does not contain is left out, and the answer says it was shortened.
• No false winners. Ask ranks counts ("the most tours"), but calls a rate higher, better or the best only when the comparison supports it. Below the minimum sample it says not enough yet (n = 12).
• No causes it cannot show. It says what changed in a period, and what was published then, never what caused it. A before-and-after comparison is "v8's period". For a tested answer it suggests an experiment.
• It says what it cannot answer. Revenue, and anything Insights does not measure, gets a plain "not available" and never an estimate.
• Every answer states its scope: the dates with their time zone, the level and the filters, for example Sep 16–22, Central Time · Lakeview · Target attainment · Website chat.

Dates. "Last week" is the last 7 complete days, ending yesterday; "this week" and "this month" run to today; "last calendar week" is the previous full week; "since v8" starts at that version's publish. Dates are read in your account's reporting time zone (a community's own zone when you ask about one community), and weeks start on Monday, both set in Settings → Insights. When a period is unclear ("lately"), Ask asks which one you mean.

Follow-up questions build on the last answer: "Now just chat", "What about Lakeview?", "Compare with the month before", "Show me the calls". The context shows as chips above each answer; remove a chip to answer again without it. New thread starts over.

Saved questions. Save a question to run it again later, and put it in your weekly digest: a Monday email with each saved question's latest answer and a link that runs it again. In compliance mode the digest carries links only. Your questions and threads are private to you, and are kept for 90 days after you last used them.

Ask runs as you: a community manager's answers name their own communities and show the others as Community B, exactly as the views do. It reads summaries and counts, never what a caller said, and it cannot change anything.

Ask comes with Conversation Intelligence. If your account does not have it, the Ask button explains what it adds and offers Talk to us.

Who sees what#

Owners, admins, editors and viewers see every level. Community managers see their own communities, and the account as a whole. Other communities appear as Community B, Community C and so on, unless the account turns on Named peers.

Settings → Insights#

• What each objective is worth. Points put playbooks with different objectives on one scale, for example a booked tour 10 and a callback 5. They apply to every community. Changing them re-values past conversations. In compliance mode, conversations whose transcripts were already deleted keep the points they had, because only their totals remain.
• Conversations a rate needs. A rate is shown only once at least 30 conversations count toward it. You can raise this, not lower it.
• Named peers. Whether community managers see other communities by name.
• Taxonomy (Conversation Intelligence): the topics, objections, intents and competitors the analyser uses. See The taxonomy.
• Built-in list (Conversation Intelligence): the senior living list or the general one, used until you publish your own taxonomy.
• Conversations analysed a day (Conversation Intelligence): past it, conversations get the code checks only until the next day. The page shows how many were used today.
• Reporting time zone and Weeks start on: how Ask reads "last week" and "this month". Unset, the zone most of your communities share.
• Ask questions a day: how many questions the account can ask each day.
• Today's rules (Conversation Intelligence): turn each rule off for the account or for chosen communities, and choose which rules also create a CRM task. See Today.

Owners and admins change these settings. Everyone can read them.

The taxonomy#

Settings → Insights → Taxonomy holds the lists the analyser labels with. Your account starts with the built-in list for your industry; that is version 1.

• Topics, objections and intents. Rename them, describe them (the description guides the analyser), add words to listen for, switch them off, mark them sensitive (care, health and money), or add your own. Labels can say {group}, {meeting} or {contact} to use your account's words.
• Competitors. Add the businesses callers compare you with, and the other names they go by. Only competitors on your list are counted.

• Drafts and versions. Changes are saved as a draft. Publish makes it the next version, and every conversation analysed from then on uses it; conversations already analysed keep the version they had. A draft with a problem (two items with the same key, an empty list, or wording about race, religion, age or similar) cannot be published. - Community topics. Each community can add topics of its own on top of the account's. Community managers edit their own communities' topics; owners and admins edit any, and can promote one into the account's draft so every community uses it.

• Re-analyse. Label up to the last 90 days again with the taxonomy in use now. Estimate shows how many conversations, the likely cost and how long it takes: it runs behind new conversations and uses at most half the daily budget each day. The earlier analyses are kept.

Owners and admins edit the account's taxonomy and re-analyse. Everyone can read it.