Retailers, marketplaces, and D2C brands

AI worker teams for order and delivery support

Retail support demand is unusually predictable and unusually low-value. The win is containing the predictable so people are free for the exceptions — and the real product decision is where you set the thresholds.

RETAILERS, MARKETPLACES, AND D2C BRANDSCONSTRAINT ENVELOPE4 sector constraints designed to firstHUMAN AUTHORITY — NEVER AUTOMATED· Goodwill beyond threshold· High-value and loyal-customer…· Fraud and chargeback decisionsACTS WITH APPROVALThresholds and gates on anything with cost or risk attachedACTS WITHIN POLICY4 blueprints start hereAutonomy widens outward only on evidence from the recordearned autonomy
The authority boundary we design to in Retail & ecommerce — autonomy widens outward from the centre, and only on evidence.

Our view

Where we think the real opportunity is

Most retail support conversations are a database lookup wearing a costume. Where is my order, can I change the address, how do I return this — the answers exist in systems the customer cannot see, and the agent is a slow, expensive query interface. That makes containment the obvious play, but it also makes containment a misleading metric on its own: a contact closed without resolution comes back, and closing it twice costs more than answering it properly once. So we instrument repeat-contact rate alongside containment and treat divergence between them as the signal that thresholds are set wrong. The genuinely strategic decision in retail is not whether to automate — it is how much money a worker may give away before a person is involved, and that is a commercial choice we will not make for you.

Operating reality

How demand actually behaves here

Volume has a shape, and the shape is what breaks teams. This is what we assume about your operation before we design anything.

  • Seasonality is a multiple, not a bump

    Peak volume can be several times baseline, arriving in a window too short to hire and train for, which guarantees that quality degrades exactly when acquisition spend is highest.

  • Customers hop channels mid-issue

    A question starts in chat, continues on WhatsApp, and finishes by email. Without shared context the customer supplies their order number three times and concludes you are disorganised.

  • Most exceptions are caused outside your walls

    Carrier delays, failed delivery attempts, and damage happen in someone else's operation, but the contact and the blame land with you.

  • Service failures are publicly visible

    Reviews and social posts mean a mishandled recovery has a reach that a mishandled call in most sectors does not.

Where the hours go

The cost we would try to move first

Almost every operation spends its most expensive time on its least valuable work. Naming that precisely is what makes a pilot measurable rather than impressive.

  • The seasonal hiring and training cycle

    Recruiting, onboarding, and quality-managing temporary agents every peak is a recurring fixed cost that produces the least experienced team at the busiest moment.

  • Order-status volume

    The single largest contact reason in most retail operations is also the one with the clearest answer available from live systems.

  • Ungoverned goodwill

    Refunds and credits issued inconsistently to defuse frustration are real margin, spent without a policy and usually without a record of why.

  • Repeat contacts on one issue

    Every second and third touch on the same problem is cost created by the first touch not resolving it, and it is the most under-measured number in retail support.

The trust envelope

What constrains the design, and what never gets automated

We design to the boundary first. A control added after launch is a control nobody believes — and in this sector the boundary is not negotiable.

Sector constraints

  • Identity before order disclosure

    Order contents and delivery addresses are personal data, and an order-lookup that skips verification is a data-exposure route.

  • Payment data stays out of the conversation

    Refunds and payment changes must route through a compliant path rather than card details being spoken or typed into a transcript.

  • Action authority limits

    Refund, credit, and address-change authority needs explicit value thresholds, because an unbounded worker with write access to money is a governance failure waiting to happen.

  • Consumer-rights timelines

    Return and cancellation rights carry statutory windows that vary by market, and getting them wrong is a legal exposure rather than a service miss.

Stays human, permanently

These are structural boundaries, not trust levels waiting to be relaxed.

  • Goodwill beyond threshold

    Discretionary spend above the configured limit is a commercial decision, and it should have a person's name attached.

  • High-value and loyal-customer recovery

    When the relationship is worth more than the transaction, a person should be making the save.

  • Fraud and chargeback decisions

    Declining a claim or flagging an account has consequences for a customer that require human accountability.

Where to start

Sequencing matters more than scope

The first blueprint should be the one whose success criteria your team already agrees on. Each step below links to the blueprint that implements it.

  1. Start

    Status and tracking, read-only, one channel

    Answer order and delivery questions from live order and carrier data on web chat. Nothing writes. You are proving grounding and identity checks before money is in scope — ideally well before peak.

    Read the blueprint
  2. Then

    Approved actions at a deliberately low threshold

    Enable address changes, returns, and reorders up to a value limit you will initially think is too conservative. Everything above it goes to approval, and the approval data tells you where the limit really belongs.

    Read the blueprint
  3. Next

    All channels, peak-ready

    Extend across messaging, email, and voice with shared context, and raise thresholds where the audit record supports it before the season starts.

    Read the blueprint

Systems we would expect to read and write

  • Ecommerce platform and order management
  • Carrier tracking integrations
  • Helpdesk and ticketing
  • Payments, refunds, and fraud tooling
  • Loyalty or customer data platform

Blueprints that apply

4 operating blueprints for Retail & ecommerce

Each one carries the full team, process, and control detail — the mechanics this page deliberately does not repeat.

Push back on this

Objections we actually hear in Retail & ecommerce

Answered as we would answer them in the room, including the cases where the right answer is that this is not for you.

Peak is when we can least afford an experiment to fail.
Which is why the first stage is read-only and runs months before peak. By the time volume arrives you are not experimenting — you have an instrumented baseline, thresholds tuned from real approval data, and a known containment rate per contact reason. Launching this in November would be a bad idea and we would advise against it.
Our customers would rather talk to a person.
For a return that went wrong, yes. For "where is my parcel", the evidence in most operations is that customers want the answer, not the conversation — the preference for a human usually shows up when self-service has already failed them. The design bet is that containment on genuinely routine contacts is what makes a person available for the ones where a person matters.
How do we stop it inventing a delivery date?
Status claims come only from live order and carrier systems, and the workers have no authority to estimate. Where the system of record has no date, the worker says what is actually known and, where useful, opens a carrier case. A confident invented date is worse than no date, and the blueprint treats it as a defect rather than a rounding error.

Other sectors

How the argument changes elsewhere

Next step

Argue with this in a working session.

Bring your own numbers. We will map your version of the operating reality, agree what must stay human, and scope the first blueprint against a measurable win — or tell you it is not worth doing yet.