VeldFire was spending R45,000/month on Google Ads with an 89% ROAS — effectively losing money on almost every rand spent. Their campaigns had 800+ keywords, most of them untargeted, cannibalising each other, and attracting the wrong buyers.
The client had tried two previous agencies. Both had added more keywords, raised bids, and produced reports that looked detailed but changed nothing. The fundamental issue was never addressed: there was no intelligent system mapping search intent to purchase intent at scale.
Their product catalogue had over 2,400 SKUs across 47 categories — tents, sleeping bags, cooking equipment, navigation tools, first aid. A human keyword strategist can manage perhaps 200–300 keywords with real precision. At 800+, they were just guessing.
We built a keyword intelligence system trained specifically on the outdoor and camping product category, South African search behaviour, and VeldFire's specific product attributes. The system didn't just find keywords — it scored intent and predicted purchase probability.
Every keyword scored on a purchase intent scale. Informational, navigational, and transactional queries separated automatically.
Automated identification of search terms that drain budget — DIY, rental, second-hand, school projects. Pruned 340 wasteful terms in week one.
Dynamic bid recommendations per keyword based on conversion probability, margin, and seasonal patterns in the outdoor market.
The engine crawls search trend data weekly. New high-intent keywords are surfaced automatically before competitors discover them.
The build took 6 weeks: 2 weeks of data ingestion and training, 2 weeks of engine development, 2 weeks of testing and calibration against live campaign data before we switched it on fully.
Full audit of existing campaign data. 18 months of search term reports processed. Product catalogue structured and loaded into the training pipeline.
Intent classification model trained. Negative keyword rules defined. Bid logic built. Integration with Google Ads API established.
Engine ran parallel to existing campaigns. Predictions validated against actual conversion data. Thresholds tuned for the SA market specifically.
Existing campaigns restructured. 340 negative keywords added. ROAS climbed from 89% to 210% in the first 6 weeks of live operation.
As the engine accumulated more data, precision improved further. ROAS reached 340% by month 10. Ongoing weekly optimisation continues.
Every rand spent on ads now returns R3.40. Previously returning less than the cost of the click.
Cost to acquire a customer fell 62% as the engine eliminated wasted spend on low-intent searches.
340 irrelevant search terms blocked in the first month alone. Budget immediately redirected to high-intent buyers.
A professional services site rebuild that turned traffic into qualified leads.
Content strategy and technical SEO for a B2B healthcare platform.
Custom AI trained on their catalogue generates 400+ product descriptions monthly.
We'll look at your current setup, identify the biggest opportunities, and tell you exactly what we'd build. No fluff.