Key points
- In retail, the customer promise covers the availability, delivery time and delivery or collection method stated at the time of purchase. It is the most concrete commitment a retailer makes to its customer.
- When the cause of a broken customer promise is not addressed, the next customer encounters the same failure. Fixing an individual order helps that customer, but leaves the execution rule that caused the failure unchanged.
- The cause of a broken promise is rarely where the customer sees the problem: a click-and-collect stockout may stem from an allocation rule, while a cancellation after an order has been accepted may stem from how the promise was calculated.
- Agentic AI makes it feasible to investigate every recurring friction point: agents connect each occurrence to its context and propose a rule change for teams to approve.
The customer promise is a retailer’s most concrete commitment: a product’s availability, delivery time and delivery or collection method, as stated at the time of purchase. Why is a product delivered late at the start of every month? Why do certain stores regularly struggle to fulfil click-and-collect orders on time? These failures are not bad luck. Teams fix each order, yet the same problem returns the following week because its cause lies in an execution rule, often far from where the customer encounters the issue. Finding that cause means examining each occurrence alongside its context — work that no one has time to do for a failure that seems minor. Agentic AI makes that work possible: AI agents continuously monitor what happens to orders, trace failures to their causes and propose corrections for teams to approve.
Why does the same customer promise keep being broken?
A broken customer promise recurs because teams fix orders one at a time. The customer receives their order, the incident is closed, and the same problem reappears the following week on other orders.
These failures occupy a grey area: frictions that are too frequent to be exceptions, but too small to justify an investigation. Many are handled manually so often that no one reports them any more: a revised delivery time communicated to a customer, or a collection method changed by hand. Individually, they cost little. Together, they erode customer trust.
Metrics show that the promise was not kept. They do not show where the cause lies — and it is rarely where teams first look.
Where is the cause of a broken promise?
The cause of a broken customer promise is rarely where the customer encounters it. The customer sees a cancellation, a delay or a stockout in store. The cause often lies earlier in the order journey: an allocation rule, a promise calculation or a stock reservation.
The three scenarios below illustrate this disconnect. They are typical situations in complex supply chains, not customer results.
| Visible symptom | Where it appears | Underlying cause | Rule changed |
|---|---|---|---|
| City-centre stores run out of a popular product mid-month, even though their in-store sales are normal. | Store | Web orders collected in store are deducted from the store’s allocation, which was sized for in-store sales alone. | Create a dedicated web allocation for click and collect; resize store allocations. |
| Orders for a new smartphone are cancelled after being accepted, in the week following each delivery. | All channels; customer service | The promise is calculated using the stock announced by the manufacturer. Deliveries fulfil orders in the order they were placed, so when fewer units arrive, the latest orders are cancelled. | Cap the promise at the likely quantity; place the rest on pre-order; introduce quotas by channel. |
| Waiting times increase for replacements of low-value products under warranty. | After-sales service | Every faulty item goes through an inspection at a central warehouse, even though the inspection never changes the decision for certain product–fault–warranty combinations. | Replace immediately, without requiring a return, for those combinations. |
In each case, the customer experience team sees the symptom in its metrics: negative reviews, customer service contacts and a falling NPS. It cannot see the rule producing the problem because that rule sits within supply chain execution. CX action plans therefore often focus on communication with the customer, while the operational cause remains.
How can agentic AI make the customer promise more reliable over time?
Agentic AI makes the customer promise more reliable by making it feasible to investigate every recurring friction point. Looking at a single order cannot explain a recurring problem. The explanation lies at the intersection of several factors: the order, the rule applied to it, the stock available at the time and a constraint no one has modelled. Reconstructing that context takes hours of specialist work for a case that appears minor. An agent is not more intelligent than the teams; it can simply be exhaustive, across every order, all the time.
This continuous-improvement process follows four steps:
- Discover: Agents monitor cancellations, delays, manual corrections and lost demand, including frictions that no one reports any more. They produce a short list of qualified signals, not a flood of alerts.
- Pattern: Agents connect occurrences to their context — applied rules, stock, carriers and customer agreements — and put forward hypotheses. Teams challenge them and add what no system records, such as a promotion applied manually or a local arrangement with a store.
- Decide: Agents propose changes consistent with the company’s decision rules and service objectives. Teams decide what to do, both for orders already affected and for the rule itself.
- Deploy: The approved rule is deployed and applies to all similar cases in the future.
Kbrw’s Agentic Execution Engine embeds this process in its agentic order management system (OMS). At a global luxury leader with more than 6,000 stores, an agent connected to the OMS investigates order problems reported by stores and identifies their causes. The system is being tested ahead of a global rollout. After two months, the client estimates that it saves each Retail Planner up to half a day per week.


