Courier routing and packer prioritization

Sequence the flight when three storefronts are live at the same dinner rush.

A multi-brand ghost kitchen running three storefronts at the same dinner rush has three independent ticket streams, three pack sequences, and one walk to the handoff point. The operator without routing logic ends up holding a courier at the kitchen door while a second bag is still on the line — one delivery late, one rating at risk, one storefront losing a tier it had been building for three weeks. Braiseflux reads the kitchen flight across all active storefronts on a ticket-by-ticket basis — POS timestamp, per-SKU prep time, courier ETA — and surfaces a ranked pack sequence per active handoff point at the packer console. The kitchen-level routing loop pairs with the menu-swap cadence on the storefront side and with the inventory reorder cadence on the procurement side, and pairs with the unified portfolio read across every channel. The broader four-agent bundle this loop lives inside sits at multi-brand restaurant management.

Start free →See pricing

The operator problem

What courier holds, envelope mix-ups and load imbalance actually cost a multi-brand operator.

Routing and sequencing automation is not a marketing-page feature; it is the only sustainable way to staff a flight-level decision that arrives on three storefronts with three independent courier ETAs against a single walk-in and a single handoff shelf. Below are the three pains that show up on every multi-brand kitchen’s weekly courier-route close, in roughly this order.

Couriers held at the handoff while the bag is still on the line.

A multi-brand kitchen running three storefronts at the same dinner rush has three independent ticket streams and one walk to the handoff. The courier for the earliest-completing sushi ticket arrives at the door while the Nashville hot ticket is still on the fryer and the shawarma ticket is still warming — and the dispatcher, who happens to be the packer, cannot seal a bag that does not exist yet. The courier waits two minutes; the marketplace records a hold against the on-time threshold; the rating lands before the courier leaves the kitchen. Across three storefronts the hold compounds: a single ranking-cost event on one storefront drags the marketplace score for the whole kitchen because the courier queue is read as one flight.

Pack sequencing that crosses the temperature envelope.

A sushi bag and a Nashville hot bag cannot share a courier run without the temperature integrity of both failing — but the kitchen packing in order-of-ticket has no way to keep the two envelopes apart once the flight goes multi-storefront. The packer assembles the bag the courier is about to grab rather than the bag the courier can carry; the courier picks up a cold item on top of a hot one; the customer opens an envelope that never should have been sealed together. The failure surfaces as a one-star review mentioning cold food — a complaint that is not a menu problem, a prep problem, or an inventory problem, but a sequencing problem that costs the storefront its tier.

Per-courier load imbalance across a single flight.

A kitchen that does not balance courier load across the full flight ends up with one courier carrying three bags and two couriers waiting at the handoff — three bags that arrive late because they cannot be carried at once, two idle couriers who cost the marketplace rank on the storefronts they were assigned to. The marketplace assigns couriers per order, not per kitchen; the balancing logic has to live on the kitchen side, against a per-courier bag cap that respects each brand’s bag format and delivery radius. An operator who runs the flight by hand on three storefronts absorbs the imbalance into the weekly on-time report and reads it as a marketplace problem rather than a sequencing problem.

How Braiseflux ships it

The routing loop: read, score, sequence, approve.

Braiseflux runs one routing loop on four steps. The agent stays the sequencer; the operator stays the approver; the per-storefront tier stays unbroken. Together they retire the courier hold, the envelope mix-up and the load imbalance — without ever replacing the operator’s dispatch authority with an autonomous one. The day-one walkthrough on how it works covers the loop in the wider agent bundle.

Read

Read the kitchen flight across all active storefronts on every ticket.

The routing agent reads three signals that compose into one flight-level decision per active ticket — POS ticket timestamp (when the order landed), per-SKU prep-time estimate (how long the ticket takes to close), and courier ETA from the delivery marketplace (when the courier walks in). The read runs against the full ticket stream in real time, not per storefront and not per brand — a kitchen with three active storefronts and three concurrent courier ETAs has a single combinatorial flight the operator cannot sort by hand at speed. The agent keeps the flight queue current on each new ticket and each ETA refresh; the queue sits at the packer console, not in the dispatcher’s head.

Score

Score each ticket against courier-arrival, per-courier load, and bag envelope.

The agent scores every active ticket against three axes that decide whether the ticket ships now or holds: the projected ticket-close minus courier-arrival gap (the hold-or-handoff decision), the per-courier bag count at the projected handoff window (so no one courier carries three while another waits), and the per-brand insulation profile (so a sushi bag never shares a courier run with a Nashville hot bag). The score lands per active handoff point, updated on every ticket arrival and every ETA refresh; the packer reads a ranked queue that is already pre-filtered on hold, load and envelope rather than a queue of every ticket that landed.

Sequence

Surface one ranked pack sequence per handoff point, first-out on top.

For every active handoff point the agent surfaces a single ranked queue — the first-out ticket at the top (the earliest-closing ticket against the soonest-arriving courier), the last ticket at the bottom, with the per-brand instructions for bag format, insulation envelope and handoff position interleaved in the same list. The packer works one sequence, not three — the cross-storefront flight is read by the agent, not by the dispatcher; the courier arrives at the handoff to a bag the packer has already sealed in courier-ETA order, not in order-of-arrival. The sequencing instruction updates in real time as courier ETAs refresh; the packer does not chase the queue, the queue updates the packer.

Approve

The operator stays the approver; the agent stays the sequencer.

Nothing ships without the operator’s eye on the five percent of flights that always need a human — a courier that has not arrived within the projected ETA window plus the hold tolerance, a per-courier bag count that breached the per-brand cap and would push the courier over its envelope, a bag-format swap a kitchen ran that the agent has not yet seen — and the operator either approves, edits, or escalates the rest. The ninety-five percent auto-ships against the flight; the five percent lands on the operator’s eye at the line-up scan. The operator stays the author of dispatch and the dispatcher-in-chief; the agent stays the radar that does not pad the operator into a flight they cannot hold by hand.

The five signals the agent reads

What the routing agent listens to on every flight refresh.

The routing agent reads from the five signal surfaces that funnel a ticketed order into a sealed, ready-to-handoff bag — across the inputs an operator would have to chase by hand on three storefronts, one POS and three marketplace courier feeds. Each signal is a pre-condition for an outbound hold-or-handoff decision; the agent surfaces a sequence only when the signal stack crosses the per-storefront threshold.

POS

POS ticket timestamps per storefront.

The cloud POS — Square, Toast, Otter — surfaces a ticket timestamp per order per storefront, keyed to the moment the kitchen receives the ticket rather than the moment the marketplace confirms the order. A ticket that landed at 6:04 p.m. against a courier ETA of 6:32 p.m. and a projected prep close of 6:30 p.m. is a 2-minute hold-zone signal; the routing agent reads the landing time on every ticket, joins it against the prep estimate, and ranks the ticket against the courier’s projected arrival rather than against the ticket’s place in the per-storefront queue. The agent never invents a timestamp; it reads the POS arrival the operator already staffs against.

Prep

Per-SKU prep-time estimates.

A sushi roll closes in six minutes from the moment the POS ticks; a Nashville hot chicken sandwich closes in nine; a shawarma platter closes in seven. The per-SKU prep-time estimate feeds the projected ticket-close side of the courier-arrival gap; the agent reads the per-SKU rate from the same prep log the kitchen staffs against and surfaces the estimate into the flight queue. A kitchen that retunes its per-SKU prep estimates from the actual close times the kitchen records drops the projected-close error inside two weeks; the flight queue stays aligned with the real kitchen cadence rather than the prep chart on the wall.

ETA

Courier ETA from the delivery marketplace.

The three marketplaces surface a courier ETA per active assignment, refreshed every ninety seconds on each courier’s position relative to the kitchen. The agent reads the ETA in parallel and weights the flight score against the projected arrival delta: a courier whose ETA moved later by four minutes pushes its ticket down the flight; a courier whose ETA moved earlier pulls its ticket up. The agent never holds a courier; the marketplace holds the courier via its ETA refresh. The agent only sorts the queue the courier finds when it walks in.

Load

Per-courier bag count at the projected handoff window.

A courier already carrying two bags from the same kitchen is a courier the agent deprioritizes for the next assignment; a courier with an open ETA window and zero current bags is a courier the agent surfaces to the packer first. The per-courier bag cap is per brand, not per kitchen: a sushi brand with a small-format bag and a tight delivery radius carries a different cap than a Nashville hot brand with a large-format bag and a farther radius. The load dashboard lives at the couriers console; the per-courier assignment queue updates in real time as courier ETAs refresh, and the cap reads against the projected handoff window, not the kitchen’s working memory.

Envelope

Per-brand insulation profile and bag envelope.

A sushi bag, a shawarma bag and a Nashville hot bag carry three temperature profiles and three insulation envelopes that do not share a courier run without the temperature integrity of at least one failing. The agent reads the per-SKU envelope profile and separates the flight assignments so the bags that share a courier run share a compatible envelope. The sequencing instruction lands at the packer console before the courier arrives, not after — the packer reads the envelope instruction inline with the flight rank rather than in a separate ticket-order pass. The envelope profile is a per-brand operator-set policy the agent never overrides.

What you get on day one

The hold, rating and tier outcomes an operator reads once the routing loop holds.

The promise of courier routing is not removing the operator from dispatch; it is removing the seams so the operator reads the entire flight in one scan and approves what ships. Concretely, here is what every Braiseflux operator gets the week the routing loop lands on the line.

  • Courier hold time drops below the marketplace’s on-time threshold. The flight-level sequencer reads the courier-arrival minus projected-ticket-close gap for every active ticket and ranks the handoff against the marketplace’s on-time window; the courier who arrives while a bag is still on the line is held in the queue, not in the kitchen. Operators in our cohort who hold the per-flight cadence read courier hold time stabilize below the marketplace’s documented on-time threshold inside a fortnight, and the rating-tier move compounds week over week.
  • One-star reviews mentioning cold food drop with the envelope profile. The agent reads the per-brand insulation profile inline with the flight rank and separates the courier runs that share a bag envelope; a sushi bag and a Nashville hot bag no longer share a courier run by accident, so the temperature integrity complaint stops compounding on the storefront that would have carried it. Operators who hold the per-brand envelope profile see the cold-food one-star rate drop inside a month.
  • Per-courier load balances against the per-brand cap. A naive sequencer that ranks tickets by courier-ETA alone ends up with one courier carrying three bags and two idle at the handoff. The flight-level loop reads the per-courier bag count at the projected handoff window and deprioritizes the courier already at cap; the kitchen ships ten bags on time and three late rather than three late and ten on time, and the marketplace ranking algorithm reads the difference.
  • The pack sequence aligns with the SKU the inventory loop confirmed is on the shelf. The flight-level loop pairs with the inventory reorder cadence on the same agent bundle — a courier that arrives for a sushi bag with the avocado the inventory loop reordered against Tuesday’s projected consumption ships the bag the reorder confirmed. The two reads stay aligned; the operator stops reconciling the pack sequence against the shelf by hand.
  • The operator stays the dispatch authority, not the dispatcher-in-chief. The agent surfaces only the sequences that pass the per-storefront hold window and the per-courier load cap; the operator stays the author of the per-brand envelope profile, the hold tolerance and the courier cap. Nothing ships without the operator’s eye on the five percent of flights that always need a human — a courier that has slipped its ETA past the hold tolerance, a per-courier bag-cap trip, a bag-format swap the agent has not yet seen.

Keep reading

Where to go next on the routing loop.

The routing loop lives inside the broader multi-brand agent bundle, and it sits next to a handful of operator field notes and sibling solution pages that walk the deeper questions operators ask before signing up. Start with the cluster field note on the three-signal flight this page describes, then move into the storefront-side sibling, the procurement-side sibling, the parent multi-brand overview, and the pricing that fits the operation.

Get started

Ship the routing cadence on your first dinner rush, this week.

Connect Braiseflux to your POS and your three storefronts, name the five input signals, and watch the routing agent draft the per-handoff pack sequence before the courier arrives. The operator stays the dispatcher-in-chief; the kitchen holds one cadence per storefront; the flight queue stops holding couriers at the door. Come back via the landing page to see the four-agent bundle the routing loop lives inside.

Open the packer queue →See the courier fleet

Want the per-flight math? Read the deep dive →