AI Agents
Discovery Express

Retail & CPG

Order status and returns eat the contact center.

Retail produces millions of small decisions a day: where is my order, can I return this, why did the delivery fail. Agents connected to the OMS, the WMS and WhatsApp answer most of them and hand the rest to a person with full context.

Pilot in production: 8 to 12 weeks, under a KPI Contract

A courier handing a grocery order to a customer at her apartment door in Mexico City at dusk

Agent Library

Three agents we recom­mend

Reference architectures show typical targets, never a result.

Typical use cases

01Order status and returns over WhatsApp connected to the OMS
02Last-mile delivery coordination and exception handling
Order 48213Illustrative
Customer Support Agent
  1. Customer

    Order 48213 was supposed to arrive yesterday and nothing came.

  2. OMSOrder 48213 read: delivery failed 16:42, carrier event: address not found.

  3. Agent

    Sorry, Ana. The driver could not find the address. Confirm the street and a reference and I reschedule for tomorrow, 10 am to 1 pm.

  4. Customer

    Calle 60 #412, next to the pharmacy. I also want to return one of the two items.

  5. OMSRe-delivery booked, tomorrow 10:00 to 13:00. Return 48213-2 authorized under the 30-day policy, label generated.

  6. Handoff to a person

    The refund amount does not match what Ana paid, so it goes to Rocío in billing with the order and this conversation attached.

Illustrative thread built from the industry's typical flows, not a client conversation. DaCodes, Sep 2026.

Customer ops

Customer Support Agent

Tier 1 support on WhatsApp, voice, chat and email, inside your CRM. Measured on cost per ticket.

KPI it moves

$6.40 to $0.85 per ticket

In production

Operations

Fleet Check-Call Agent

Voice check calls to drivers, logged in the TMS with GPS context, at $0.035 per call on the DaCodes voice stack.

KPI it moves

80 to 90% of check calls automated

Reference architectureTypical target

People & talent

Voice Interviewer

Structured voice interviews at scale, scored and written to your ATS. Packaged as Reclutalia.

KPI it moves

1,000 calls per month to 1,000 per hour

In production
KPI sources
  • Customer Support Agent, Mexican fintech, 90-day pilot, Proof Ledger Sep 2026.
  • Fleet Check-Call Agent, Typical target for a reference architecture, not a result. DaCodes, Sep 2026.
  • Voice Interviewer, AI Interviewer case, Proof Ledger Sep 2026.
  • Maturity labels reviewed September 3, 2026.

Proof

The case behind the recom­mendation

Every figure sits in the Proof Ledger. When there is none yet, we say so.

SorianaPlatforms

90.9%

successful deliveries

from 3.4% with manual dispatch

Retail & CPG · México

Successful deliveries from 3.4% to 90.9%

Verified by Proof Ledger, Sep 2026

The Soriana figure comes from the last-mile platform DaCodes built, not from these three agents; the agents are the next step on that platform.

Where the agent lives

Inside the systems you already run

Your system of record stays the truth. Nothing lives outside your cloud.

Channel order follows each market, WhatsApp first in México. DaCodes, Sep 2026.

Channels

  • Voice
  • Email
  • Web chat
  • WhatsApp
Customer Support Agent

Systems of record

  • POS
  • OMS / e-commerce
  • WMS
  • Salesforce
  • HubSpot
  • Last-mile carriers

Fleet Check-Call Agent

The last mile, by voice

The agent calls the driver, hears the delay and moves the delivery window before the customer has to ask.

Live call
The agent asks the driver for status on the next delivery of the route.
TMS
The new ETA is written to the TMS, not to a side spreadsheet.
WhatsApp
Customer and dispatcher hear about the change at once.

Demo data, not a client operation. DaCodes, Sep 2026.

Check call · Unit 42
Demo data
Live call · 00:41

  • Good morning, this is the dispatch assistant for Route 18. Are you on schedule for the 14:00 delivery?

  • Running late, about twenty five minutes. Traffic on the highway.

  • Understood. I am moving the delivery window to 14:25 and notifying the customer. Anything else to report?

  • No, all good.

ETA updated in TMS

14:00 → 14:25 · customer notified on WhatsApp · dispatcher alerted

Own voice stack · about $0.035 per 15-min call, internal measurement, Aug 2026

TMS

Regulation

Regulation we design for

  • PROFECO consumer protection (México)
  • LFPDPPP data protection (México)
  • CFDI 4.0 for invoicing
1

Approvals

Sensitive actions stop at a person you name.

2

Audit trail

Every read, write and escalation logged with prompt version, model and approver.

3

Data residency

Your AWS account, your region. PII masked before any model call.

Design inputs, not legal advice. Information security certified under ISO/IEC 27001:2022, December 2025.

FAQ

Questions from this industry

No. It reads order status, return policy and delivery events, and writes the return or the re-delivery back into the OMS with scoped permissions. Your systems stay the truth; the agent is the conversation on top of them.

Peaks are the reason to have the agent. The runtime scales horizontally, the daily cost ceiling is set per campaign in the Agent Console, and the pilot KPI is measured over the whole period, peaks included, never on a quiet week.

Because store and warehouse hiring is high volume and seasonal. In the AI Interviewer case a team went from 1,000 screening calls a month to 1,000 an hour (Proof Ledger, Sep 2026); your recruiters still make every decision.

Other industries

All industries

Which process do we automate first?

Two discovery sessions, no cost, and you leave with two or three quick wins whether or not you build with us.

We reply in under one business day.

Loading the form