How Foxtly’s AI Found a $3,200/Month Leak in 2 Minutes (And What It Told Us About SMB Marketing)

Last February, a DTC skincare brand founder reached out to Foxtly. Her agency had been managing her Meta Ads for 14 months. Spend: $18k/month. Results: stagnant. ROAS oscillating between 2.1x and 2.6x despite the agency’s monthly “optimization reports.”

She wanted a second opinion. We plugged Foxtly into her account. Two minutes later, the AI flagged six distinct issues, one of which was silently costing her $3,200 every single month.

This is the case study. Numbers are real; brand name changed for privacy.

The Setup

• Brand: “LuminaSkin” (DTC skincare, 4 SKUs, $2.8M ARR)
• Ad spend: $18,000/month on Meta, $4,000/month on Google
• Agency: $3,200/month retainer for managing both
• ROAS trend: 2.3x blended, flat for 9 months
• Goal when agency was hired: Scale to $5M ARR within 18 months

From the founder’s perspective: “We’re spending, agency is pushing buttons, but nothing’s changing.”

What the Agency Said

Monthly report (paraphrased):

“This month we optimized bid strategies, refreshed 3 creatives, and tightened audience targeting. ROAS is stable at 2.4x. We recommend a $2k/month budget increase to scale further.”

Business as usual. Agency not lying, but also not finding the fire.

What Foxtly’s AI Found (In 2 Minutes)

We connected Foxtly to Meta, Google, Shopify, and Klaviyo. The AI ran a diagnostic across the full account. Here’s what it surfaced:

Issue 1: Audience Overlap (The $3,200 Leak)

The agency had set up 9 active ad sets. Foxtly’s AI cross-referenced all audience definitions and found:

• Ad Set A: “Lookalike 1% based on purchasers, 180-day”
• Ad Set B: “Lookalike 1% based on purchasers, 365-day”
• Ad Set C: “Custom audience: Website visitors 30 days + Interests: Skincare”
• Ad Set D: “Lookalike 2% based on purchasers + Interests: Clean beauty”

Meta’s Audience Insights tool showed these 4 ad sets had 68% audience overlap. They were auctioning against each other on the same 420k people.

Impact quantified by Foxtly:

• $7,800/month spent in overlapping auctions
• Estimated 40% of that spend = inflated CPM from bidding against self
• Waste: $3,120/month

The agency had built this structure 14 months ago and never re-audited. No human intuition caught it. An AI running cross-audience overlap analysis caught it in 30 seconds.

Issue 2: Broken CAPI

Foxtly’s AI noticed the Event Match Quality score was 4.1 (should be 7+). Root cause: the Conversions API integration had broken 6 weeks prior during a Shopify app update. Nobody noticed.

Impact: Algorithm was matching only 38% of conversions to users. Meta was essentially blind to 62% of conversion signal. Optimization was running on bad data.

Fix: reconnect CAPI, improve match key fidelity. Expected lift: 15-25% reported conversions.

Issue 3: Creative Fatigue

AI noticed 4 of 11 active creatives had frequency above 4.2 and CTR declining week-over-week for 5+ weeks.

The agency had launched “3 new creatives this month” — but 3 per month doesn’t outpace fatigue at this spend level. Required: 8-12 new creatives monthly.

Issue 4: Wrong Bidding Strategy

All ad sets were on Lowest Cost bidding. Given LuminaSkin’s AOV distribution (60% orders between $40-$70, 30% between $70-$120, 10% above $120), the algorithm was optimizing for the cheapest customers — lowering AOV over time.

AI recommendation: switch to Value-Based Optimization (VBO) to find higher-LTV buyers.

Issue 5: Landing Page Speed

AI ran Core Web Vitals check on top landing pages. Mobile LCP was 4.2 seconds (should be <2.5). Root cause: a third-party review widget loading synchronously, blocking page render.

Estimated impact: 15-20% conversion rate loss from slow load.

Issue 6: Attribution Window

Meta was reporting on 1-day click attribution (Meta changed defaults in late 2025). Brand’s actual purchase cycle: 4-12 days from first click. Reporting was severely under-counting.

Fix: adjust attribution to 7-day click + 1-day view. Immediate ROAS reading lifted 28%.

The Prioritized Fix Plan

Foxtly’s AI ranked the 6 issues by impact x effort:

1. Fix audience overlap: 1 hour, $3,120/mo savings → do first
2. Reconnect CAPI: 2 hours, 15-25% conversion lift → do second
3. Fix attribution window: 5 minutes, reveals true ROAS → do immediately
4. Launch more creatives: ongoing commitment, fixes fatigue
5. Switch to VBO: 30 minutes, long-term LTV lift
6. Fix landing page speed: 4-6 hours dev time, CRO lift

Founder executed 1-3 herself that week. Asked us to help with the ongoing creative and technical work.

Results After 90 Days

Before fixes:

• Monthly spend: $18,000
• ROAS: 2.3x
• Monthly attributed revenue: $41,400

After 90 days with Foxtly managing:

• Monthly spend: $18,000 (same)
• ROAS: 3.7x
• Monthly attributed revenue: $66,600
• Additional revenue per month: $25,200

And critically: she fired the agency. Net savings: $3,200/mo retainer + $3,120/mo waste from overlap + $22,000/mo additional attributed revenue from fixes. Total impact: ~$28k/month.

What This Case Study Reveals

This wasn’t a bad agency. They were doing what most agencies do: monthly check-ins, some creative, some bid adjustments. They just missed the big structural issues because:

1. They audited monthly, not continuously. Issues compound between checks.
2. They lacked cross-platform AI pattern recognition. Overlap analysis isn’t human-scalable across 9+ ad sets.
3. They weren’t incentivized to look for big leaks. Agencies are paid to “manage,” not to audit themselves out of a job.
4. They were under-invested in technical monitoring. CAPI break detection, Quality Score alerts, landing page monitoring — these take infrastructure.

The fundamental shift: continuous AI monitoring catches what monthly human reviews miss. Not because AI is smarter, but because it never stops looking.

Common Pattern: The Hidden Overlap Problem

After running Foxtly diagnostics on 400+ SMB ad accounts, audience overlap is the #1 most common major issue. 67% of audited accounts had >25% overlap on spending audiences.

Why it persists:

• Agency creates structure on setup, rarely revisits
• Overlap gradually increases as audiences grow (a 180-day lookalike grows to overlap with 365-day)
• Nobody checks the overlap tool because it’s buried in Audience Insights

Estimated aggregate waste across audited accounts: $1.2M/month in unnecessary self-bidding.

Common Pattern: Broken Tracking

Second most common: broken or degraded CAPI. 41% of audited accounts had Event Match Quality scores below 6.0. Of those, 28% had CAPI that had actually stopped firing entirely due to app updates, domain changes, or cert expirations.

Consequence: algorithms running on partial data. Inflated CPLs. Scaling ceilings that shouldn’t exist.

What You Can Do Today

You don’t need Foxtly to do this. Any disciplined operator can:

1. Weekly audience overlap check: Go to Meta Ads Manager, use Audiences tool, check overlap between top-spending ad sets. Any pair >25% = consolidate.

2. Monthly CAPI health check: Events Manager → Data sources → Your pixel → Event Match Quality. Target: 7+.

3. Weekly creative fatigue check: Sort active ads by spend, check frequency. Anything above 3.5 with flat or declining CTR → rotate out.

4. Quarterly landing page speed check: Google PageSpeed Insights, check mobile LCP on top pages. Fix anything above 3s.

5. Monthly attribution window review: Ensure you’re reporting on a window that matches your buyer behavior (not whatever default Meta most recently changed to).

Taken together, these 5 checks would catch 70-80% of the issues we find in AI audits. They’re free. Most SMBs don’t do them because nobody’s scheduled it.

The Broader Implication

This case study isn’t unique. It’s representative. We’ve run similar diagnostics on 400+ SMB accounts with strikingly consistent patterns:

• 67% have material audience overlap
• 41% have broken/degraded tracking
• 85% have creative fatigue issues (<5 new creatives/month at relevant spend)
• 53% have landing page speed issues
• 34% are on wrong attribution window

Most of these are fixable in <10 hours of work. Average impact when fixed: 25-50% CPA reduction or ROAS lift.

The point isn’t “hire Foxtly.” The point is: continuous monitoring catches what periodic human reviews miss. Whether you buy software to do it or discipline yourself to audit weekly, do something.

What Didn’t Happen

Notable: nothing in this case study involved fancy creative strategies, secret audience insights, or breakthrough optimization hacks. It was all fundamentals. Audience overlap. Tracking health. Creative pace. Page speed. Attribution accuracy.

Boring fundamentals, applied rigorously. That’s the game.

Final Word

Most SMBs running ads have a $2k-$5k/month leak they haven’t found yet. They might have great creative. They might have the right target audience. But somewhere in the account, money is evaporating silently.

Finding the leak is 80% of the value. Fixing it is the other 20%. Most operators — and most agencies — skip the finding step.

If you want the same AI audit Foxtly ran on LuminaSkin’s account, it’s free for 7 days. If you’d rather audit yourself using the 5 checks above, that also works.

Either way, look. Audits expose things that neglect hides.