Case Study: Reducing Cost Per Lead With ChatGPT Ads in Financial Services
How a US financial advisor matching platform took a new channel from three times its paid search cost per approved lead to below it.
80%
Reduction in cost per approved lead
Month 4
Cost per approved lead fell below the paid search benchmark
35%
Cheaper per approved lead than paid search at steady state
The Client
A matching platform in one of the most expensive lead categories there is.
The client operates a US platform that matches individual investors with vetted financial advisors. Revenue depends entirely on lead quality: a lead only counts once an advisor on the other side of the marketplace accepts it, so volume without qualification is worse than no volume at all. That single fact governs every media decision the business makes.
Financial advisory is also one of the most competitive and expensive categories in paid search. The client had spent years building an efficient search programme and knew precisely what an approved lead was worth to them. Any new channel would be judged against that benchmark, and the benchmark was hard to beat.
The Challenge
When an efficient search programme sets a bar the new channel cannot clear.
When ChatGPT Ads became available, the client wanted to test it early rather than wait for the category to fill up. The strategic case was strong. Their buyers research conversationally for weeks before choosing an advisor, asking open questions about fees, fiduciary duty and retirement timing, which is exactly the behaviour the channel reaches.
The first month contradicted the strategy completely. Cost per approved lead came in at more than three times the client's established paid search benchmark. On a straight last-click reading, the channel looked like a straightforward mistake, and most advertisers would have stopped there.
3x
Cost per approved lead against the paid search benchmark in month one.
1%
Click-through rate, in line with the client's paid social channels.
A new channel with no established benchmarks in the category.
Approved lead economics that punish volume without qualification.
An incumbent paid search programme already running efficiently.
No keyword targeting available, so existing search craft did not transfer.
What our audit found
The account was built with search logic, and nothing measured the outcome that mattered.
Two things were wrong, and only one of them was the campaign. The account had been built with search logic: tightly themed ad groups and context hints written as keyword lists. ChatGPT Ads has no keyword field, and the targeting input is a freeform description of the conversations a product belongs in. Fed a list of terms, the platform had very little to work with.
The second problem was measurement. Approved status is decided downstream, days after the click, by an advisor accepting the lead. Without that outcome wired back into reporting, the account was optimizing towards raw form fills, which in this category is the fastest way to buy expensive rubbish. The early cost per approved lead was real, but nothing in the account was set up to improve it.
Context hints written as keyword lists rather than conversation briefs.
Ad group structure imported from search, too granular for the platform.
Approved-lead outcomes not connected back to campaign reporting.
Optimization signal set to form fills rather than qualified outcomes.
The Solution
How we turned it around.
Rewriting the Targeting as Language
We rebuilt the targeting from scratch, replacing keyword lists with written descriptions of the situations a prospective client is in when advisor matching becomes relevant. A person working out whether their employer plan is enough, someone who has just changed jobs and holds an orphaned retirement account, someone approaching a decision about fees.
That meant collapsing a large number of narrow ad groups into a small number of sharply described ones. Each hint was written as a brief and revised on evidence, not expanded into more variants. The discipline is closer to audience planning than to search management, and it is the single lever with the most influence over what the platform serves.
What we shipped
- Conversation-led ad group architecture.
- Hints written as briefs and revised weekly.
- Deliberately few ad groups, each sharply defined.
- Intent stage matched to landing experience.
Wiring the Real Outcome Back In
We installed the conversion pixel, standardised URL tagging across every destination, and connected advisor acceptance from the CRM back into campaign reporting, so the account could be judged and optimized on approved leads rather than form fills.
This also changed how the channel was reviewed internally. Reporting showed trend alongside snapshot, which matters enormously for a channel that reaches people early in a decision. A last-click report would have recommended shutting the test down in week three.
What we shipped
- Conversion pixel and tagging standardised before scale.
- CRM acceptance data joined to campaign reporting.
- Approved leads set as the optimization target.
- Trend reporting alongside snapshot reporting.
Iteration, Not Patience
The cost curve did not bend because anyone waited. Every week we rewrote context hints against qualified lead data, rotated imagery and headlines within the platform's character limits, and pruned the ad groups that were buying the wrong conversations.
Budget rose only as the qualified lead rate justified it. Scale mattered, since the platform needs volume before its delivery makes sensible decisions, but scaling an account that has not been fixed only buys expensive leads faster.
What we shipped
- Weekly context hint rewrites against qualified lead data.
- Continuous copy and imagery rotation.
- Ad group pruning on qualified outcomes.
- Budget scaled behind evidence, not ahead of it.
The Numbers
Outcomes we can talk about.
Cost per approved lead fell by roughly 80% over six months. It crossed below the client's paid search benchmark in month four and settled at about 35% below it.
Both figures use the same CRM definition of an approved lead. We take total spend by source in a calendar month and divide it by the approved leads the CRM attributes to that source in the same month, rather than relying on either platform's own reported conversion count. The comparison does not depend on an attribution model, because both channels are scored the same way.
The most useful detail is what did not move. Click-through rate stayed around 1% throughout, in line with the client's paid social channels, and the on-site conversion rate barely shifted either. The gain did not come from better clicks or a better landing page. It came from reaching better qualified conversations, which is a targeting and measurement problem rather than a creative performance one.
The comparison deserves one caveat. The client's search programme is eleven years old and runs at roughly three and a half times the monthly budget of the ChatGPT account. Beating a mature, heavily worked search programme at under a third of its spend is a real result, but it is not proof the channel holds at that scale. Cost per lead usually rises as budget grows, because the best inventory goes first.
The wider lesson is about review windows. At ninety days this account was still behind paid search and would have failed a quarterly review judged on cost alone. What it had was three months of slope. Judged on the direction rather than the number, it was already working.
80%
Reduction in cost per approved lead
Month 4
Cost per approved lead fell below the paid search benchmark
35%
Cheaper per approved lead than paid search at steady state
At fourteen or thirty days the numbers said stop. The only reason this channel exists in the account today is that the review was set for ninety days before anyone spent a dollar.
- Head of Digital Marketing, Intelegencia
What We Built
What's Next
From a proven channel to a repeatable measurement standard.
The account continues to run on a weekly testing cadence, with context hints rewritten against qualified lead data as the mix of conversations shifts. The next phase focuses on first-party audience activation and on extending the same approved-lead measurement discipline across the client's other acquisition channels.
Frequently Asked Questions
About This Project
The questions teams usually ask when they want to run a similar engagement.
The account was built with paid search logic, and ChatGPT Ads does not have a keyword field. Context hints written as keyword lists give the platform very little to work with, and approved-lead outcomes were not yet connected to campaign reporting, so nothing was optimizing towards the metric that mattered.
The Real Numbers
Need real numbers? Let's talk.
We kept the names off the page. The story is real, the outcomes are real, and we're always happy to walk a serious team through the rest of it.
Start a Conversation→More from Marketing
Reducing ACoS and Scaling Sales for an Amazon Seller
Scaling B2B Lead Generation for an Electrical Equipment Exporter
Boosting Conversion: A 22% ROAS Lift for a Global Luxury Beauty Leader
Insights from our blog
Digital MarketingChatGPT Ads: Six Months In, Our Cost Per Lead Beat Paid Search
We have run ChatGPT Ads since March 2026. Month one cost over three times our paid search benchmark per approved lead. It now runs about 35% below it. Here is what moved the number.
Digital MarketingHow Generative AI is Revolutionizing Ad Creatives at Scale
Discover how AI-powered creative production helps advertising teams move faster, personalize campaigns, and scale what works.
Digital MarketingGoogle AI Overviews are Changing Product Search. Is Your Store Ready?
Want your products featured in Google AI Overviews? Explore ecommerce SEO strategies that make product information clearer, more accurate, and easier for AI to understand.