Three actors. Six stages.The case never closes.
The system places the burden of investigation, tracking, and follow-up on the customer. Amazon's posture is passive at every stage.
Frontline associates see the same information the customer already has. They cannot override policy, schedule follow-ups, or differentiate a first contact from a fifth.
No persistent case record survives a contact. No commitment is tracked. No alert fires when a deadline misses. The architecture guarantees the loop.
DNR-related contacts dominate Amazon Customer Service repeat volume. The metrics below combine DNR-specific volume with system-wide context that compounds DNR resolution failures.
Metrics noted as system-wide cover all CS issue types. They illustrate the structural communication gap that compounds DNR resolution failures specifically. · Sources & methodology
A customer's path from the moment a package is reported delivered but missing, through the system that quietly forgets them at every stage. Read each column top to bottom to feel one moment; read each row left to right to see the pattern.
Each row is one dimension of the experience: what the customer does, feels, and what's happening behind the scenes.
Read left to right. Every column is one stage of the journey, from package missing to resolution or escalation.
Pointing or tabbing to any stage lights up the whole column so you can read one moment top to bottom.
Find out where my package actually is
Get someone at Amazon to own this
Close this and decide what I trust next time
Customer left to self-investigate. No system action.
Self-service ends in repetition. 8 of 11 push past it.
FRD enforced. Contact closes. Nothing carries forward.
No callback, no notification, no system memory of the wait.
New associate, full re-explanation, sometimes turned away again.
Path A succeeds late. Path B and C lose context across handoffs.
Tracking page, delivery photo
Order detailPhoto (if available)Help & Contact Us flow, chatbot
Self-service treeChatbotLive chat with associate
ChatPromo credit (sometimes)No surface. Customer left watching.
No proactive commsLive chat with new associate
ChatOrder detailRefund / replacement confirmation, or transfer
Refund flowTransfer to ECR / DARTWhat the map implies
These are not bugs. They are architectural choices the system makes by default, and each one is a guarantee that the DNR journey will fail in exactly the ways Part 3 documented. They compound.
Every contact is treated as a new, isolated event. There is no case record that survives past one interaction, no thread the next associate can pick up, no concept of "this customer is mid-journey." 4.05M DNR repeat contacts a year are the receipt for this choice.
Example: at Stage 5, the associate must manually search 8+ fragmented systems (AC3 Past Contacts, SIM, annotations, email, refund history) to reconstruct what Stage 3 already established.
Associates make commitments during contacts ("we'll investigate," "you'll hear back," "a refund has been issued") that no system tracks, validates, or fulfills. 65% of VPI escalations involve false statements or unexecuted commitments.
Example: a transcript says a refund was issued; account actions don't show one. The customer calls back five days later because the credit never appeared.
The system places the entire burden of tracking, remembering, and advancing the resolution on the customer. Amazon's posture is passive at every stage. The customer is the only actor who remembers the case exists.
Example: zero proactive communication during the 48-hour wait. The customer must self-track when the FRD window expires. No reminder, no countdown, no check-in.
When work IS happening behind the scenes (SIM tickets filed, resolver teams investigating), neither the customer nor the next frontline associate can see it. Work happens in silence, resolves in silence, or fails in silence.
Example: 8.5M SIMs received resolver responses with no written correspondence back to the customer. The investigation closed; the customer never heard.
The system treats a customer's 1st contact and their 14th contact identically. There is no mechanism to detect that a case is failing, that a pattern is emerging, or that intervention is needed. 54% of VPIs involve pattern detection failures.
Example: one VPI customer contacted CS 360 times because every order was delivered outside their requested window. Each contact was treated as new. No escalation triggered.
The DNR experience is not failing because of associate errors, training gaps, or policy edge cases. It is failing because the system architecture treats every customer interaction as a stateless, isolated event. No persistent memory. No commitment tracking. No proactive communication layer. No escalation intelligence. No feedback loop. The system cannot learn, cannot remember, and cannot act without a customer initiating every step. This will produce the same outcomes next year, at the same scale, until the infrastructure changes.
| Dimension | Metric | What it means |
|---|---|---|
| Customer cost | 4.05M repeat contacts/year | Each one is a customer forced to re-initiate because the system forgot. |
| Associate cost | 3-5 DNR cases/day | Same script regardless of history. Associates repeat identical diagnostic steps because the system provides no signal that this is a repeat. |
| Operational cost | 19.1M SIMs / 8.5M silent | Massive backstage work that never reaches the customer. Effort without outcome. (System-wide; illustrates the infrastructure gap that compounds DNR.) |
| Trust cost | 69% of VPI escalations | When cases reach VP attention, the root cause is almost always "the system forgot." (System-wide.) |
| Escalation cost | 54% of VPIs | Pattern detection failures. The system cannot detect its own circular behavior even at 360 contacts on the same issue. (System-wide.) |
Seven priority areas emerge from the research. Each addresses a named pattern from Part 4. Each carries a quantifiable cost of inaction.
Pattern 1 · No persistent memory
Without it: repeat-contact volume persists at 8.69M annually across top 4 drivers; associates continue manually piecing context from 8+ systems; cognitive burden and error risk remain high.
Pattern 1 + 3 · Amnesia + Customer-as-PM
Without it: customers continue re-explaining across contacts; CSAT erosion persists; newer associates particularly struggle under live contact pressure.
Pattern 2 + 4 · False promises + Silent backstage
Without it: 85% of associates continue working around artificial 7-day constraints; complex cases remain inadequately tracked; missed commitments erode trust.
Pattern 4 + 5 · Silent backstage + Flat system
Without it: manual escalation processes persist; resolver-team SLA adherence remains unmonitored; duplicate case risk continues.
Incubation feedback · 2.3/5 satisfaction
Without it: 62% navigation difficulty persists; adoption barriers remain; system investment underutilized.
Pattern 3 · Customer-as-PM
Without it: customers continue initiating unnecessary contacts for status; trust erodes; zero proactive communications during waiting periods.
Tactical design feedback
Without it: associates distrust AI outputs and continue manual verification; AI investment ROI diminished.
The current architecture guarantees failure. The cost is quantified. The incremental approach has been tested and found insufficient. What follows is not optimization. It is replacement.
Given the current architecture, the following five outcomes are not bugs. They are guarantees. Case Management replaces each one. Same six stages, the same customer, the same missing package, re-walked with the new architecture active.
No persistent case record survives past a single contact. The customer becomes the carrier of context.
A case opens at first system touchpoint and travels through every interaction. By the time the customer reaches a live associate, the system already remembers.
No system monitors whether verbal commitments become actions. A refund promised is not a refund issued until someone manually verifies it.
Every promise auto-extracts from the transcript, validates against policy, and lands as an obligation with an owner and a deadline. The contact can close. The commitment cannot.
No communication layer connects resolver-team actions to customer-facing updates. 8.5M SIMs resolve in silence.
Commitment Escalations monitors the SLA throughout every wait period. At 75% of the window, the owner gets an alert. At miss, severity escalates and routes to a specialized team. The 48-hour wait is no longer black-box.
No escalation intelligence detects accumulating contacts or circular handoffs. 360 contacts. 8 in one day. No trigger.
Case Insights detects patterns at three levels: cross-customer (same ASIN failing for many), within-case (this case past its expected resolution threshold), between-team (case bouncing back to a team that already handled it). A 5th contact is no longer treated the same as a 1st.
Context is scattered across 8+ fragmented systems with no consolidation layer. The associate's only option is to ask the customer to re-explain.
Every inheriting associate sees the full narrative on contact acceptance. Prior contacts. Commitments and their status. Actions taken and outstanding. The case closes only when the Closure Decision Service confirms terminal outcome (delivered, refunded, or replaced and verified).
The associate stops carrying the case in their head. The customer stops carrying it on their phone. The system carries it.