Private beta · first 6 months free

Stop Chargebacks Before They Happen

Microscope is AI-driven chargeback prevention for Shopify and DTC merchants. A four-stage risk engine — Radar, Sniper, Judge, Report — watches every order, blocks the fraud, and keeps good revenue flowing.

No spam. One email when your access is ready.

See the engineNo credit card. One email when your access is ready.
$33.8B
lost to chargebacks worldwide in 2025
61–80%
of chargebacks are friendly fraud
0.9%
Visa risk threshold that triggers fines
$265B
in good orders falsely declined yearly
Live decision streamScore 0.71

Order #48213

$486.00

Decision

review

Handled by: Judge · LLM deep dive

Evidence chain

  • Device shared with 6 flagged accounts
  • BIN velocity +420% vs 24h baseline
  • Address cluster A1 — 11 orders in 40 min
Scanning order streamhourly sweep · 4 anomaly classes
Radar→Sniper→Judge→Report
The problem

Fraud is a tax on every DTC brand

Chargebacks, friendly fraud, and false declines are three leaks in the same pipe — and playing it safe on one makes the others worse.

$33.8B

Global chargeback losses

Merchants worldwide lost $33.8B to chargebacks in 2025 — fees, lost goods, and dispute labour all land on your P&L.

61–80%

Friendly fraud dominates

Most disputes come from customers who received the order and claim they never placed it. Rules can't catch that; evidence can.

0.9%

The Visa line you can't cross

Cross 0.9% and you enter Visa's chargeback monitoring programmes — fines, remediation fees, and eventually processing risk.

$265B

Real revenue, falsely declined

$265B of legitimate orders are rejected every year. Blocking harder is not a strategy — deciding better is.

2–3×

Cross-border multiplies risk

Cross-border chargeback rates run 2–3× domestic, and the domestic signals your current model trusts are simply absent.

The engine

Four stages, one closed loop

Radar finds it, Sniper gates it, Judge interrogates it, Report closes it. Discovery and adjudication stay deliberately separate — so you can widen the search without loosening the standard.

Stage 1

Radar雷达

Wide-area anomaly discovery

Every hour, Microscope sweeps your order stream for four classes of statistical anomaly — plus any subgroup with a weak authorization rate.

  • Velocity spikes across BIN, IP and email domain
  • Clustering and collusion across address, device and payment account
  • Multi-dimensional rare combinations — under 0.01% of the global population
  • Subgroups whose chargeback rate runs ≥5pp above the market

Detect only — never act

Radar raises alerts and hands candidate leads downstream. No order is approved, held, or declined at this stage.

▸A wide-area radar: broad discovery, wide ingress.

Stage 2

Sniper狙击

Deterministic policy adjudication

Validated online policies score orders in real time, allowed to act only while three dynamic gates hold.

  • Chargeback rate ≤ max(0.5%, min(2%, 0.3 × trailing-7-day median))
  • MIN(sep) ≥ 5pp
  • Complaint rate ≤ market × 0.9

The core discipline

Search and gates stay separate: you may change how you search, never how you judge. Thresholds are earned from data, not tuned by vibes.

▸Precision sniping: strict handling, strict egress.

Stage 3

Judge研判

AI deep dive on the grey zone

Mid-band scores and high-value orders go to an LLM for three-dimensional deep research before anything happens to them.

  • Account identity — secondary weight
  • Cluster and collusion — heaviest weight
  • Shipping address and payment semantics — secondary weight

Never decline on score alone

A score is only a referral signal. Any decline must name the order-level evidence fields that justify it — no exceptions.

▸A microscope for single-order adjudication.

Stage 4

Report上报

Closure and feedback loop

Decisions land on the panel with an order-level evidence chain, and every miss or over-turn is fed back into the system.

  • Decision dashboards with an auditable, order-level evidence chain
  • Bad-case feedback: false positives mine new guard rules
  • Missed chargebacks are backfilled into the next detection cycle

Auditable by default

Every action carries the exact fields that produced it, so finance, risk and support argue about facts — not about a black box.

▸Mission closure plus a closed feedback loop.

Output

Five decision actions

Every order lands on exactly one action with its evidence chain attached. There is always a citable reason — never a bare score.

approve

Approve

Signals are clean. Ship the order.

approve_3ds

Approve + 3DS

Route through 3-D Secure to shift liability.

verify

Verify

One lightweight confirmation: CVV, address, or order confirmation.

review

Review

Hold for a human, with the evidence pack already attached.

decline

Decline

Blocked, and you can see the exact fields that triggered it.

Shadow Mode ships with every plan: Microscope scores your traffic for the first 30 days without touching a single order.

Product form

Not a plugin — a decision layer

One core, three ways to plug it in. Microscope lives in the window between an order being created and being sent for authorisation, and returns one decision action together with the evidence behind it. It is not inside the authorisation path, and it replaces nothing in your stack.

How it actually ships

This is the integration shape during beta: the REST API can be wired up exactly as described, while the Shopify app and the audit export open with the plans that include them.

Four stops of one call

1Stop 1

Mount point

POST /v1/score

The order exists; nothing has gone for authorisation yet. One call returns a single decision action — that is the entire integration, not a migration.

2Stop 2

Delivery form

API / app / panel

The same core behind three doors: the REST API is the primary surface, the Shopify app needs no code, and the panel plus audit export is the face your compliance review reads.

3Stop 3

Data contract

12 in / 12 out

What you send is order and payment metadata. What comes back is a verdict, the evidence chain, the gate readings behind it, and how long it took.

4Stop 4

What you keep

audit export

Every decision leaves a log entry made of citable fields. Export it for your own compliance review, or quote it when a dispute comes back from the issuer.

Three delivery forms, one core

REST API

From Starter

One call per order; back comes the action and its evidence chain.

Shopify app

From Growth

Install and go — covers the order flow before checkout, no code to write.

Panel & audit export

Every plan

Decisions, evidence chains and gate readings — searchable and exportable.

What we explicitly are not

  • Not a payment processor, not an acquirer
  • Not inside the payment authorisation path
  • Replaces no part of your existing stack
  • Processor-agnostic: Stripe · Adyen · Braintree · Checkout.com · Shopify Payments

The fields in the artefact

These fields are the entire contract between you and Microscope; the raw payload is in the worked-example section.

Request · what you send
order
order_idamountcurrency
payment
binlast4avs
customer
email_hashaccount_age_days
device
fingerprintip_country
shipping
countrypostal

BIN and last four only — full card numbers are never stored, and emails arrive hashed.

Response · what comes back
verdict
order_idscoreactionstage
evidence[]
fieldsignalvalueweight
gates
cb_ratemin_sepcomplaint_ratio
timing
decided_in_ms

On a decline, evidence is never empty: no named field, no decline.

Show the three diagramsThe four stops above, drawn as structure: where it sits in the payment flow, what one call exchanges, and how the four modules connect.

Where it sits in the payment flow

Not one link of the authorisation chain changes — the decision layer hangs beside it, trading one API call for one disposition.

Where it sits in the payment flowMerchantOrder createdAuthorisedAcquirersnapshot inaction outMicroscope · advisory decision layerPOST /v1/scoreaction + evidence[]
  • Bypass: takes no payment, replaces no link
  • Only in this window: order created, not yet authorised

What one call exchanges

Twelve fields in on the left, twelve back on the right. The middle step is the only thing you change.

What one call exchangesRequest · you send 12Decision coreResponse · you get 12orderpaymentcustomerdeviceshippingverdictevidence[]gatestimingPOST /v1/scorescore · action

Two deliberate constraints

  • BIN and last four only — full card numbers are never stored, emails arrive hashed
  • On a decline evidence is never empty — no named field, no decline

How the four modules connect

Discovery and adjudication are held apart by the gates, so the search can widen without the standard loosening. The dashed edge below is what makes it a system rather than a script.

How the four modules connectThree gatesall must holdRadar雷达Alerts onlySniper狙击Gated actionJudge研判Per-order reviewReport上报Closes the loopFeedback: false positives mine guard rules · missed chargebacks refill the next cycle

Microscope is a risk advisory and decisioning layer — not a payment processor or an acquirer, and not inside the authorisation path. Every decision carries the fields that produced it, and the final call stays with your team.

Worked example

One order, end to end: 87 seconds

The card at the top of this page is an order that reached the engine with a score of 0.71. Here is what happened to it next — and what each possible ending would have cost you.

Illustrative data

We are in private beta, so this is a worked model rather than a merchant case study. The fields, the gates and the decision path are exactly what ships; the order is a stand-in.

Order #482132026-09-24 · 03:14:22 UTC
Amount
$486.00
Channel
Shopify · cross-border
Initial score
0.71 · mid-band

Through the four stages

Stage 1T+0s

Radar雷达

The hourly sweep flags the order twice: it sits inside address cluster A1, and its BIN is running +420% against its own 24-hour baseline.

Stage 2T+2s

Sniper狙击

The three gates are checked and all three hold — holding only permits the policy layer to act, it does not oblige a decline. The order is escalated to deep research.

Stage 3T+9s

Judge研判

The LLM deep dive runs its three-dimensional research: the device fingerprint matches 6 accounts already flagged, and the payment instrument resolves into the same cluster.

Stage 4T+87s

Report上报

The order lands on the review queue with its evidence chain attached. Every field that produced the decision is listed, citable and exportable.

Evidence chain

Order-level and citable — the same fields a dispute response would quote back to the issuer.

  • device_fp d41f…9c — shared with 6 flagged accounts
  • bin_velocity +420% vs 24h baseline, same BIN
  • address_cluster A1 — 11 orders in 40 minutes

Decision path

review1 human checkFraud confirmed — blocked

No order is declined on a score alone. A decline has to name the order-level fields that justify it — so the reviewer argues about facts, not about a black box.

What those 87 seconds were worth

$486
Fraud stopped before the goods left the warehouse
$1,240
Good orders released in the same sweep — 3 that a rules-only stack would have declined
87 s
To a decision a human signed off on

One blocked chargeback already covers a month of Growth — the same arithmetic as our pricing page, applied to a single order.

Integration

This is the entire integration

One call per order. You get back a decision action and the evidence chain behind it. Nothing in your stack is replaced, and Shadow Mode runs the same call without acting on the answer.

Request
POST https://<your-gateway>/v1/score
Authorization: Bearer $MS_KEY
Content-Type: application/json

{
  "order_id": "48213",
  "amount": 486.00,
  "currency": "USD",
  "payment":  { "bin": "453911", "last4": "0912", "avs": "N" },
  "customer": { "email_hash": "…", "account_age_days": 0 },
  "device":   { "fingerprint": "d41f…9c", "ip_country": "US" },
  "shipping": { "country": "US", "postal": "10001" }
}
Response
{
  "order_id": "48213",
  "score": 0.71,
  "action": "review",
  "stage": "judge",
  "evidence": [
    { "field": "device_fp", "signal": "shared_with_flagged_accounts",
      "value": "d41f…9c", "weight": "high" },
    { "field": "bin_velocity", "signal": "baseline_multiple",
      "value": 4.2, "weight": "medium" },
    { "field": "address_cluster", "signal": "orders_in_40min",
      "value": 11, "weight": "high" }
  ],
  "gates": { "cb_rate": 0.006, "min_sep": 0.07, "complaint_ratio": 0.71 },
  "decided_in_ms": 87
}

Endpoint shown as the Beta integration shape — the field list you receive at onboarding is authoritative. Shadow Mode uses this exact call for 30 days without taking any action.

A modelled scenario based on the engine's documented behaviour. Not a customer testimonial and not a performance guarantee — outcomes depend on your data quality, configuration and processor behaviour.

Pricing

Priced to pay for itself

One blocked chargeback covers a month of Growth. Beta merchants lock their rate for a full year.

Beta

Early-bird beta: your first 6 months are free

Month 7 onwards, early-bird merchants keep 50% off for their first full year. No card required to join the waitlist.

Starter

Small shops getting started

$99/month

billed monthly, cancel anytime

  • Up to 1,000 orders / month
  • Real-time risk scoring API
  • Core risk dashboard
  • Email support
  • 30-day Shadow Mode (scores only, never blocks)
Join waitlist
Recommended

Growth

Growing brands

$299/month

billed monthly, cancel anytime

  • Up to 10,000 orders / month
  • Advanced risk analytics
  • Custom rule configuration
  • Priority support
  • Shopify app integration
  • Chargeback Dispute Assist
Join waitlist

Scale

High-volume merchants

$499/month

billed monthly, cancel anytime

  • Unlimited order volume
  • Dedicated account manager
  • Custom ML model training
  • SLA guarantee
  • White-label option
Join waitlist

Beta pricing is locked for 12 months. Enterprise volumes, marketplaces and multi-store groups — talk to us.

FAQ

Questions merchants ask first

Straight answers, including the ones a vendor usually dodges.

Shadow Mode runs Microscope alongside your current stack for 30 days. It scores every order but takes no action, so nothing in your business changes. At the end you get a report with precision, recall, and the exact orders we would have caught or held. You compare that against your real dispute records — accuracy is measured, not promised.

No, and you should distrust anyone who does. Microscope is an advisory decisioning layer: it reduces chargebacks and false declines, but it does not sit inside your processor's authorization path. Every decision ships with an evidence chain so your team keeps the final say.

Under 30 minutes for a single API call. Most merchants start in Shadow Mode with one endpoint, then enable automated actions tier by tier. Shopify merchants can install the app and skip code entirely.

We are processor-agnostic — Stripe, PayPal, Braintree, Adyen, Checkout.com, Shopify Payments and others. If you can send order and payment metadata, we can score it, and nothing in your stack has to be replaced.

TLS in transit, AES-256 at rest. We never store full card numbers — only tokens and derived risk features. Access is role-scoped, every decision is logged, and the audit trail is exportable for your own compliance reviews.

Early-bird merchants keep their terms: months 1–6 free, then 50% off your first full year, locked for 12 months. No auto-upgrade, no card required to join the waitlist, and you can leave whenever you want.

Waitlist

Get early access to Microscope

Join the beta and we'll wire up Shadow Mode on your order stream — 30 days of scoring before a single order is touched.

No spam. One email when your access is ready.

Beta · first 6 months free · cancel anytime