What problem does it solve?
"Where is my parcel" questions are a large share of what a carrier's contact centre handles: Chronopost says routine enquiries such as parcel tracking and delivery times make up 40 per cent of its incoming contacts. Many of those contacts come from someone who is not the carrier's customer: the recipient did not choose the carrier, often does not have an account and only has a tracking number. Volumes follow ecommerce peaks (Evri calls the weeks from Black Friday to just before Christmas its busiest period of the year), so contact centres are busiest exactly when the network is under the most pressure.
The answer is usually already in the tracking data, but tracking events are written for the network, not for people ("exception, consignee not available", "held at depot"). What the recipient wants is an action: deliver it tomorrow, leave it with a neighbour, send it to a pickup point, pay the customs duty, or open a claim because the parcel is damaged or never arrived. A first generation chatbot could show the tracking page; an agent connected to the delivery management system can change the delivery, and knows when a claim or a complaint needs a person.
This page covers the carrier's side: the recipient contacting the parcel company. The retailer's side, where the shopper asks the web shop about an order, a delivery change or a return, is covered by the order status and returns agent.
- Chronopost reports that routine enquiries such as parcel tracking and delivery times represent 40 per cent of its incoming contacts.CX Award Innovation: Chronopost Wins Gold for AI Agent (2026)
- Evri describes the period from Black Friday in late November to just before Christmas Day as its busiest period of the year for parcels.Evri dials up new automated phone line to quickly connect customers to UK advisors, as part of a £46m total investment (2023)
How does it work?
- Identify the parcel. The agent takes a tracking number, a phone number or an order reference, and on messaging channels it can start from the notification the recipient already received.
- Translate the tracking. It reads the scan events and the network status (depot, route, weather or peak delays) and explains in plain language where the parcel is and what happens next, including a realistic delivery window.
- Offer the delivery options. Within the carrier's rules and the shipper's settings, it offers redelivery on another day, a safe place or neighbour, a pickup point or locker, or a change of address, and confirms the change in the delivery management system.
- Handle exceptions. For customs holds it explains what is needed and links to payment of duties; for damaged, late or missing parcels it collects the details and photos and opens a claim or an investigation with the right team.
- Hand over well. Claims above set values, suspected fraud or theft, complaints and repeated failures go to a human agent with the tracking history and what the agent already tried.
- Audience
- Customer facing
- Autonomy
- Supervised agent
- Adoption
- Early adopters
- Channels
- Web chat, Phone and voice, WhatsApp, Mobile app, SMS
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Containment rate | Too few to pool | 50% to 85% | 2 | 1 organization, 1 vendor |
| Contact deflection | Too few to pool | 50% | 1 | 1 organization |
| Interactions handled | Not pooled | 432,000 | 1 | 1 organization |
| Response time reduction | Too few to pool | 40% | 1 | 1 organization |
Value drivers: Lower cost to serve, Customer experience, Speed and cycle time, Inclusion and access.
Indicative value
A parcel carrier delivering 200 million parcels a year
EUR 800,000 to EUR 12 million
Human handled contact cost avoided per year
How this is calculated
Formula: parcels * contactRate * trackingShare * containment * costPerContact. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Parcels delivered per year parcels, parcels per year | 200,000,000 | 200,000,000 | The reference carrier. |
| Assisted customer contacts per parcel contactRate, contacts per parcel | 0.01 | 0.03 | Editorial assumption; replace with your own contact rate. |
| Share of contacts about tracking, delivery times and delivery changes trackingShare, fraction of contacts | 0.4 | 0.5 | The low end matches Chronopost, which reports that routine enquiries such as parcel tracking and delivery times make up 40 per cent of incoming contacts. The high end, which adds delivery change requests, is an editorial assumption; replace with your own contact reasons. |
| Share of those contacts the agent resolves containment, fraction of contacts | 0.5 | 0.8 | Conservative against the benchmark on this page (Chronopost reports 85 per cent of simple requests resolved end to end). |
| Cost of a human handled contact costPerContact, EUR per contact | 2 | 5 | Editorial assumption for a blended phone and chat contact. Replace with your own fully loaded cost. |
What it leaves out: Gross avoided contact cost only. It leaves out the cost of running the AI and the integrations, fewer failed deliveries from redelivery and redirect options, lower claim handling cost and the effect on shippers who choose carriers partly on recipient experience.
Who already uses it?
7 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Chronopost
France · Logistics and transportation · 2026
Chronopost, a French parcel carrier and business unit of the Geopost group, runs Léonard, an autonomous generative AI agent developed with its internal Robot Factory. It handles routine enquiries such as parcel tracking and delivery times, which together make up 40 per cent of incoming contacts, takes actions directly in operational systems and pre qualifies complex cases for customer advisors. It won a Gold CX Award for innovation in June 2026, when Geopost said the rollout would extend across all services from July 2026.
- Containment rate: 85%, simple requests
"Chronopost achieves 40 per cent faster response times and resolves 85 per cent of simple requests end to end with its autonomous agent, Léonard."
Claimed by: organization - Response time reduction: 40%
"Chronopost achieves 40 per cent faster response times and resolves 85 per cent of simple requests end to end with its autonomous agent, Léonard."
Claimed by: organization
PostNL
Netherlands · Logistics and transportation · 2025
PostNL, the Dutch postal and parcel company, reports in its 2025 annual report that it renewed its customer service chatbot and introduced conversational AI to support service agents, to give faster and more consistent answers to common questions such as "where is my parcel?", and that it migrated to a new integrated conversational platform. The report names the chatbot Daan, available around the clock, and states PostNL's ambition to become an AI first organisation.
No outcome disclosed.
DPD Belgium
Belgium · Logistics and transportation · 2020
DPD Belgium (DPD Belux) launched Phil in April 2020, a chatbot on dpd.be and its mobile site that lets consignees ask where their parcel is and get instructions to redirect it, with a link to the service team when it cannot answer. In March 2022 DPD Belgium replaced Phil with Ruby, a self learning chatbot available by typing, clicking a suggestion or speech, that answers questions such as "My parcel has not been delivered" or "When will my parcel be delivered?", helps change a delivery address or time by handing the consignee to the myDPD app, and helps find a pickup point from the consignee's location.
No outcome disclosed.
DPD Deutschland
Germany · Logistics and transportation · 2019
DPD Germany launched Red in October 2019, an AI controlled chatbot on dpd.de that lets parcel recipients check the status of a shipment on desktop and mobile, with a link to the service team when it cannot answer. Red later added redirection options and a deposit permission (leave the parcel without a signature), and from 2021 it also lets shippers and recipients book a parcel shipment. DPD Belgium runs a similar consignee chatbot, launched as Phil in 2020 and replaced by the self learning Ruby in 2022.
No outcome disclosed.
Evri
United Kingdom · Logistics and transportation · 2018
Hermes UK, now Evri, launched an AI chatbot called Holly in November 2018 to automate parcel tracking, signature questions, diversions and confirmation of delivery for recipients. In 2023 it added a voice assistant on a dedicated phone line that identifies the caller's active parcels from their phone number and routes them, with an automatic callback within 24 hours; the release says the phone line complements a recently introduced chat service. Evri's November 2024 release credits a new chatbot and the callback feature with a marked increase in consumer satisfaction. The sources do not say whether Holly itself still runs.
- Interactions handled: 432,000, from the November 2018 launch to October 2019
"Since the initial Chatbot launch in November, Holly has had a total of 432,000 chats resulting in a reduction of contacts to customer service operatives by 50%."
Claimed by: organization - Contact deflection: 50%, contacts to customer service operatives, November 2018 to October 2019
"Since the initial Chatbot launch in November, Holly has had a total of 432,000 chats resulting in a reduction of contacts to customer service operatives by 50%."
Claimed by: organization
DPD UK
United Kingdom · Logistics and transportation · 2024
DPD UK had used an AI element in its online customer service chat for several years. In January 2024, after a system update, a customer trying to trace a parcel got the chatbot to swear, call itself useless and criticise the company, and the screenshots spread widely on social media. DPD said an error occurred after the update and that the AI element was immediately disabled and being updated.
No outcome disclosed.
Aramex
United Arab Emirates · Logistics and transportation · 2022
Aramex launched Aramex Bot, a WhatsApp chatbot built by MENA conversational AI provider arabot, giving customers real time shipment and courier tracking by typing a reference or shipment number. Besides tracking, the bot gives delivery time information, lets customers arrange a shipment return, and finds the nearest Aramex branch, with a hand off to a live agent for complex queries. Since the bot's launch in October 2018, arabot reports that half of shipment inquiries have been processed by the chatbot and more than 8 million customers have been served.
- Containment rate: 50%
"Since the launch of Aramex's WhatsApp Business Bot by arabot in October 2018, 50 percent of shipment inquiries have been processed by the chatbot solution and more than 8 million customers have been served."
Claimed by: vendor - Users served: at least 8 million
"Since the launch of Aramex's WhatsApp Business Bot by arabot in October 2018, 50 percent of shipment inquiries have been processed by the chatbot solution and more than 8 million customers have been served."
Claimed by: vendor
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Tracking events and exception codes mapped to plain language explanations and next steps
- Delivery options and restrictions per shipper and product (signature, age check, high value)
- Claim rules (liability limits, evidence needed, deadlines) with an owner
- Contact reasons by tracking status from the contact centre
Systems to integrate
- Track and trace and the delivery management system (redelivery, redirect, pickup point, safe place)
- Claims and investigations system
- Customs and duty payment for cross border parcels
- Notification platform (SMS, email, messaging, app)
- Contact centre platform for handover with the tracking history
Complexity: Medium
Tracking answers are straightforward once scan events are mapped to plain language. The work is in the actions (redelivery, redirect, safe place, claims) through the delivery management system, verification of the recipient, the shipper's restrictions and peak season load.
- 1
Start from the contact reasons
Take a year of contact reasons and split them by tracking status. Most "where is my parcel" contacts map to a handful of statuses; write the explanation and the allowed next actions for each.
- 2
Connect the actions, not just the tracking page
Integrate redelivery, change of delivery point, safe place and claim intake with the delivery management system, with the shipper's restrictions applied (for example no redirect for age restricted or high value goods).
- 3
Separate recipients from shippers
Recipients need tracking and delivery changes; business shippers need bulk status, pickups, invoices and claims. Give them different authentication and different allow lists.
- 4
Plan for peak and bad weather
Load test at peak season volume and prepare network wide messages (weather, strikes, depot backlogs) that the agent uses to answer honestly instead of repeating an estimate that has already slipped.
- 5
Guard the brand
Put input and output guardrails on every turn and test the agent against provocation and prompt injection, with a regression set that runs after every model or prompt change.
- 6
Test before customers do
Build test conversations per tracking status, per exception and per language, including claims the rules do not allow and attempts to redirect someone else's parcel, and run them on every change.
Guardrails
- Delivery changes only through an allow list of actions, respecting the shipper's restrictions and proof of delivery rules
- Verification (tracking number plus postcode or a one time code) before changing a delivery or showing the full address
- Claims above a set value, suspected theft or fraud, and complaints go to a human
- Input and output guardrails against abuse, prompt injection and off brand answers, tested after every change
- Personal data (names, addresses, phone numbers) masked in logs and model prompts
KPIs to instrument
- Containment per tracking status and per exception, counting repeat contacts within seven days as not contained
- Share of delivery changes completed by the agent and the delivery success rate after a change
- Handover rate and handover reasons, especially during peak weeks
- Customer satisfaction on contained conversations versus human handled ones
- Claims opened by the agent that were later rejected or reopened
Human in the loop
Humans handle claims above the agent's limit, investigations of missing parcels, suspected fraud and complaints. A team reviews a sample of contained conversations each week, watches handover reasons during peak, and approves every new action or change to claim rules.
Common failure modes
- The assistant says something that becomes the story
- A customer provokes the bot into swearing or criticising the company and the screenshots go viral, as happened to DPD in January 2024 after a system update. Keep guardrails on every turn and rerun adversarial tests after every update.
- Estimates that are not true
- The agent repeats a delivery estimate the network already missed. Use network status and exception events, and say when a date is uncertain.
- Changing the wrong parcel
- A redirect is made by someone who is not the recipient. Verify before any change and respect the shipper's restrictions.
- Containment that is really abandonment
- The recipient gives up and complains to the retailer or on social media. Measure repeat contacts across the carrier and the retailer where you can, and satisfaction per status.
What are the risks and rules?
EU AI Act
Limited risk (transparency)
Article 50(1): an assistant that talks directly with recipients must be designed so they know they are interacting with an AI system, unless that is obvious from the context. It is not high risk: explaining tracking, changing a delivery and taking in a parcel claim fall under none of the Annex III areas (it does not decide on public assistance benefits, creditworthiness, insurance pricing or employment), and it involves no practice prohibited by Article 5.
Guidance
- Article 50, transparency obligations for providers and deployers of certain AI systems (European Union, Europe). Recipients must be informed that they are interacting with an AI system unless this is obvious from the context.
- Parcel companies, complaints and where to get help (Ofcom, Europe). Ofcom rules require UK parcel delivery companies such as DPD and Evri to have a complaints procedure, and its complaints guidance (in effect from 1 April 2023) says complainants should be told the channels, the process and how long it takes, and that trained staff handle the complaint. The agent's complaint handover should follow that procedure.
Controls to put in place
- AI disclosure at the start of every conversation
- Verification before any delivery change, with an audit trail of who changed what
- Guardrail and adversarial regression tests rerun after every system, model or prompt update
- Claim and compensation limits per action, with human approval above them
- Monitoring of handover reasons and complaints during peak weeks
When it went wrong elsewhere
- Incident 631: Chatbot for DPD Malfunctioned and Swore at Customers and Criticized Its Own Company. In January 2024 a customer tracing a parcel got the DPD UK chatbot to swear and criticise the company after a system update; DPD disabled the AI element. Guardrails and regression tests must be rerun after every update, not only at launch.
Frequently asked questions
- How many parcel contacts can an AI agent resolve?
Chronopost reports that its agent Léonard resolves 85 per cent of simple requests end to end with 40 per cent faster responses, and that routine enquiries such as parcel tracking and delivery times make up 40 per cent of incoming contacts. Hermes UK (now Evri) reported 432,000 chats with its chatbot Holly between its November 2018 launch and October 2019, and a 50 per cent reduction in contacts to customer service operatives.
- What went wrong with the DPD chatbot?
In January 2024, after a system update, a customer got the DPD UK chatbot to swear and criticise the company, and DPD disabled the AI element. The lesson is to keep output guardrails on every turn and to rerun adversarial tests after every update.
- How is this different from an order status agent?
The carrier's agent serves recipients who often have no account and only a tracking number, and it changes the delivery itself. A retailer's order status agent serves its own shoppers and also handles cancellations, returns and refunds.
How to cite this page
Blits.ai AI Use Case Library, "AI agent for parcel tracking and delivery exceptions", last verified 27 September 2026, https://www.blits.ai/ai-use-cases/parcel-tracking-and-delivery-exception-agent. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 30 September 2026: Added deployments: Aramex.
- 29 September 2026: Added deployments: DPD Belgium.
- 27 September 2026: First published