> ## Documentation Index > Fetch the complete documentation index at: https://docs.apollospace.ai/llms.txt > Use this file to discover all available pages before exploring further. # Logistics & transportation > Freight, last-mile, and 3PL teams drowning in "where is my order?" and carrier chasing — with the sensitive line always staying human. It's 7am and the truck is late. Before you even open the system, three customers have already messaged on WhatsApp asking "where's my shipment?", a carrier never answered yesterday's request for an update, and a delivery that should leave today is stuck waiting on a proof of pickup from last week. None of it is hard — it's just a lot, all at once, and every minute without an answer is a customer calling again. **What leaves your plate:** answering where an order is, chasing the carrier, triaging the incident to the right owner, collecting the shipping documents, and flagging a shipment before it misses its window all start happening on their own — with context, within budget, and with the sensitive part always waiting on a human. ## How it works, in one line Logistics is, at bottom, **a status problem that changes constantly, plus people chasing people.** The agent doesn't drive the truck — it makes sure the right answer exists at the right moment, that nobody has to remember to chase the carrier, and that a problem reaches its owner before the customer complains. ```mermaid theme={null} flowchart LR A[Trigger
customer asks, shipment delays,
document missing] --> B[Agent pulls context
from Company Brain + CRM + order] B --> C[Decides + uses tools
Chats, WhatsApp, CRM, Routines, Docs] C --> D{Sensitive?
money, SLA,
contract} D -->|No| E[Answers / chases / logs / alerts] D -->|Yes| F[Proposes + calls a human
with full context] E --> G[Becomes memory
the next delivery already knows] F --> G ``` ## Concrete flows ### 1. "Where's my order?" — answered on WhatsApp from the record **Trigger:** a status question arrives on [WhatsApp](/en/integrations/whatsapp) or [chat](/en/features/chats) — "does my shipment leave today?", "what's the tracking code?". The agent identifies which order the person means, pulls the record from the [CRM](/en/features/crm) and the shipment details from the [Company Brain](/en/features/brain) — carrier, ETA, last update, manifest — and answers right away, in plain language: where it is, when it arrives, the tracking number. The interaction is logged as a [CRM activity](/en/features/crm), so the next person to open the card sees the customer has already been answered. When the data **doesn't exist** or is stale — the last position is two days old — the agent doesn't invent an "it's on the way." It flags the information as outdated and kicks off the chase flow (below) instead of pushing an empty answer. The win isn't "a bot that replies." It's that **the customer gets the position immediately, on their channel**, without anyone stopping the operation to open three systems — and the answer comes from the real record, not a guess. ### 2. Chase the carrier — and log what comes back **Trigger:** an order has gone longer without an update than is acceptable, or a customer asked for a position you don't have yet. The agent assembles the chase with what matters — order number, invoice, last known status — and sends it to the carrier contact by [email or WhatsApp](/en/features/outbound), with [Marcus](/en/agents/marcus) handling the send. You decide whether to require **human approval before each send** or to let routine chasing run. When the carrier replies, the agent reads the update, writes the new position into the [order record](/en/features/crm), and if it changes the customer's ETA, raises a [notification](/en/features/rotinas) in the bell of whoever owns the account. ```mermaid theme={null} flowchart TD A[Order with no update] --> B[Agent assembles the chase
from the order record] B --> C[Marcus sends
email / WhatsApp to carrier] C --> D{Replied?} D -->|Yes| E[Logs new position in CRM] D -->|Not in time| F[Re-chases + alerts the account owner] E --> G[Becomes memory] F --> G ``` The win: **the chase always goes out, on time, without anyone remembering** — and every carrier reply becomes a searchable record, not an email lost in one person's inbox. ### 3. Triage an incident — to the right owner, with a proactive alert **Trigger:** a delivery goes wrong — damage, loss, a refused receipt, a wrong address. The agent classifies the incident from the record and the [Company Brain](/en/features/brain) (what each kind of occurrence requires), opens a [task](/en/features/boards) on the incidents board already carrying the context — order, customer, carrier, what happened — and **assigns it to the right owner** for that kind of case. In the same step, it raises a [proactive notification](/en/features/rotinas) so the owner knows **before the customer chases**, not after. **Why this matters:** the cost of an incident isn't only the freight — it's the time until someone notices it exists. Triaging it on the spot, with the case already in the right person's lap, cuts that time from hours to seconds. Anything that needs a decision — a refund, opening a claim, changing a promised deadline — keeps waiting on a human (see below). ### 4. Collect and file the shipping documents **Trigger:** a pickup or delivery happens and generates paperwork — proof of delivery, invoice, receipt, manifest. The agent gathers the documents that arrive by [chat](/en/features/chats), WhatsApp, or email, attaches each to the right [order](/en/features/crm), and organizes them into [folders](/en/features/pastas) by customer, route, or period — so that when someone needs the proof of delivery from three weeks ago, it's one search away, not buried in a WhatsApp group. What's filed also feeds the [Company Brain](/en/features/brain): the next question about that shipment already has somewhere to look. ### 5. The routine that flags before the window slips **Trigger:** a [routine](/en/features/rotinas) the agent schedules for itself — hourly, or every morning — to sweep the open shipments. On each run, the agent compares each shipment's ETA against its promised window and **separates the ones at risk** of running late. For each, it raises a [notification](/en/features/rotinas) in the bell, suggests the next action (chase the carrier, warn the customer, reschedule) and, if you've allowed it, fires the chase from flow 2. The result is to stop discovering a delay through the customer's complaint and start knowing about it **while there's still time to act.** A routine never has a hectic Monday. It runs at the set time, every time, and turns "we only noticed it was late when the customer called" into "the system flagged it at 7am and we were already chasing." ## What stays human The agent **proposes**, never decides alone, when the matter is: * **Money** — a refund, compensation, a claim payout, a discount, any financial adjustment with a customer or carrier. * **An SLA change** — any commitment that alters the deadline or condition promised to a customer. * **A commercial proposal or term** — freight pricing, a carrier contract, any commitment that becomes an obligation. * **Heavy external, legal, or regulatory communication** — formally opening a claim, a notice, a response to a regulator. * **Governance decisions and destructive actions.** In those cases the agent prepares everything — the draft, the context, the recommendation — and **waits for a person's approval.** Every action stays in the auditable trail, with author and rationale. ## Where to start Upload to the [Company Brain](/en/features/brain) your incident process, your carrier contacts, and what each kind of incident requires. That's where the agent pulls the answer and the classification from — without it, there's nothing to consult. Hook up [WhatsApp](/en/integrations/whatsapp) and let the agent answer "where's my order?" from the record, escalating what it doesn't know. Start with human approval before each external reply and loosen as it proves itself. Create a [routine](/en/features/rotinas) that sweeps open shipments and flags the ones at risk. One agent, one cadence — and autonomy grows like a ratchet, not a leap. ## Next steps The general operations pattern — triage, follow-up, and SLA — of which logistics is a high-volume case. The cron of the operation — the window sweep and the chase that run on their own. Where the order, the customer, and every update stay live and searchable. The channel where "where's my order?" arrives — and the answer comes from the record.