FlyB Hunt AI

Sidebar: Communication → Inbox — the purple Hunt AI button in the Inbox toolbar.

Module: Inbox — active on every account from the start.

Plans: searching is included in Plus and Prime, and each search costs Flights. On Check-in (Free) you can open Hunt AI, write and save Hunt Rules — the search button asks for an upgrade.

Somewhere in this week’s conversations there is the person who asked for a price and never got a proposal, the one who said “let me talk to my partner” and went quiet, the customer who complained twice. Nobody is going to re-read three hundred threads to find them.

FlyB Hunt AI does. You describe who you are looking for in one sentence — a Hunt Rule — and it reads every contact who was active in the period you chose: their conversations on every channel, your team’s notes and tasks, their appointments, the e-mails they opened and clicked, where they are in your funnels, their custom fields. Each contact gets a score for how well they match. The ones who match become a list right in your Inbox queue, best match first, ready to be worked with every action the queue already has.

Where it is

Open the Inbox and press the purple Hunt AI button in the toolbar, beside the filters. The Hunt AI screen takes the middle of the screen — where the Inbox overview is while no conversation is open — and the conversation list on the left stays as it was. The FlyB Hunt AI card at the bottom of the overview opens it too.

The × at the top right of the screen (or Esc) goes back to the overview; a conversation you open from a Hunt Rule’s list comes back to Hunt AI when you close it. While a search is running, the toolbar button spins and shows its progress, wherever you are in the Inbox; while a Hunt Rule’s list is in the queue, a dot on the button says so, and the line above the list names the rule. On a phone, tap the button to open the screen and use the back arrow to return to the list.

The period

The two dates at the top of the screen — from and to, days only — scope everything below them. The first day starts at 00:00:00 and the last one ends at 23:59:59, on your own clock. The screen opens on the last seven days through tomorrow, and the shortcuts beside the dates switch to 7, 30 or 90 days or to the current month.

The counters

Four numbers for the period, each compared with the period of the same length right before it (an arrow and a percentage — or “no data in the previous period” when there is nothing to compare with):

They follow the same visibility as the rest of the Inbox: if you see only your own contacts, or only some WhatsApp numbers, the counters count only what you see.

Hunt Rules

A Hunt Rule is three things:

Rules belong to your organization: everybody on the team sees the same Hunt Rules.

Your first rule is already written

The first time an account opens Hunt AI with no rule at all, one is already filled in: Interested in a meeting — “Contacts who showed interest in our products or services but have not booked a meeting yet”, at 75%. Change anything you like, or press Search with Hunt AI as it is.

Below your rules, Ready-made rules offers more starting points — asked about price, ready to buy, waiting for our reply, unhappy customers, going cold. Each opens as a new rule you can edit before saving. Duplicate copies an existing rule to try a variation.

Writing a good question

The AI answers the question you wrote, read literally — so write it the way you would brief a colleague who will read every conversation for you:

• One sentence, every condition explicit. “Asked for a quote and has not received a proposal” works better than “hot leads”.
• Say what should be absent, too. “…and has not booked a meeting yet” is a condition the AI checks, using the contact’s appointments.
• “Our products” means yours. The AI knows your organization’s name and sector, so a question can speak in the first person.
• Leave counting and dates to the period. Ask for what happened, and use the period to say when: “complained about delivery” over the last 30 days, rather than “complained more than twice since the 3rd”.

Searching

Search with Hunt AI saves the rule and starts the search. Its price in Flights is on the button and right above it, before you press it — see What a search costs. The search runs on FlyB’s servers: you can open conversations, leave the screen or close the tab, and the list updates by itself when it finishes. A bar shows how many contacts have been analysed so far, and Cancel search stops it.

A period in which no contact was active has nothing to analyse: the button stays off and the screen asks for a longer period, so an empty search never replaces a rule’s list.

Which contacts are analysed

Every contact who was active in the period, and not only the ones with a conversation:

• contacts created in the period;
• contacts who wrote, or were written to, on WhatsApp or FlyB Chat;
• contacts with an internal note, an annotation or a task created, edited or due in the period;
• contacts with an appointment booked in the period or taking place in it;
• contacts who clicked a link in an e-mail in the period;
• contacts who placed an order in the period.

A contact created long before the period is analysed when they did any of the above inside it. Each search analyses up to 2,000 contacts; when a period has more, the most recently active are analysed and the screen says how many were left out.

What the AI reads about each contact

The period’s messages (on every channel), notes, tasks, e-mails (subject, opened, which buttons were clicked) and funnel moves — plus the contact’s current situation, whatever its date: upcoming appointments, open tasks, pinned notes, the funnel stage and live subscriptions — and the record: where the contact came from, the campaign that brought them, company, city and state, the domain of their e-mail address (and whether it is a free provider like Gmail or an organization’s own), the country code of their phone (+55, a phone number from Brazil), qualified flag, Lead Score, tags and custom fields. Facts that depend on arithmetic (has an upcoming appointment in 2 days, the last message is the contact’s and has not been answered, has never booked) are worked out by FlyB and handed to the AI as plain sentences.

Only what your rule is about

Before the first contact, Hunt AI asks the AI one question about your rule — not about any contact — to decide which parts of a record bear on it. Every other part is left out of every record sent: for “contacts from Brazil who opened an e-mail” the AI reads the profile, the e-mails and the custom fields, not a word of the conversations. The record says that a part was left out on purpose, so its absence is never taken as “nothing happened”. The profile and the computed facts are always sent.

The same question finds the parts your rule cannot be met without. A contact with nothing at all in one of them is given 0% straight away, without being sent to the AI: for “came from Instagram and booked an appointment”, a contact who never booked anything cannot match. It is still an analysed contact — on the score histogram, and in the price. A rule about something that did not happen (“never booked”) never rules anybody out this way: there, an empty part is the match.

The rule’s last search says what was left out and how many contacts were ruled out, under its score histogram. Less to read is also a better reading: an AI model judges more accurately when the record holds only what the question is about.

What it never reads: the contact’s name, e-mail address, phone number or street address — of the e-mail and the phone, only the domain and the country code above. They are not sent as fields, and they are removed from the words too — an e-mail address, a phone number or a document number typed in a message, and the contact’s own name, reach the AI as [e-mail], [number] and [name]. Nothing obtained from Google is sent either: contacts imported by Active Search contribute none of the details Google provided, and Google Meet links are removed from messages.

What a search costs

A search costs 1 Flight per 100 contacts analysed, with a minimum of 1 — every block of 100 that is started counts as a whole one:

The rate on screen is the one that applies: the line above the button reads Cost: N Flights (1 Flight per 100 contacts, minimum 1) · Balance: N Flights, and the button carries the same number.

Charged once, before it starts

The whole price is debited from the organization’s Flights when the search starts — counted on the contacts it is about to analyse, which are chosen at that moment — and it is one line on the Flights Statement, Hunt AI search. If more contacts became active in the period between the moment the screen showed the price and your click, the search is not started and nothing is charged: the screen shows the new price and you search again.

Searches are paid from the organization’s balance, whoever runs them — personal lists included.

Contacts ruled out without the AI — those with nothing in a part the rule requires, see Only what your rule is about — are analysed contacts and count in the price like any other.

What comes back

When the search ends, the Flights of what it did not do are given back automatically, as a second Hunt AI search line on the statement:

The search history and the rule’s last search show what each search cost and what was refunded.

Not enough Flights

If the organization’s balance does not cover the price, the button stays off and the screen says how many contacts your balance does cover. Shorten the period at the top of the screen until fewer contacts are active, or buy Flights — the button is right there. See Buying extra Flights.

After a downgrade

An organization that drops back to Check-in keeps everything it made: its Hunt Rules, and the lists of the searches it ran while it was paying — still on screen, still in the queue, still re-cut by the minimum match. Only new searches wait for an upgrade.

The list

Pick a rule and the conversation list becomes its list: the contacts of its latest search at or above its minimum match, best match first. A line above the list names the rule, how many contacts it holds and when the search ran; its × brings back the queue you were on before.

Each card carries its score — the purple percentage beside the time. Hover it to see how sure the AI was (high, medium or low confidence) and where it found the strongest evidence: conversations, notes, tasks, appointments, e-mails, the CRM pipeline, purchases, custom fields or the contact’s details.

Everything else is the Inbox you know: open a contact to read and answer, and use the bulk actions — assign, tag, add to a funnel stage, mark qualified, resolve — on the whole list or part of it (Select all takes the whole list, not only the contacts loaded on screen). The views in the toolbar’s list still work, counting inside the list: Unread within the hunt, Waiting on us within the hunt.

Changing the minimum match

Every contact a search analysed keeps its score — not only the ones that made the list. So moving the minimum match and saving re-cuts the list immediately, with no new search. While you drag the slider, the rule tells you how many contacts of the last search the new value would keep, and the score histogram shows how all of them spread from 0% to 100%, with your minimum match drawn across it.

Searching again

A new search replaces the rule’s list when it finishes. Until then the previous list stays; a search that fails or is cancelled never touches it. Changing the question does not change the list on its own — the rule warns that its list still answers the previous question until you search again.

Search history keeps each rule’s latest searches: when, by whom, over which period, how many contacts were analysed and found, and why a search failed.

Who sees what

Hunt AI reads only what you may see, and a list is never shown to somebody whose profile could not have produced it:

• Members who see the whole organization — every contact, every WhatsApp number and every funnel, with the Notes & Tasks and Financial permissions — share one list per rule: a search by any of them replaces it for all of them.
• A member with any of those restrictions works on a personal list: their searches analyse only the contacts, conversations, funnels and data their profile reaches, and only they see the results. The screen says so.

Reading Hunt AI follows the Inbox’s view permission. Writing, searching, cancelling and deleting rules follow the Inbox’s edit permission — the same that assigns and resolves conversations. Deleting a rule removes its searches and the lists of every member; contacts and conversations are not touched. See Settings & Team.

Limits

When a search fails

In every case the rule keeps its previous list, and every Flight the search cost is refunded.

A search can also be refused before it starts — nothing is charged then:

What Hunt AI does not do

• It never writes to anybody. It finds people; what you send them, and whether, is yours — from the conversation, a template, or the bulk actions.
• It never changes a contact. It does not tag, qualify, move or assign anyone by itself, and it does not touch the Lead Score. To have the AI act on its own — move the lead, assign it, answer it — when a contact writes or enters a pipeline stage, use an AI condition in a CRM stage automation (Prime).
• A score is a judgment, not a fact. It is generated by an AI model from what the record shows, and can be wrong — read the conversation before acting on it, and never use a score as the sole basis for a decision with legal or similarly significant effects on a person. See Terms of Use, section 8.

Privacy

The AI that scores your contacts is Jev, a decision model by TypeSafe AI, Inc., listed in Sub-processors. It processes each contact’s record under contract with FlyB, only to return the score, and TypeSafe does not train its models on the requests it receives. What reaches it — and what never does — is described in What the AI reads about each contact. See the Privacy Policy.