Faster experimentation: 84 variants in four weeks vs. 12-18 months before.
COMPANY OVERVIEW
The company, at a glance.
FOUNDED
1850
INDUSTRY
Financial Services and Payments
EMPLOYEES
~77,000
GEOGRAPHY
130+ countries
use case
Landing page personalization for the consumer cards acquisition funnel
SCALE OF OPERATIONS
145M+
Cards in force
$65.9B
FY2024 revenue
130+
Countries accepted
~$200B+
Market capitalization
A landing page personalization proof of concept across the consumer cards acquisition funnel: 9 pages spanning premium travel, cash back, and no-fee card products. Four intelligence types deployed simultaneously: keyword-based, location-based, repeat visitor, and multi-armed bandit optimization. Each representing a different layer of buyer intent on a single enterprise domain.
We quickly adopted Fibr AI to align our ads with personalized landing pages, achieving great results with minimal effort. We're excited to scale this across our campaigns for increased growth and improved team productivity.
A
Abhishek Bhushan
Senior Manager, Performance Marketing · Livspace
In a Nutshell
The bottleneck was never strategy. It was velocity.
A Fortune 100 global payments network had the data, the intent signals, and the ad assets to personalize its entire card acquisition funnel. What it didn't have was a way to act on any of it without sequential engineering, design, legal, and compliance cycles; a process that turned a single CTA test into a multi-month project. With Fibr, the team built and compliance-cleared 84 personalization variants across 9 pages in four weeks, without a single engineering ticket for variant creation.
THE CHALLENGE
The growth and performance marketing team ran one of the most sophisticated paid search programmes in financial services: 20+ keyword clusters on the flagship card alone, tight segmentation, strong brand equity, and high traffic volumes. The targeting was not the problem.
The problem was everything that happened after the click. Testing a single page variant required coordination across engineering, design, legal, analytics, and compliance, all sequential, not parallel. A single CTA test could take a full quarter from hypothesis to live traffic. At that pace, the team could test roughly 8 to 12 hypotheses a year across all pages combined.
The cost of mismatch
Hundreds of potential optimizations sat untested in backlogs not because they lacked value, but because the operational machinery to test them at scale didn't exist. Meanwhile, despite granular ad targeting, every keyword segment landed on the same static page. A visitor searching for lounge access and one searching for hotel credit, on the same premium card, saw identical content. The ad promised relevance. The page didn't deliver it. The team sensed conversion could improve and blamed page design. Fibr's analysis connecting keyword clusters to scroll depth and exposure rate showed the real gap was intent-to-content mismatch, not aesthetics.
THE SOLUTION
Four intelligence types. One domain. Built in parallel, not in sequence
Fibr's approach centered on proving operational velocity at enterprise scale building everything a sophisticated internal team would want to test, simultaneously, without adding engineering load.
STEP 1
The Analysis: where intent was getting lost
The top five keyword clusters per page were selected by click volume, together covering 64 to 86 percent of traffic on each page. Each cluster was mapped to a specific buyer intent: a high-volume lounge-access keyword cluster representing one buyer, a grocery cash-back cluster representing an entirely different one. Personalization copy was sourced exclusively from each cluster's own ad headlines and descriptions never invented, ensuring message match and giving compliance a clean audit trail.
STEP 2
The Signals: keyword, geography, returning visitor, and algorithmic optimization
Four distinct intelligence types were deployed simultaneously on the same domain: keyword intent mapped from search term to ad asset, geolocation across five major US cities for non-paid-search visitors, returning visitor detection via cookie, and multi-armed bandit optimization allocating traffic algorithmically across element combinations. This was not four separate projects staged over four quarters. It was one parallel build.
STEP 3
The Setup: structural changes, not just copy swaps
Beyond message-matched copy, Fibr restructured page elements based on what the data showed. A benefits section was reordered to lead with the specific benefit matching keyword intent dining content first for dining-intent traffic, grocery content first for grocery-intent traffic. Generic category labels were replaced with specific earn-rate callouts. A below-the-fold value statistic was promoted to a more visible position as part of a structured test.
The New User Experience
A visitor arriving from a lounge-access search saw lounge-led messaging sourced directly from that ad's own copy. A returning visitor saw a different experience than a first-time visitor on the same page. A visitor browsing from a different city saw destination imagery matched to actual travel pattern data from that city, not generic stock photography. All four intelligence types operating on the same domain, simultaneously, without one engineering ticket separating them.
The results
Building 84 variants through the team's existing process: one engineering ticket, one design review, one legal review, and one QA cycle per variant was estimated to take 12 to 18 months of sequential work. With Fibr, the same 84 variants were built, compliance-cleared, and made sandbox-ready in four weeks. That is roughly a ten-to-fifteen-times improvement in experimentation velocity the exact multiple the team set out to validate.
variants built
↑21 %
Personalization variants shipped across 9 pages in the acquisition funnel.
time to sandbox-ready
↓ 33%
From data receipt to variants ready for cross-functional review.
engineering tickets
↑21 %
For variant creation. Engineering only handled IP whitelisting pre go-live.
intelligence types
↑100 %
eyword, location, repeat visitor, bandit deployed all at once
Cohort standouts
Value-based comparison messaging improved engagement consistently across segments. Geographic headline personalization drove the highest conversion rates of any cohort tested.
KEY INSIGHTS
INSIGHT 01
The real bottleneck in enterprise personalization is operational velocity, not strategy.
The team already had the keyword clusters, the scroll-depth data, and the ad assets needed to personalize their funnel. What was missing was a way to act on all of it simultaneously instead of one sequential engineering-design-legal cycle at a time. Collapsing that sequence into a parallel workflow was the entire unlock conversion uplift becomes the validation, not the headline.
INSIGHT 02
Sourcing copy exclusively from existing ad assets solves two problems at once
Every variant's language came directly from the ad that triggered it never invented. This guaranteed message match between ad and page while also giving compliance and legal teams a clean, auditable source for every line of copy, which matters enormously at companies where every word is reviewed.
INSIGHT 03
Compliance-safe personalization needs a reusable language framework, not case-by-case review.
Discovery verbs paired with "Your Offer" cleared review. Commitment verbs paired with the same phrase did not, because they implied a guarantee the underlying offer couldn't make. Turning that distinction into a standing rule rather than re-litigating it variant by variant is what let the compliance review scale to 84 variants without becoming the new bottleneck.
