From break-even to 2.08 ROAS in 42 days.
Weekly blended ROAS, pulled directly from the ad account. It crossed break-even in week three and kept compounding as the governance system fed cleaner data into bidding.
Spending hard, earning nothing
$80K–$150K/month against the world's largest aggregators, with ROAS at 1.02 and no negative-keyword governance.
Restructure + query governance
Full rebuild around booking intent, a custom AI classifying every search term daily, and value-over-cost bidding.
2.08 ROAS, owned by the client
ROAS doubled, CPA down 41%, and a 1,800-term negative list plus full change log handed over.
01 The problem
FlightSearchDirect — a flight comparison and meta-search platform — was competing head-to-head against some of the largest travel aggregators in the world on Google Search. Despite budgets of $80K–$150K/month, ROAS had fallen to 1.02, meaning the account was barely breaking even on every dollar spent. The fundamental issue was keyword targeting quality: campaigns were triggering on large volumes of irrelevant and low-intent search queries — non-flight terms, informational searches, and competitor brand searches with no commercial intent. With no negative keyword governance and no system for continuously classifying incoming search terms, wasted spend compounded daily.
Three moves that reversed the curve.
No bigger budget. We fixed the foundation, automated the governance, and pointed bidding at profit.
Rebuilt around booking intent
Every campaign restructured into tightly themed ad groups by route type, airline brand and booking intent — broad-match waste cut entirely.
Custom query-governance system
An automation layer classified every search term daily, promoting converters and pushing waste to negatives, each tagged 'Added by AI'. Negatives grew 200 → 1,800+.
Value-over-cost bidding
Switched to conversion-value bidding with hard CPA floors, so the algorithm optimised toward profitable bookings, not raw volume.
Measurement rebuilt first
Conversion tracking was rebuilt to count completed bookings rather than clicks — so every optimisation rested on a real number.
Same budget. Different account.
Every figure measured from the live ad account at the start of week one and the end of week six.
03 The results
ROAS recovered from 1.02 to 2.08 — more than doubling the return on every dollar spent. CPA reduced 41%. The negative keyword automation system was the most impactful change: within 60 days of deployment, wasted spend on irrelevant queries fell by 67%, releasing that budget for high-intent terms. The account went from barely breaking even to generating meaningful margin on every campaign.
Six weeks. Forty-two days.
Discovery → measurement → execution → compounding.
Audit & measurement
Pulled 90 days of search-term data, identified 4,200+ wasted query types, and rebuilt conversion tracking to count bookings.
Restructure
Killed the broad-match campaign and built tightly themed ad groups by route and intent. First 600 negatives shipped.
Automation goes live
Query Governance deployed — daily classification of every term, each change tagged 'Added by AI' with an auditable log.
Bid logic switched
Value-over-cost bidding with CPA floors. The algorithm finally had clean data to optimise against.
ROAS crosses 2.0
From 1.02 to 2.08 ROAS, CPA down 41%, and a 1,800+ negative list that keeps growing.
What the client kept.
Most agencies keep their logic opaque on purpose. We hand it over — documented, portable, and owned by the client.
- ✓47-page change logEvery optimisation, dated and explained.
- ✓1,800+ negative keywordsThe full governed list, exportable.
- ✓Query-classification rulesThe logic behind every 'Added by AI' action.
- ✓Rebuilt conversion trackingCounting bookings, not clicks.
- ✓Restructured campaign blueprintAd groups by route and booking intent.
- ✓Value-based bidding configCPA floors mapped to real margin.
04 Why it worked
An account spending $80K–$150K/month at 1.02 ROAS is a burning building. The AI search term classification system was what made recovery possible at that scale — manual management simply cannot keep pace with the volume of new queries at $150K/month. Building an automated system that auditably classifies and acts on every search term daily is the difference between an account that drifts and one that improves continuously.
They handed us a 47-page log of every change they'd made. No other agency has ever done that. The query-governance system is now part of how we think about Search.