Shield

Canonical interaction model · v1

Customers feel disruptions before we know they exist.
Shield closes that gap.

Three actors. Six stages. Blast radius determines autonomy — not system confidence.

  1. 01 / 04

    Cross-domain correlation is the unowned middle

    No source-owning team correlates signals across payment, fulfillment, devices, and CS in real time. Shield does. PRFAQ FAQ 6

  2. 02 / 04

    Blast radius scales autonomy, not confidence

    The AI knows what to do at every scale. The question is whether the stakes require human approval first. PRFAQ FAQ 3

  3. 03 / 04

    Invisibility is a feature

    At low blast radius, the customer never knows. Trust is preserved by absence, not by communication. PRFAQ HTD 1

  4. 04 / 04

    Recovery beats prevention when prevention fails

    Proactive resolution scores 12 points higher in satisfaction than reactive (82.5% vs 70.1%). The recovery layer earns its build cost. PRFAQ Appendix A

$453M in reactive cost. 40 minutes to even know.

2026 projections from PRFAQ FAQ 1 and Appendix A. Each number is the gap Shield is built to close. Sources in the sidecar; underlying study deck and Appendix D pending PM forward.

$0M Annual cost of reactive operations $84.9M CS contact + $368M concessions, 2026 projection. PRFAQ FAQ 1
0 min Time to know an issue exists Median today. Shield targets <15 min. PRFAQ FAQ 1
0 pts Trust gap, proactive vs reactive 82.5% vs 70.1% satisfaction. n=1,022, May 2026. PRFAQ Appendix A
0M Customers affected per year 1,900 incidents, 30M associate contacts (2026 pace). PRFAQ FAQ 1

PRFAQ stats marked "needs sourcing" until PM forwards underlying study deck and Appendix D. · Sources & methodology

A pattern builds. Shield sees it. A human decides. Trust holds.

Six stages through time, read left to right. The customer is invisible to the disruption; Shield sees the pattern across domains; a human approves the response inside the 15-minute window; trust holds. The same shape applies at every blast radius — what changes is the weight of Stage 4.

How to read this

Rows are layers

Each row is one dimension of the experience. What the customer does, feels, sees, and what's happening behind the scenes.

Columns are time

Read left to right. Every column is one stage of the journey. The bar at the top of each stage shows where you are in the arc.

Hover a stage to focus it

Pointing or tabbing to any stage lights up the whole column. Use it to read one moment vertically, top to bottom, across every layer.

Outlined items open detail

A ring opens a popover with evidence. A framed thumbnail enlarges a screenshot, or every screen for that stage.

Phases Story chapters
Tension builds
Action & resolution
Stages Time flows left to right
1The failed paymentCustomer experience
2The pattern buildsGrowing impact
3Shield sees the patternAI detection
4The human decidesHuman-in-the-loop
5Proactive recovery at scaleCustomer outreach
6Trust strengthenedAfter
What you'd see Illustration per stage
Narrative Story beats
The failed payment

A customer's card declines at checkout. They retry. They retry again. Three of their last six transactions have failed, all on Visa. They wonder if their card is broken.

The reveal

Visa and Mastercard auth failures elevated. Amazon Pay unaffected. Root cause candidate: a deployment 47 minutes ago. The pattern was there all along; only Shield could see it.

The loop closes

Outcome attributed. Pattern enters the playbook. Next time auth-failure rates climb on a single network, the system may handle it autonomously.

Actors Human
Customer
CS associate
Operator
Customer
Customer
Digital
Shield
Shield
Shield
Shield
Shield
Shield
Customer Actions What they do
Retries the card Card declines at checkout. Retries twice. Considers a different card.
Calls CS Asks if their card is broken. Associate hears similar reports but has no root cause.
Reads the message Receives a notification: the issue is fixed, a credit is on the account, no action needed.
Returns to checkout Payment goes through. Customer never had to chase the resolution.
Customer thoughts In their words

"My card was just declined. Let me try again."

"Third time. Is my card broken?"

"They caught it before I even called."

"Honestly impressed they caught that so fast."

Emotion How it feels
Anxious
Frustrated
Relieved
Trusting
Behind the scenes What Shield is seeing
Watching A single auth failure. Within normal variance.
Watching Auth-failure rate climbing on Visa. Mastercard climbing too. Amazon Pay flat. None individually alarming.
Threshold crossed Visa & Mastercard auth failures elevated. Amazon Pay unaffected. Root cause: deployment 47 min ago. Confidence high.
Recommendation "Roll back the deployment. Run the recovery campaign." Operator approves inside the window.
Executing Affected customers notified. Credits applied. Stuck transactions reprocessed.
Learning Outcome attributed. Pattern enters the playbook. Next deployment-borne auth regression may handle autonomously.
Support Processes Source-owning systems
Payments Tech
Payments Tech · AC3
Payments Tech · CS DI
CS Value Engine · deployment-rollback tooling
CJM · AC3 · Concessions Orchestrator (proposed)
Continuous Attribution
Phase 0 reality check What's live today
Today, the journey stops here

Phase 0 detection (Perfect Mile delivery POC, Jul-26) is the live reference for cross-domain detection. Payments onboarding targets Sep-26. Stages 4–6 are Phase 2 vision — recovery levers tagged TBC in PRFAQ Appendix E.

What the map implies

Four patterns the storyboard reveals.

The journey shows what's broken about today; these patterns name it. Each one shows up across multiple stages and across multiple incidents. Shield is built to break each one.

Pattern 01

Detection lag

Detection trails customer experience by minutes at best, days at worst. The MX deployment failure ran six days before anyone noticed; the IN payment-delay state ran for two years. Customers know first.

Stages 1–2: customer is retrying; CS associate is fielding contacts. Shield (Stage 3) has not yet crossed threshold.

Pattern 02

Silent sufferers absorb the cost

About 55% of customers experiencing a delivery delay never contacted CS. They absorb the disruption silently, shift behavior quietly, and the business never sees the trust loss in real time.

PRFAQ Appendix D: silent-sufferer cohorts spend $0–$16 less in the 30 days following an unresolved incident; lowest spend tier loses $21.50.

Pattern 03

Generic remediation, regardless of context

When recovery happens, it is one-size-fits-all. The Belgium carrier strike sent identical €5 gift cards to 21,000 customers regardless of order value, urgency, or history. The remediation acknowledges the disruption but doesn't fit the customer.

PRFAQ FAQ 1 Problem 2: the €5 amount was decided on what finance was anecdotally felt to approve, not what would resonate with customers.

Pattern 04

No closed loop

Outcomes aren't measured. The same incident shape recurs, and the system can't tell whether last time's response worked. Each event is a fresh investigation, not the next iteration of a learned playbook.

PRFAQ Appendix H: post-recovery CSAT, customer-journey metrics, and downstream-spend impact are all "not measured" today. Repeat-incident rate runs at ~21%.

How it plays out

A real Amazon payment incident, before and after Shield. Same disruption shape; opposite outcomes.

Build Shield. The journey is the proof.

Phasing follows PRFAQ FAQ 10. Phase 0 proves the detection layer with one delivery use case. Phase 1 builds the response interface and onboards payments. Phase 2 closes the loop with the customer-facing recovery engine.

Now through Sep-26 · Phase 0

Prove

Demonstrate that faster detection delivers measurable value. Perfect Mile delivery POC (Jul-26), recovery experiment (Aug-26), CS incident scorecard (Sep-26). By end of phase: detection approach validated, recovery impact measured, data to justify dedicated investment.

Oct-26 through Mar-27 · Phase 1

Detect + Respond

Build the production interface incident teams use to handle bridges, resolvers, action items, impact quantification, associate alerts. Onboard payments into the detection layer. First production-grade detection and response capability.

Q1-27 through Q3-27 · Phase 2

Recover

Customer-facing recovery engine: personalized outreach, segment-tailored credits, channel-of-choice notification. Closed-loop attribution measures cost per incident and trust impact. The vision in the storyboard becomes operational reality.

Resourcing ask: two two-pizza teams (BI / engineers / scientists) per PRFAQ FAQ 8. Critical dependencies: CS DI ingestion, CS Concessions tooling refresh (proposed Concessions Orchestrator, currently unfunded), CJM outbound rails, AC3 in-line associate context.