Skip to main content

Child & Interaction Tickets

A ticket rarely lives in isolation. During its lifecycle it can spawn child tickets for backend work, and — while that backend work is in flight — interaction tickets to keep the customer cared for. This page explains both, and why the AI goes quiet while a ticket is open.

Child Tickets​

A child ticket is a sub-ticket created during a parent ticket's lifecycle to handle one small, well-defined part of the work — an objective — that a human or backend team needs to complete before the AI can continue.

A good way to think about it: the parent ticket is the whole customer request; a child ticket is one specific question or task that has to be answered by a person (often a backend team) before the AI can move forward — for example "confirm whether this order shipped" or "approve this exception."

When a child ticket is created​

While the AI is working a ticket, a workflow can decide the next step must be handled by a human. When that happens, Flowcall:

  • Creates a child ticket in queued, carrying the objective — the parameter to fill, an instruction for the agent, an optional hint, and either a free-text answer or a set of selectable options. If the objective is set to wait in monitoring and the order data says the team can't act yet, the child starts in Monitoring instead.
  • Routes the child to the right team (often a backend team), based on the objective's tags.
  • Names it after the parent (for example T-1024_c1, T-1024_c2).
  • Inherits the parent's channel, order, and workflow, and starts its own SLA timer.

If an unresolved child for the same objective already exists, Flowcall reuses it rather than creating a duplicate — objectives are matched by name, ignoring case and spacing.

If the parent has an owner, automation cannot create the child. Flowcall assigns the parent to its owner for action instead; only the owner may decide to create a child manually.

What happens to the parent​

When a child ticket is created, the parent moves to blocked and is unassigned — it's now waiting on the child's answer and can't progress on its own. This is true whether the parent was in progress, queued, or assigned beforehand.

Objectives are handled one at a time. When a new child is created and an older child is still queued or assigned, that older child is pushed back to blocked — unassigned, its SLA timer stopped, and its previous assignee remembered as the preferred agent — so exactly one child is ever actively worked. The newest objective is the one in front of an agent.

When a child is resolved​

The child's current assignee—or an admin—submits the response and an optional resolution. Being involved in the ticket or owning its parent is not enough to resolve the child. Flowcall then:

  1. Marks the child resolved by agent and copies its answer back onto the parent, under the objective's parameter name. Any resolution text is added to the parent as a note.
  2. Looks at the parent's remaining open (non-interaction) children:
    • If one is already queued, assigned, or monitoring, nothing is promoted — the parent stays blocked and that child continues.
    • If only blocked children remain, the oldest is promoted to queued (with a fresh SLA timer) and the parent stays blocked.
    • If none remain and the parent is unowned, the parent is unblocked back to in progress — the answer flows into the AI workflow, the AI resumes the conversation, and the customer is served again.
    • If none remain and the parent has an owner, the parent moves to assigned and returns to its current owner. The AI does not resume; the owner reviews the child's response and decides what happens next.
  3. Auto-closes any open interaction tickets attached to that parent.

When a parent ticket is resolved directly, any of its still-open (non-interaction) child tickets are auto-closed — moved to closed, unassigned, with a resolution note saying why. When a parent is marked stale, its eligible unresolved children are marked stale instead, with assignments cleared and their activity and SLA timers ended. Open interaction tickets are closed in both cases.

Sharing a resolution with the customer​

When resolving a child, the agent can choose to share the resolution with the customer — it appears to them as a note — or keep it internal. Backend objectives are usually internal; the AI weaves the answer into its reply.

Crucially, a shared resolution is not sent to the customer verbatim. The agent's answer and any note are handed to the AI, which rewrites them in your account's customer-facing voice before sending. A backend agent can jot a terse, internal-sounding note — "courier says address incomplete, need landmark + alt number" — and the customer still receives a polished, on-brand message. This is why backend teams don't need to be trained on customer tone: they supply the facts, the AI handles the phrasing.

Child ticket history​

On a ticket that has had more than one child ticket, the ticket panel shows a Child ticket history timeline below the child currently being worked. It reads as a chain of handoffs, so an agent picking up the ticket can see what already happened without opening each child:

  • Route. The teams the ticket passed through, in order — for example Tech → Part Booking → Tech → CRM L2 — with a summary of how many handoffs there were, how long they took, and which teams got the ticket more than once.
  • Each handoff. Who sent it, from which team, to which team, and when. A 2nd time / 3rd time tag marks a ticket going back to a team that already had it.
  • Why it was handed off. The note written when the child was created sits on the handoff, together with any ticket note the same agent added in the 30 minutes before it. A replacement with no note at all says No handoff note.
  • What happened while a team had it. How long the team held the ticket, who worked it, its answer or resolution, and the ticket notes written during that time.

Long chains show the first handoff and the latest three; Show earlier expands the rest. The same timeline appears under closed tickets in Customer Tickets.

Waiting for order data (Monitoring)​

Some child tickets can only be worked after something happens outside Flowcall. For example, a customer asks "where is my order?", the AI finds the order was returned to origin (RTO), and the order should be re-dispatched — but your warehouse team can only re-dispatch it once the return actually reaches the warehouse. A Monitoring child ticket waits for that, out of the queue, and then reaches the team on its own.

How it works:

  • Created in Monitoring only while the data says to wait. When the workflow creates the child, Flowcall checks the objective's monitoring conditions against the parent ticket's data points (for example return_status is one of rto_marked, rto_in_transit). If they all hold, the child starts in Monitoring; if the data already shows the return arrived, the child goes straight to the team's queue.
  • The parent is blocked and the AI holds the conversation. As with any child ticket, the parent moves to blocked. If the customer writes in, the AI tells them the configured monitoring message (or a short "we're tracking this and will act as soon as the update arrives" message) instead of promising that an agent will join.
  • Order re-evaluation releases it. Whenever the order is re-evaluated — a tracking or order webhook (ClickPost, Shopify, your custom webhooks) or an agent clicking Re-evaluate — Flowcall refreshes the order data points and checks the conditions again. Once they no longer hold, the child moves to queued for its team, its SLA timer starts, and it's assigned like any new ticket.
  • The timeout always releases it. If the data never changes, the child moves to the same team's queue when the monitoring timeout (72 hours by default) passes.
  • No manual release. A monitoring child can't be assigned or moved to the queue by hand: the order data is the single source of truth. If you suspect a missed webhook, open the child (or its parent) and click Re-evaluate — it fetches the latest order and tracking data and tells you whether the child was released or why it is still monitoring, with the auto-release time.

The ticket log records when the child entered monitoring, the conditions it is waiting on, and whether re-evaluation (with who or which webhook triggered it) or the timeout released it.

Setup. In the Data Library Objective Library, open a human-agent objective and turn on Wait in monitoring. Add one or more conditions — a data point, an operator, and a value; all must hold for the child to wait — then the timeout in hours and, optionally, the message the AI gives the customer while the child waits. Use order data points for the conditions: re-evaluation refreshes order data, so a value collected from the customer would never change. You can also ask the Copilot to set it up, for example "keep the warehouse redispatch child in monitoring while return_status is rto_in_transit, release after 5 days." Monitoring applies to children the workflow creates; children an agent creates manually go straight to the queue.

Awaiting information (keep the child open)​

Sometimes the agent working a child ticket can't answer yet — they're waiting on the courier, another team, or the customer. Instead of submitting a resolution, they open the dropdown next to Submit and choose Awaiting information. The parent and its open queued, assigned, and blocked children move to an awaiting-response state together and are unassigned. Resolved children stay resolved.

This differs from Snooze, which pauses the child for a set number of hours and then brings it back automatically. Use Awaiting information when you're waiting on someone else's input, and Snooze when you just want to defer the work.

A child can also be put into an awaiting state by the parent: when a parent ticket is marked awaiting response by system, its queued, assigned, and blocked children are marked awaiting by system too and unassigned. Workflow refreshes preserve this pause until the family is reactivated.

Removing awaiting response from the parent or any awaiting child clears it across the family. A customer reply does the same. Children that were queued or assigned return to the queue; children that were blocked return to blocked. The parent returns to blocked while child work remains, or to queued when there are no open children.

Request information from the customer​

Sometimes the agent can't finish because they need something from the customer — a new date to reschedule a technician visit, a delivery address, a photo reference. Instead of messaging the customer themselves and remembering to check back, the agent clicks Request from customer on the ticket and picks what they need.

  • The ticket leaves the queue. The ticket the agent was working on (the active child, or the main ticket when there is no child) is set to blocked and shows Waiting for customer input. A small AI-owned request ticket appears in the family while the request is open.
  • The AI asks the customer. It writes a short question and sends it on WhatsApp and/or email. On WhatsApp it uses a normal message when the customer wrote in the last 24 hours, and the selected approved template otherwise.
  • Replies go to the request, not the conversation. While the request is open, the customer's WhatsApp replies and email replies on the request thread are read by the request. An unrelated email on another thread still reaches normal support. If an answer is incomplete or invalid (for example a date in the past), the AI asks a follow-up question. If the customer asks for a person or cancels, the request ends straight away.
  • The ticket comes back with the answer. The value is saved on the ticket (shown under Customer provided) and in the ticket's data. Then the ticket returns to the queue, to the same agent when they are available if that option was ticked. The AI conversation is not restarted.
  • No reply. The AI sends reminders. At the timeout the ticket either returns to the queue with a note, or the whole ticket is closed, as chosen when the request was sent.

See Request Information from the Customer for setup and the full behaviour.

From the open request, an agent can view the Conversation, Enter answer themselves (for example when the customer phoned in), or Cancel request. While a request is open, adding a child ticket, marking the ticket awaiting and snoozing are not allowed.

Setup. An objective appears under Request from customer once Agents can request this from the customer is ticked on it in the Data Library (or the Copilot sets it up). There you choose the answer type (text, number, yes/no, date, date and time), the WhatsApp template and its variables, reminders, timeout and the default timeout behaviour. Agents need the Request information from customer permission, which agent roles have by default.

Automatic Resolution from Tracking Updates​

Many child tickets exist only because an order is in flight — a dispatch is delayed, a shipment is "where is my order," a delivery is stuck. Once the shipment actually moves, the ticket is moot. Flowcall closes these out on its own when your logistics provider sends a tracking update, so the backend team doesn't spend its day manually closing tickets the carrier has already resolved.

When a tracking webhook arrives — from ClickPost, Shiprocket, or a Shopify fulfillment event — reporting that the order was dispatched, is out for delivery, was delivered, or went RTO, Flowcall does two independent things:

  • Tells the customer. It sends the WhatsApp status update your account has mapped to that tracking event (see Broadcasts → Events). The same event won't message the same customer twice within 24 hours.
  • Re-evaluates the open ticket. It re-runs the ticket's workflow against the fresh order state. If the original concern is now satisfied — the delayed order has shipped, the "where is my order" is now delivered — Flowcall auto-resolves an unowned parent ticket (as resolved by AI) and cascades auto-close to its child and interaction tickets. An owned parent cannot be auto-resolved; it is assigned to its owner for action instead. If the concern still isn't satisfied, nothing is closed and the ticket stays with your team. Any monitoring child whose conditions no longer hold is released to its team's queue.

Re-evaluation is de-duplicated per tracking event for 24 hours, so a burst of near-identical carrier updates only triggers one pass. This only auto-resolves tickets whose remaining work the tracking update actually makes unnecessary — it never closes a ticket that still has open questions.

In the ticket timeline, a resolution or monitoring release caused by re-evaluation carries an Order re-evaluation tag naming what triggered it: the webhook (for example Shopify orders/fulfilled or Clickpost tracking-update), a manual Re-evaluate, or a bulk refresh.

Resolving Many Child Tickets at Once​

When dozens of child tickets pile up on a backend team — say fifty "delayed dispatch" orders that all shipped over the weekend — resolving them one at a time is tedious. Flowcall lets a team clear them in a single pass with an Excel round-trip, from the Ticket View.

  • Export Child Tickets downloads your open child tickets as a workbook with one sheet per objective type (all the "delayed dispatch" children on one sheet, all the "need alternate number" children on another). You can filter the export by task, team, or assigned agent — only options that actually have open child tickets are offered — and non-admins only see child tickets assigned to or involving them. Each row carries the ticket, its parent, the customer, the order, and the ticket's captured fields for context.
  • The team fills in the answer column (named after that sheet's objective) and an optional Resolution Note offline. Rows left blank are skipped, so partially filled sheets are fine.
  • Import Resolutions re-uploads the workbook. Each filled row runs the same resolution as an individual submit. The row is accepted only for the child's current assignee or an admin, and assignment is checked again during background processing. An unowned parent can return to the AI workflow; an owned parent returns to its owner. A single Share resolution note with customer choice applies to the whole batch. Rows are processed one at a time with a small delay so customer messages stay staggered, and you can watch progress as it runs.

Export handles up to 2,000 child tickets at a time and each import up to 500 rows; split larger sets into multiple files.

Tip: If your logistics provider can send Flowcall a tracking webhook, you often don't need to bulk-resolve unowned delivery tickets at all — tracking updates can auto-resolve them. Owned parents always return to their owner for the final decision.

Why the AI Stops Responding​

When a ticket is open with a human — and especially when a backend child ticket is in progress — the AI stops auto-responding to the customer. This is deliberate: it prevents the AI from talking over a human who is mid-investigation or contradicting an answer the backend team hasn't finished yet.

Practically, the customer's AI is switched to manual while the human handles things. Any messages the customer sends in the meantime are held rather than auto-answered. So the customer isn't met with silence, the AI sends a short acknowledgement in the customer's language — that their request is with your team, and when they'll join if you're outside opening hours. It sends at most one every 15 minutes, and none while an agent has replied in the chat in the last 30 minutes on WhatsApp or live chat. For an unowned parent, the AI is re-enabled automatically once the work is resolved or the blocking child is answered and the parent returns to in progress. An owned parent returns to its owner after the final child instead, so the AI remains paused.

AI auto-replies can also be paused at the account or channel level independently of tickets — see Settings → Channels.

Interaction Tickets​

Backend work takes time. A backend team might need hours to confirm something, and while the parent ticket sits blocked, the AI is paused — so a customer who messages in the meantime would otherwise get silence. Interaction tickets exist to handle exactly that situation.

An interaction ticket is created when a customer reaches out while a backend ticket of theirs is open and the AI is paused. It routes that customer interaction to a front-line team so someone can respond, without disturbing the backend work still underway.

When one is created​

An interaction ticket is created when all of the following hold:

  • The customer has an open backend ticket (a ticket on a backend team that isn't itself an interaction ticket).
  • That backend ticket has been waiting at least a configured delay — interaction tickets are meant for the gap when the backend team is taking a while, not for instant replies.
  • The customer sends a message during that window.

Interaction tickets must be enabled for your business, with a delay and a designated interaction team configured.

How it behaves​

  • It's created in queued and linked to the main ticket (the parent of the backend work, or the backend ticket itself) so the context stays connected. It's named after the main ticket (T-1024_i1, T-1024_i2).
  • There is at most one open interaction ticket per customer on a main ticket — Flowcall won't pile up duplicates if the customer sends several messages.
  • It's routed to the configured interaction team (a front-line team) and assigned like any other ticket.
  • It carries the customer's mood at the time it was raised, which is also logged on the main ticket.
  • Resolving an interaction ticket does not unblock or resolve the underlying backend work — it only handles the customer's immediate message. Unlike a child ticket, it never changes the parent's status.
  • When the underlying work finishes — the backend child is resolved, or the parent is resolved, snoozed, marked awaiting response, or marked stale — open interaction tickets are moved to closed automatically.

In short: child tickets push work down to a backend team and block the parent; interaction tickets keep the customer attended to while that backend work runs.

Linked Contacts​

Sometimes a different customer — usually someone from the same household, on another phone number or email address — reaches out about an order that already has an open ticket. Instead of opening a second ticket for the same problem, Flowcall can attach them to the existing ticket as a linked contact: an interaction ticket that belongs to the new customer.

When a contact is linked​

The AI links the new conversation as soon as it knows the order, before asking any further questions or running any actions. All of the following must hold:

  • Link other contacts is turned on in Ticket Lifecycle, with an interaction team selected.
  • The conversation is on WhatsApp or email (email accounts in passive analysis mode are not linked).
  • Another customer has an open ticket for the same order that is with your team, whatever request it was raised for — queued, assigned, blocked, or awaiting a reply. Tickets still being handled by the AI are not linked. If the order has more than one open ticket, the newest one is used.

If the order differs, or the customer asks for an agent before giving the order, they get their own ticket as usual.

What the customer sees​

What happens depends on the mode chosen under Link other contacts.

Link straight away (the default). The customer gets one acknowledgement that the team is already working on a request for this order — naming the channel and time it was raised, such as "on WhatsApp on Oct 3 at 4:15 PM" in your business timezone — and that their message has been added to it. The other customer's name and contact details are never mentioned. Further messages are held for the agent, just like any customer whose ticket is with a human. On email, later emails on the same thread are attached to the linked ticket without another reply.

Ask first. The customer is told the team is already working on a request for this order, with the same channel and time, and asked whether they would also like to be connected with the team. On WhatsApp the message has Connect me and No, thanks buttons; on email they are asked to reply if they want to be connected. Their own AI ticket is closed as resolved by AI, without a CSAT survey. Their next reply decides:

  • No, thanks, or any clear decline: they get a short confirmation that the team will carry on with the existing request. No ticket is opened.
  • Anything else — Connect me, a yes, new details, or a complaint: they are linked as a linked contact and told the team will reply, including when if you are outside opening hours.
  • If the open ticket has closed in the meantime, their message is handled as an ordinary new conversation.

If the customer seems frustrated, angry or asks for an agent on the same turn as the question, they are linked straight away instead of waiting for their answer. The offer is remembered for 24 hours: if they write about the same order again in that time, after declining or without answering, they are linked without being asked again. On email, only a reply on the same thread answers the question.

How it behaves​

  • The linked contact is routed to the interaction team and assigned like any other ticket. It is not assigned to the agent working the main ticket.
  • It appears as its own card in the new customer's chat and in the assigned agent's My Tickets. The card shows the main ticket it is linked to, its status, and who is working on it.
  • The main ticket's card lists it under interaction tickets with a Linked contact tag, and the main customer's row in the chat list shows a link tag with the number of linked contacts.
  • A linked contact has its own SLA. Its timers and breaches show on its own customer's row and never on the main customer's.
  • It can be replied to and resolved, but it cannot be marked awaiting a reply or snoozed — those actions would affect the other customer's ticket.
  • It closes automatically whenever the main ticket moves on: a child ticket is resolved, or the main ticket is resolved, closed, snoozed, marked awaiting response, marked stale, or closed by the customer's exit keyword. Because linked contacts are usually from the same household, the update sent to the main customer covers them too; Flowcall does not message the linked contact separately.

Linking cannot be undone. If a contact was linked by mistake, the agent resolves the linked ticket and helps the customer in the chat.

  • Ticket Lifecycle — the statuses these tickets move through.
  • Ticket View — working child and interaction tickets in the ticket panel.
  • Agent Assignement & Team — backend vs front-line teams, and how each of these tickets is routed.
  • Workflows — where objectives that become child tickets are configured.