.svg){kind=icon}
.svg){kind=icon}
.svg){kind=icon}
.svg){kind=icon}
.svg){kind=icon}
Key takeaways:
- Server-side rendering of live rates placed above the fold reduced bounce from 51.7% to 17.0% overnight.
- The first question on your form determines whether a veteran starts or leaves, so design it as a one-tap identity question, not a data entry field.
- 84–97% of VA IRRRL ad traffic is mobile, so test on iOS Safari before launch.
- Reducing bounce rate from 52% to 15% cuts effective cost per click (CPC) by 43% with no changes to ad spend.
- Track first-question clicks as a micro-conversion from day one.
We spent $10,000 running Google Ads for VA IRRRL leads. We tested four different landing pages, analyzed 1,000+ session recordings in Microsoft Clarity, and tracked every click, bounce, and form interaction across six weeks of live traffic.
Here’s a look at everything we learned.
Why we ran this experiment
The VA interest rate reduction refinance loan (IRRRL) is one of the most valuable refinance products available to veterans. It’s a streamlined refinance with no appraisal, minimal paperwork, and consistently lower rates than conventional refinance products. Based on all of that, generating qualified leads for VA IRRRLs at a reasonable cost might seem easy. It’s not.
Most mortgage advertisers running Google Ads for VA refinance traffic make the same mistake: they invest heavily in the ad and almost nothing in what happens after the click.
We wanted to understand exactly what happens when a veteran clicks a VA IRRRL ad, lands on a rate comparison page, and either converts into a lead or disappears. So we ran a controlled experiment across four landing page variants, collected behavioral data from every session, and let the numbers tell us what works.
The setup
Campaign: Google Search Ads targeting VA IRRRL keywords
Duration: June–July 7, 2026
Total sessions analyzed: 1,159
Primary tool for behavioral analysis: Microsoft Clarity session recordings
Conversion metric: Workflow start
We defined "workflow start" as any click on the page. In a multi-step mortgage lead form, that first click — answering the first question — is the strongest leading indicator of whether a borrower will complete the form and become a lead. If they don't click, they don't convert. It's that simple.
Top keywords by spend:
Our CTR averaged 14% across the campaign, roughly three to four times the mortgage industry average of 3–5%. The ads were working. But when it came to converting the people who clicked into leads, the landing pages were the problem.
The four landing pages we tested and what the data showed
We ran traffic to four page variants simultaneously, using the same ad copy and keywords across all four:
The only variable was the landing page design. And that taught us a lot.
*Full campaign, 624 sessions, June 8–28, 2026
The numbers above contain two important surprises. Understanding them is the core lesson of this entire experiment.
Lesson 1: A high bounce rate can hide good data
/get-va-refi-rates-2 received the most traffic in our campaign (364 sessions, 58% of all clicks). It also had the worst bounce rate: 34.9% of visitors left in under three seconds.
That means more than one in three people who clicked our ad, and for whom we paid approximately $8.52 per click, spent less time on our page than it takes to read a single sentence.
At 364 sessions and $8.52 CPC, approximately 127 of those sessions were pure waste — visitors who saw a loading state or an incomplete page and left immediately. At $8.52 per wasted click, that's roughly $1,081 in spend that generated zero engagement.
The culprit? The rate widget was loading via JavaScript after page render. When Googlebot or a real user hit the page, the rate table was empty. Veterans searching for "VA IRRRL rates" landed on a page that showed no rates. They left.
The takeaway: We moved rate data to server-side rendering so actual rate numbers appear in the initial HTML response, rather than injected after a JavaScript load. On June 24, when we added static rate content to the landing page, median session duration jumped from 2 seconds to 27 seconds — a 1,250% increase — overnight.
Lesson 2: The first question on your form is the conversion event
The most important design decision on a VA IRRRL landing page is not the headline, the hero image, or the CTA button color.
It’s the first question your form asks.
Here’s a look at how dramatically that first question affected our workflow start rates across pages:
- /serve-button (37.8% workflow start rate): This page opened with branch of service selection displayed as large emoji icon buttons: 🪖 Army, ⚓ Navy, ✈️ Air Force, and so on. Low friction. Visual. Personal. It didn’t feel like a form. It felt like a question a friend would ask.
- /get-va-refi-rates (31.0% workflow start rate): Our original page featured a more traditional form opening. We still saw solid performance, and this page had the highest engagement depth of all variants. Specifically, sessions that started here averaged 3.51 clicks and had the highest over-20-second rate (45.1%).
- /quote-button (7.7% workflow start rate): Here is the most instructive data point in the entire experiment. This page had the lowest bounce rate of all four variants at 3.8% and 42.3% of visitors stayed over 20 seconds. People were reading the page and staying. But only 7.7% clicked anything.
That combination — high engagement time, near-zero bounce, near-zero clicks — has one explanation: the first interactive element on the page was not tappable on iOS Safari. Visitors were reading, scrolling, and trying to tap buttons that were not registering their touch. They were not abandoning. They were stuck.
With 19 of 26 sessions on iOS Safari, a broken tap target on the first question meant 92.3% of visitors hit a silent dead end.
The takeaway: In a mobile-first VA campaign (our audience was 84–97% mobile depending on the day), a tap target bug on the first form element is a conversion catastrophe that looks like a design problem in the data.
Lesson 3: Engagement time isn’t enough (you need clicks)
Early in our campaign, we were encouraged by /quote-button's low bounce rate and high time-on-page numbers. It looked like a winner on surface-level metrics.
But workflow start rates tell a different story. A veteran spending 65 second on your page without clicking anything is not an engaged lead. They’re a frustrated visitor who can’t figure out how to interact with your form.
The click distribution on our best-performing pages showed a clear bimodal pattern: visitors either clicked nothing (bounced) or clicked deeply (5, 10, 15, 20+ clicks, going through multiple form steps). There was almost no middle ground.
This means your form has a binary outcome for most visitors. They either get started and keep going, or they never start at all. The entire conversion problem is concentrated at the first click.
The takeaway: If you can get a veteran to answer the first question, the odds of them completing the form increase dramatically.
Lesson 4: Adding rates to the page changes everything
On June 23, we made a single change to one of our landing pages: we added live interest rates to the above-the-fold content.
Here’s what happened to session metrics the next day:
A single change. One day. Median session duration increased by 1,250%. The bounce rate dropped by two-thirds.
Veterans searching for "VA IRRRL rates" want to see rates. When they land on a page that shows them actual VA IRRRL rates immediately — not a loading spinner, not a form with no context — they stay, read, and engage.
The takeaway: This is the most important data point in our entire experiment. Your landing page headline can promise rates. Your page actually has to deliver them above the fold, in the HTML, before JavaScript runs.
Lesson 5: Traffic quality varies by day, so build for your worst day
Across our campaign, we noticed significant day-to-day variation in engagement metrics that had nothing to do with page design. Here’s the daily breakdown for our best-performing page variant (new serve-button design, July 4–7):
Monday was consistently our weakest day across every page variant throughout the campaign. Saturday traffic was mixed (holiday weekend effect). Sunday and Tuesday were our strongest days.
This is likely a traffic quality pattern rather than a page issue. Monday mortgage searchers may include more casual browsers and competitive researchers. Sunday and Tuesday searchers tend to be more intent-driven.
The takeaway: If you’re making landing page decisions based on a single day's data, you may be optimizing against noise. Run any page variant for a minimum of 5–7 days across a full week before drawing conclusions.
Lesson 6: Your CPC is much lower than you think, but only if you fix the bounce
Our average CPC was $8.52. That’s actually remarkably low for VA mortgage keywords, which typically run $15–$40 per click. This was likely due to our keyword match types capturing some broader, lower-competition queries.
But effective CPC is what actually matters. When 35–52% of clicks bounce in under 3 seconds, your effective CPC on engaged sessions is significantly higher.
Here is how effective CPC changed as we improved our pages:
By reducing bounce rate from 52% to 15%, we effectively reduced our cost per engaged session by 43% — without changing a single bid or budget setting.
The takeaway: The cheapest way to improve your Google Ads ROI is often not in the campaign. It’s on the landing page.
What we would do differently from day one
If we were starting this campaign over today, here is the setup we would launch with based on everything we learned:
- Server-side rendering of rate data above the fold: Show actual VA IRRRL rates in the HTML on page load. Don’t rely on a JavaScript widget to inject rates after render. If a veteran can’t see rates within two seconds of landing, they will leave.
- Open with an identity question, not a data question: "What is your branch of service?" with large emoji buttons is a better first question than "What is your ZIP code?" or "What is your loan amount?" Identity questions feel personal. Data questions feel like a form. Design for the former.
- Test for iOS Safari specifically before spending a dollar: 84–97% of our traffic was mobile. The majority of that was iOS Safari. Tap targets, button rendering, and touch event handling on iOS Safari behave differently than on Android Chrome or desktop. Test every interactive element on an actual iPhone before launching.
- Track workflow starts, not just conversions: If your only conversion event is a completed lead form, you won’t have any data to optimize against for weeks. Track first-question answers as a micro-conversion from day one. It gives you 10x more signal to work with.
- Don’t send the same ad to four landing pages without tracking at the URL level: We used “utm_term” and “gad_campaignid” in our URLs, but iOS privacy changes caused a large portion of sessions (86.8% at one point) to lose keyword attribution. Use “utm_term={keyword}” explicitly in your Google Ads URL template. Don’t rely on auto-tagging alone.
The bottom line
We spent approximately $10,000 and ran 1,159 sessions across six weeks. Here is the single most important thing we learned:
Veterans clicking VA IRRRL ads want to see rates. If you show them rates immediately, they stay. If you don't, they leave. And you pay for the click either way.
