Research, Conversation Design and UX Writing
Simplifying the B2B support chatbot for FedEx
What we did
- Benchmarking
- AI agents
- Intent restructuring
- Meta guidelines
- UX audit
- Conversation design
- UX writing



The problem
Solve it in the chat. Only send people elsewhere when the bot really can’t.
That was FedEx’s brief. Their B2B bot was live on WhatsApp, but customers struggled to use it, and support tickets and emails kept coming in.
-
FedEx had launched a WhatsApp bot for its business customers, but most people couldn’t find what they needed in it. Queries still ended up as tickets and emails. We were brought in to make the bot easy to understand and resolve those queries in the chat.
-
We led the benchmarking, and I set up the AI agent that did the heavy lifting on research. From there, I drove the IA, the user audit, the new flows and the copy.
-
We restructured the IA and redesigned tracking, pickups and delivery changes, so each one ends with a confirmation inside WhatsApp.

-
- 42% fewer support tickets and emails
- 68% of conversations resolved in the chat
- Pickups booked 60% faster
Where we started
FedEx had a support bot live on WhatsApp for its business customers, and it wasn’t taking any load off the support team. People got lost in it and raised tickets or sent emails instead. Our job was to bring those down by resolving as much as possible inside the chat.
We started by mapping the current information architecture. It went four levels deep, repeated the same tasks in different places and sent people back through menus.
We redesigned it around the three things customers come to do. Nothing sits more than three levels deep, and every task lives in one place.
The approach
We studied the market, understood the intents the bot already had, learned what WhatsApp allows and listened to the people using it. The audit told us to fix the intents one at a time. The language came last.
Market research
We benchmarked FedEx against the four strongest support bots on WhatsApp in India, on 10 parameters.
- 01Entry and discoverability
- 02Functional scope and depth
- 03Intent architecture and navigation
- 04Tone of voice and personality
- 05Exception handling and fallback
- 06Channel transition
- 07In-chat UI and message design
- 08Resolution quality and closure
- 09Human agent escalation
- 10Proactive communication
Setting up the research agent
I set up an AI agent to do the heavy lifting. We gave it a clear brief, the sources to read and a fixed format for what it sent back, so every output could be checked.
- 01Briefed it with our 10 parameters and FedEx’s goal of fewer tickets
- 02Asked for scored tables and grouped lists, never open summaries
- 03Checked a sample of every output by hand before using it
Understanding the current intents
The agent mapped every intent in the current bot and grouped them by what customers came to do. We validated each group with real tasks, and this became the redesigned IA.
- Duties, taxes and customs
- Invoices, refunds and claims
- FAQs into one short list
- Three back to menu steps
- Tracking repeated in FAQs
- Delivery changes repeated in FAQs
- Tracking as one status card
- Pickups with a single form
- Manage Delivery with six options
Understanding Meta guidelines
FedEx wasn’t sure what WhatsApp allows. We studied Meta’s guidelines ourselves, and the agent turned every limit into a quick screen the team could design against.
Interviews and audit
We interviewed 24 of FedEx’s business customers, grouped them into 8 archetypes and had each of them complete 12 tasks on the current bot. AI helped us synthesise the sessions into a list of issues for every archetype.
The pattern was clear. Tracking, pickups and delivery changes caused most of the trouble, so we took them on one intent at a time.
Improving the verbiage
Last, we rewrote every message. One idea per message, key details in bold and a clear next step at the end.
Thank you for your input. Your details have been successfully verified. Please confirm if you’d like us to schedule pickup from this location: DLF EXCLUSIVE FLOORS. If you wish to schedule a pickup from a different address, kindly contact Customer Care team to update the pickup location.
📦 Choose when and where you would like to schedule your pickup.
Solving the primary intents
Tracking, pickups and delivery changes made up most of the bot’s traffic and most of the tickets. We redesigned each one end to end, so it closes inside WhatsApp with a clear confirmation.
Track a package
Intent 01The old reply put everything into one paragraph. We split it into a status card customers can scan, and moved package details one tap away.
Shipment status for AWB 871987641234: Your delivery has been rescheduled. Estimated Delivery: Monday 9:30 AM to 12:30 PM. Current Status: Bethpage, NY - Fedex facility, Alternate delivery requested. Origin: Greer, SC. Destination: New Hyde Park, NY. Last Scanned Location: DUBAI WORLD CENTRAL (DWC) AE 11:01 AM. Packaging Type: Package. Total Weight: 1.5 kg (3.3 lb). Dimensions: 17 x 17 x 15 cm. Package Count: 1. Note: Your package may be late due to local delivery restrictions. Last Updated: 4 April 2026 4:42 AM.
10:02What went wrong
- 01Status, package details and the note in one block
- 02Nothing stood out, so the delay got missed
- 03The only way forward was back to the main menu


Schedule a pickup
Intent 02The old flow asked for details one message at a time and sent address changes to customer care. The new one collects everything in a single WhatsApp form and confirms with the AWB and pickup reference.

What went wrong
- 01Two menus before the task even started
- 02AWB and postal code asked for one by one
- 03Long notes that most people skipped
- 04A different address meant calling customer care



Manage a delivery
Intent 03Customers can change a delivery in six ways, and each one has its own flow. Before, every option opened with long paragraphs. Now each one is a few short steps that end in the same confirmation.
- 01Long, legal-sounding paragraphs before every choice
- 02The same warning repeated across messages
- 03A plain line of text as the only confirmation






Impact
42% fewer support tickets and emails
Tracking, pickup and delivery queries now close inside WhatsApp instead of reaching the support desk.
68% of conversations resolved in the chat
Up from about a third before the redesign. Customers leave with a confirmation, not a phone number.
Pickups booked 60% faster
One WhatsApp form replaced a string of back and forth messages.
35% fewer drop-offs mid-conversation
A simpler IA, with fewer choices at every step, kept people moving to the end of their task.
Get in touch
Want the full walkthrough?
There’s more to this project than one page can hold. Leave me a message and I’ll take you through the research, the agent setup and every flow we redesigned.
shivamar275@gmail.com
LinkedIn