Case studies — seven advertiser programs, audited numbers

Flagship program

Meridian Fintech cut cost per funded loan by a third.

Finance · lending · CPA on approved applications · IN and AE · eleven months and counting. Offer OFR-4821.

Challenge

Cheap leads, expensive loans.

Meridian's in-house CPL buying was hitting its volume target every month and missing its cost-per-funded-loan target by a wide margin. A 61% rejection rate at underwriting meant they were paying for eleven leads to fund four loans, and their agency was optimising to the lead, because that was the only number it could see.

Approach

Move the payable event down the funnel.

We repriced the program from CPL to CPA on approved applications at $24.00, which shifted the rejection risk from Meridian to us and to the publishers — and gave everyone the same incentive for the first time. Then we rebuilt the pre-qualification flow so income band, employment type and existing obligations were collected before the KYC step rather than after it, and wired their underwriting decision back to us as a postback so publishers could see approval rates by sub-ID within the hour.

Pre-qualification questions reordered; drop-off moved from step four to step one, where it is cheap.
Underwriting outcome postbacked per application, exposing approval rate by sub-ID in near real time.
Publishers with sub-40% approval rates capped rather than removed, then coached back up.
Scrub window fixed at 12 days and published on the offer page, ending the monthly reconciliation argument.

Result

Fewer applications, far more funded loans.

Raw application volume fell in the first six weeks and Meridian's board asked hard questions about it. By week ten, funded volume had passed the old peak on 38% less spend. Lead validity settled at 91%, and cost per funded loan has stayed within a narrow band for eight consecutive months.

Before and after: cost per funded loan

Six months CPL baseline vs six months CPA

−34%

CPL baseline CPA on approved applications

Cost per funded loan improved

−0%

$228 → $148 blended, IN and AE

Approved volume

0.0×

Versus the CPL baseline peak

Lead validity

0%

Phone-verified, duplicate-suppressed

Fraud blocked pre-billing

0.0%

Of raw clicks, rolling 90 days

Six more programs

Different verticals, the same argument about quality.

Each of these started with a metric the advertiser could not defend internally. Three headline numbers apiece.

Finance · payments

OrbitPay grew funded wallets on a strict fraud budget

Wallet sign-ups were plentiful and worthless: half never funded. We gated payout on a first deposit of ₹500 or more within seven days, and cut 96 sub-IDs whose deposits reversed inside the chargeback window.

3.1×Funded wallets

−41%Cost per funded wallet

1.4%Reversal rate

OFR-4902 · CPA · IN, AE, SG

Gaming · casual

Playloop scaled to 380K installs without wrecking D7

Install volume was cheap and hollow. We gated payout on tutorial completion, killed 140 sub-IDs in the first fortnight and moved the CTIT threshold to catch click injection before the install was ever counted.

380KInstalls in 90 days

28%D7 retention held

$1.85Blended CPI

OFR-3390 · CPI · IN, ID, BR, PH

E-commerce · home

Sable Commerce built a 620-publisher affiliate channel

From no affiliate revenue at all to 19% of total online sales in eleven months, on a 9% CPS with a 30-day cookie, coupon-partner controls and a nightly reconciliation against their commerce feed.

19%Of online revenue

620Active publishers

$74Average order value

OFR-5107 · CPS · IN, SG, GB

Travel · OTA

Voyara recovered demand across three peak seasons

Travel intent is spiky and their fixed CPS rate could not follow it. We introduced seasonal rate cards published four weeks ahead, so publishers could plan inventory instead of guessing.

2.8×Peak-season bookings

$0.74Network EPC

5.5%Commission on stay value

OFR-2745 · CPS · IN, AE, TH

EdTech · courses

Nimbus EdTech priced acquisition on trial-to-paid value

Every trial was worth the same to their old CPA ceiling. Once we had the day-41 breakeven curve per geo, we could pay 2.2 times more for a trial in AE than in BR and still improve blended margin.

D41Cohort breakeven

+27%Trial-to-paid rate

3.6×90-day ROAS

OFR-6188 · CPA · IN, AE, SEA

Finance · insurance

Kavach Insurance moved from CPL volume to issued policies

Motor and health leads converted at wildly different rates and were paid identically. Splitting the offer by product line and paying on issued policy stopped the cross-subsidy inside one quarter.

+63%Policies issued

−29%Cost per policy

4.4×Program ROAS

OFR-5533 · CPA · IN

The long version

Three programs, written out properly.

Including the parts that did not work, because those are the parts worth reading.

Gaming · casual · mid-core

Playloop Games — buying retained players, not installs

Playloop came to us after a burst campaign delivered 210,000 installs in three weeks and a day-7 retention figure of 6%. The installs were real in the sense that the MMP counted them, and worthless in the sense that nobody opened the game twice. Their VP of Engineering had already worked out that a large share of the installs were click injection — the click-to-install time distribution had a spike in the first ten seconds that no organic behaviour explains.

We restructured the offer so payout fired on tutorial completion rather than on install, set a CTIT floor that discarded anything under fifteen seconds, and clustered devices by hardware fingerprint to find farms running the same handset image. In the first fortnight we cut 140 sub-IDs, which removed about 22% of volume and roughly 3% of post-tutorial events — the ratio that told us we had cut the right ones.

What did not work: our first attempt gated payout on a day-3 session, which pushed the publisher payment cycle out far enough that three good partners left for a network with faster terms. Tutorial completion was the compromise — a same-day event that still correlates with retention at 0.71.

Over the following 90 days Playloop took 380,000 installs at a blended $1.85, with D7 retention holding at 28%. That is 4.6 times the retained-user volume of the burst campaign, at 31% lower cost per retained player.

E-commerce · home · retail

Sable Commerce — building a channel from zero

Sable had no affiliate channel at all. They had been quoted an eighteen-month runway by two other networks and were sceptical that publisher supply existed for homeware at a $74 average order value. The first commercial question was whether to allow coupon partners, because coupon traffic is easy to recruit and very good at claiming credit for sales that were going to happen anyway.

We opened the program to content and review publishers first, at 9% CPS with a 30-day cookie, and held coupon partners back for four months. When we did admit them, it was on a reduced 4% rate with a last-click rule that only paid if the coupon click was the first touch in the session. That single rule is why the channel's incrementality survived scale.

Reconciliation runs nightly against Sable's commerce feed, so returns and cancellations reverse inside the 14-day scrub window rather than surfacing as a dispute six weeks later. Publishers dislike reversals but they dislike surprise reversals a great deal more.

Eleven months in, 620 publishers are active in a given month and the channel accounts for 19% of Sable's total online revenue, at a blended EPC of $0.88.

Finance · insurance

Kavach Insurance — paying for policies, not paperwork

Kavach ran a single CPL offer covering motor, health and term products at one flat payout. Motor leads converted to issued policy at roughly 14%, health at 6% and term at under 3% — so publishers rationally sent whatever was easiest to generate, and Kavach's cost per issued policy drifted upward every quarter while their lead cost looked flat.

We split the offer into three product lines with separate payouts, then moved each to CPA on issued policy. Motor was straightforward. Health required a longer scrub window because underwriting takes time, and we published 14 days rather than the 7 we would have preferred. Term we paused entirely for six weeks: no publisher could make the economics work at any payout we could justify, and pretending otherwise would have burnt goodwill.

We also ran a six-week geo holdout across four states to measure incrementality, because insurance has strong organic search demand and last-click was flattering the channel. The measured lift came in at 21%, lower than the ledger claimed and high enough to keep investing.

One quarter after the split, policies issued were up 63% and cost per policy was down 29%. Term relaunched in month five at a payout that works, on two publishers rather than forty.

All seven programs

The numbers in one place.

Conversions are post-scrub. ROAS is against confirmed revenue reported by the advertiser. Fraud blocked is the share of raw clicks rejected before billing.

Advertiser program results by vertical, pricing model, conversions, ROAS and fraud blocked
Advertiser	Vertical	Model	Conversions	ROAS	Fraud blocked
Meridian Fintech	Finance · lending	CPA	118,400	5.2×	7.9%
OrbitPay	Finance · payments	CPA	246,900	4.8×	9.1%
Kavach Insurance	Finance · insurance	CPA	61,700	4.4×	5.6%
Playloop Games	Gaming · casual	CPI	380,000	2.9×	8.4%
Sable Commerce	E-commerce · home	CPS	204,300	6.1×	4.2%
Voyara	Travel · OTA	CPS	88,600	3.4×	6.3%
Nimbus EdTech	EdTech · courses	CPA	142,800	3.6×	5.1%

Figures cover FY25–26 to date and are reconciled against advertiser-reported revenue. Blended network approval rate over the same period was 87.4%.

What partners say

Written by the people who signed off the spend.

Six weeks in, our application volume was down and I had to defend that to a board that had only ever been shown lead counts. The reason I could defend it is that every rejected conversion had a reason code attached. We argued about data instead of vibes.

Rohan MehtaHead of Growth, Meridian Fintech

They told us the day-3 gate was a mistake before we noticed it ourselves, and proposed the tutorial event instead. Vendors do not usually volunteer that their own idea cost you three publishers. That is the reason we renewed.

Daniel SrinivasanVP Engineering, Playloop Games

Holding the coupon partners back for four months was not a popular decision internally. It is the single reason the channel still shows incremental revenue at 19% of online sales rather than cannibalised revenue at 30%.

Priya NairPerformance Director, Sable Commerce

Programme statistics

0 Advertiser programs run since FY24

0.0× Median program ROAS across verticals

−0.0% Median CAC reduction in the first 90 days

0.00% Postback delivery over 12 months

Advertisers

Programs currently live on the network.

Seven advertiser brands across five verticals, plus five managed disciplines running alongside them.

Meridian Fintech

OrbitPay

Kavach Insurance

Playloop Games

Sable Commerce

Nimbus EdTech

Voyara

AppsFlyer

Adjust

Branch

Shopify

Salesforce

Your program next

Bring the metric you can't defend.
We'll show you where it's leaking.

Send us ninety days of campaign data and the Delhi team will map it against live network benchmarks on a call, Monday to Friday, 10:00 AM – 7:00 PM IST.

Book a program review WhatsApp us