Competitive landscape

Everyone holds one piece. Only the gateway holds the judgment.

The defensible position is gateway-level recovery with suppression intelligence - which pure-software players cannot match because they lack the failure reason and the risk signal.

Where everyone sits

Surgical & measured recovery →

has both → the whitespace

Relay Recover

Razorpay

Static cart recovery

Dunning tools

Shopify apps

Automation

Payment-layer signal →

X · Payment-layer signal

Acts on the live decline reason + gateway fraud/RTO risk, in real time - not blind and after-the-fact.

Y · Surgical & measured recovery

Suppresses the wrong targets and proves incremental lift vs a control group - not spray-and-pray, reported gross.

Conceptual map - each player's position reflects its scores in the capability matrix below, not a precise metric. The top-right needs both axes; only a gateway-native agent sits there.

The wedge

None of the vendors swept advertises propensity-based suppression or control-group incremental-lift reporting - each markets gross recovery, which overstates causal impact.

Scope of the claim

Public positioning only (product pages, June 2026). A vendor may run holdouts internally; holdout measurement itself is standard practice - the wedge is making it the ONE reported number.

What would falsify it

One published control-measured lift number from any competitor - it does not exist in the sweep today.

Capability matrix

Capability	Relay Recover	Static cart recovery	Razorpay	Shopify apps	Dunning tools	Automation
Real-time, at the break	●	○	◐	○	○	◐
Reason-aware fix (decline code)	●	○	◐	○	◐	○
Gateway risk signal (fraud + RTO)	●	◐	◐	○	○	○
Propensity-based suppression	●	○	○	○	○	○
Control-group incremental proof	●	○	○	○	○	○
India UPI / COD / NPCI native	●	●	●	○	○	◐
One unified agent (not siloed)	●	◐	○	○	◐	○

Swipe the table sideways to see every column.

● yes · ◐ partial (scoped: gateway-side only, recurring-card only, or split across products) · ○ no. Public product pages, June 2026.

The gap, per player

Cashfree (the host)Right signals, separate boxes - no joint reasoning, no suppression, no lift proof.

RazorpayRecovery split across 4+ products; reported gross, not incremental.

PayU IndiaRetries the rails, never the customer; no causal lift.

StripeMature dunning, but not UPI/India-native; reports gross.

AdyenNetwork retry for card recurring; no suppression, not India-native.

Shopify apps (Klaviyo, Postscript, WhatsApp)Fires on abandonment, blind to the decline code.

Dunning tools (Chargebee, Churnkey)Subscriptions only; can't touch one-time UPI/COD.

Generic automation (Zapier, n8n)Plumbing, no judgment - structurally can't be surgical.

Product naming last verified June 2026 from public product pages: Razorpay = Agent Studio (Claude Agent SDK) + UPI Autopay Revenue-Protect + Failed Payment Recovery + Magic Checkout - not one product. RTO risk = “RTO Intelligence”, distinct from the fraud product “RiskShield”. Stripe's “55%” is the combined suite, not Smart Retries.