Module 12 — Circle Intelligence

Every conversation tells you something. Here's what you can do with it.

Lina doesn't just answer questions — it pays attention. Every chat and phone turn is read for what the person wants and how they feel, the moment it happens. Those signals build up into a fuller picture of each person, get sorted into simple groups, feed reports on where people come from, and power side-by-side tests — without storing a single message, and without slowing down a single reply.

← Back to the platform overview

The sensing layer

What people want and how they feel, read turn by turn — no message text saved.

A simple checker runs on every public chat and phone turn, in the same step that produces the bot's reply. It looks for keywords — no AI model, no extra delay — and tags each turn with a category for what the person wants and a direction for how they feel. The tags are saved; the actual message text is not. This is the base layer everything else reads from: profiles, groups, and the task list all use these tags, never the raw conversation.

Privacy by design: the tag store keeps only the tags themselves (for example, volunteer, positive) plus a random session or call ID. No message text, no name, no phone number, no email ever enters this layer. Personal info stays in your CRM; tags stay in their own store — always kept apart.

Intent categoriesinterested — said something showing general interest in the organization or its work. volunteer — said they want to help out. donor — said they want to give. urgent — said something time-sensitive or urgent. hostile — an aggressive or abusive message (this brings in a human right away).

Sentiment directionpositive, negative, or neutral — tracked separately from what they want. A person can be both positive and urgent, or hostile and negative, or interested with no particular feeling attached.

PrecedenceWhen more than one tag could apply to the same turn, this is the order that wins: hostile > urgent > donor > volunteer > interested. The most important tag is the one that gets saved — so one confusing message can't create a misleading profile.

Fail-openIf the tagging step fails or can't save, it never breaks the chat or phone call itself. This layer only watches — it never changes a bot's reply and never sends an alert on its own.

Circle profiles

One full profile per person, built from what they do across chat and phone.

Circle holds one record per person, matched first by email, then phone, then session ID. As a person chats and calls, their record fills in over time: how many sessions, how many calls, which channels they've used, when they were last seen, whether they've given, and the history of tags from the sensing layer. The profile combines who they are (from your CRM) with what they actually do — a fuller picture than either gives you alone, and your team never has to tag anything by hand.

No duplicate contacts. A person who chats on Tuesday and calls on Thursday is treated as the same record — matched by email, then phone, then session — not as two separate people.
What they do, not just who they are. The profile tracks how many times they've engaged, which channels they used, and what they've said they want — real behavior, not just name and contact fields.
Automatic, not manual. Profiles fill in on their own as conversations happen — your team never has to tag or sort anything; the system just watches and records.
Lifecycle tracking. When they were first seen, when they were last seen, how engaged they are, and which channels they use — so you can tell a first-time visitor from a frequent repeat contact at a glance.

Personas

Lasting group labels — found in patterns, never assigned by hand.

Personas boil a person's whole profile down into one short, lasting label you can act on. They're worked out automatically from their tag history, how engaged they are, and which channels they use — no manual tagging, no survey, nothing to import. A persona isn't a guess; it's a pattern the system actually observed across several conversations. And because it's based on live signals, the label updates as a person's behavior changes.

High-intentPeople who've said they want to give or help out, more than once — the top-priority group for your team to follow up with.

Multi-channelPeople who've reached out through two or more channels (chat, phone, email). They tend to be more invested — and more likely to respond when you follow up.

At-riskPeople who went quiet after showing real interest before — they spoke up, then nothing. Flagged on the task list as a possible lapse worth addressing.

Repeat callerPeople who call far more often than they chat — useful for deciding who to prioritize on the phone and for staffing decisions.

Derived, not assignedPersonas are worked out fresh each time, from the profile's current state — they update on their own as more signals come in, with no batch job and no manual re-sorting needed.

Attribution

How far you reach, where people overlap, and what converts — reported honestly.

This reporting tool answers the questions organizers and business operators actually ask: how many people have we reached on each channel, how many are on more than one, and how engaged is our base overall? It's built directly from Circle — it adds up real, observed facts, not a model's guess — and it tells you plainly when a number needs a caveat.

Channel reach. How many distinct people you've reached on each channel — chat, phone, email — plus how many show up on two or more.
Engagement depth. How many people fall into each level of engagement — one session, five sessions, ten or more — so you can tell a wide, shallow reach apart from a smaller, deeply engaged group.
Conversion tracking. Gift and sign-up rates by channel, with an honest heads-up: since most gifts come in by phone, phone numbers will look better than email or chat — that's just how the channels mix, not a real ranking of which is "best." The report says this plainly instead of hiding it.
No guesswork. These numbers are a roll-up of real facts already in Circle — last channel seen before a conversion, engagement history — not a model's prediction. What you see is exactly what happened.

Experiments

A/B testing — the same visitor always gets the same version, off by default, no slowdown.

This tool lets you run controlled tests on chat and phone content — different opening messages, different points where a human gets pulled in, different timing for follow-up — with a sound way to split visitors and tally results. Each visitor is sorted into a group using a fixed calculation based on their session ID, so the same visitor always sees the same version, with nothing extra stored and no added delay. Tests are off until your team turns one on.

Because the grouping is calculated, not stored, there's nothing to keep in sync between services. A session with many back-and-forth turns always lands on the same version — not because of a database lookup each time, but because the calculation gives the same answer every time. No cookies, no session lookup, no delay.

Off by default. The testing tool is there but does nothing until your team turns a test on. Turning on Circle Intelligence does not also turn on testing.
Consistent grouping. Which version a visitor sees is worked out from their session ID — it stays the same across turns and channels, with nothing extra to store and no lookup needed each turn.
Results that connect. The system matches who saw each version to who actually converted, by session ID — so you get real conversion rates per version, not just how many people saw it.
Storage stays bounded. The list of who-saw-what is capped so it can't grow forever — the same storage approach used everywhere else in the platform.
Ready for live content tests. There's a documented way to vary live chat content by version, but it's not turned on by default — so test traffic can't accidentally change your real site content.

Dispatch board

One list of everything that needs a person's attention — from every channel.

The Dispatch board pulls together everything that needs a human, from every channel: chat hand-offs, phone callbacks, at-risk people flagged by the tagging layer, and messages waiting for approval in the send queue — all in one place, with one count, so your team knows exactly what to work on the moment they log in. If any one source breaks, it just counts as zero — it never breaks the whole board.

Everything in one place. Chat hand-offs, phone callbacks, flagged people, and pending approvals are all counted together — not spread across four separate panels.
Built to keep working. If any one source has a problem (an outdated file, a network hiccup), it just counts as zero — the board still loads and never breaks.
Always current. The count is refreshed every time the dashboard loads, so your team sees what's true right now without refreshing a separate panel.
Linked to alerts. The same things that show up on the dispatch board can also be set to send Smart Alerts by email or Telegram — so your team finds out even when they're not logged in.

Editions & availability

An advanced capability we turn on with you.

Circle Intelligence is an advanced capability of both the Campaign and Business editions. Once it's turned on for your install, the sensing layer runs on every chat and phone turn right away — no setup needed to start collecting signals; reporting, groups, and testing are each opt-in from there. Because we turn it on per organization, contact us and we'll scope it to your account:

Campaign edition

Circle & voter insight

Track what voters and your Circle want, across chat and phone. Find the helpers and donors who are most ready to act. See where engagement is coming from across your channels, with honest notes on the numbers. Run A/B tests on content voters see. Use the dispatch board to prioritize follow-up with the people most likely to convert.

Business edition

Customer insight

Track what customers want, across support chat and the phone front desk. Spot repeat callers and customers who use more than one channel. See where incoming volume and conversions come from, by channel. Run tests on your support messaging. Use dispatch to triage the customers who need attention before they leave.

Live demo

Watch it answer real questions, right now

This demo is the exact same system real customers get. Try it as the Campaign edition or the Business edition.

Get started See the Campaign edition demo → See the business demo →

Who it’s for

For teams that want to understand the questions, intent, and sentiment in conversations already happening.

How it works in 3 steps

1
Choose your plan and sign up. If an invitation is required, request a code first.
2
Add your organization’s knowledge and brand in the dashboard. See the setup guide.
3
Share your web address and start answering from your knowledge. Phone and staff drafting tools are available on eligible plans; compare features.

Get started