A true story about blended ROAS, multi-touch attribution, and the most expensive mistake business owners make with their ad budgets. Last year I had a client. Let's call him David. Not his real name. Real story. David sold a premium Aussie product. He was spending around $12,000 a month on ads. Seven on Meta, five on Google. One Monday he rang me, very pleased with himself. "Nick, I'm cutting Meta. Meta's ROAS is 1.8. Google's is 6.4. Why am I paying Zuckerberg to lose money?" It sounded like the most logical thing a human being could possibly say. My gut said don't do it. I'd seen this exact pattern half a dozen times. The healthy brand search. The suspiciously high Google ROAS. The Meta number that looked worse than it really was. It had the fingerprints of a feeder channel doing quiet, unglamorous work. I told David. He insisted. The client is the client. So I let him do it. Six weeks later, revenue had dropped 41%. Google's spend had doubled. Google's ROAS had collapsed from 6.4 to 2.9. Branded search had quietly cratered. His Shopify dashboard looked like a man holding a melting ice cream in the rain. He hadn't cut the bad channel. He'd cut the engine feeding the good one. The number that actually matters Here's the question David never asked. While his blended ROAS was sitting at 4.0, why did it matter that Meta looked weak? It didn't. That's the whole point. Blended ROAS is total revenue divided by total ad spend across every channel in the same period. That's the entire formula. It doesn't care what Meta claims. It doesn't care what Google claims. Platforms don't get a vote. The denominator is total money out. The numerator is total money in. The bank account decides. Before David cut Meta: $12,000 spend, $48,000 revenue. Blended ROAS of 5.0. After: $10,000 spend, $29,000 revenue. Blended ROAS of 2.9. If the blended number is healthy, the machine is working. Full stop. You don't need to surgically optimise the channel that looks ugliest in isolation. You need to keep the whole thing humming. Channel reporting is never 100% accurate The Singular ROI Index 2026, a global mobile ad benchmark, found that Meta campaigns measured under multi-touch attribution show up to 50% higher ROAS than the same campaigns measured under last-click. Industry overlap analysis suggests 30 to 60% of conversions across multi-channel accounts involve more than one channel touching the customer, meaning a meaningful share of sales get claimed by multiple platforms at the same time. Then Meta changed its attribution model in March 2026, redefined what counts as a click, and most accounts saw their reported numbers drop overnight. Nothing about the actual business changed. Only the dashboard did. Meta sees Meta. Google sees Google. Neither sees the customer who watched a Reel, forgot the brand name, Googled it three days later, abandoned a cart, opened an email on Sunday, and finally bought on Tuesday. If you optimise to a number that's wrong by a margin you can't see, you'll make confident decisions that destroy your business. Like David did. This is where human judgment earns its keep You can buy software that promises to fix attribution. Triple Whale. Northbeam. Rockerbox. They're useful. They're also not the answer on their own. They give you better data. They don't tell you what to do with it. The call David needed wasn't in any dashboard. It was the call that said: "Your blended ROAS is 5.0. Your brand search is climbing. Your Meta number looks bad in isolation because Meta is doing the work Google is getting credit for. Don't touch it." That call comes from having watched this exact movie play out across hundreds of accounts and knowing how it ends. This is what years of doing the job actually buys you. Not certainty. Pattern recognition. Knowing which weak-looking channels are doing real work behind the scenes, and which weak-looking channels are genuinely weak. A junior media buyer reads the dashboard and reacts. Someone who's seen the pattern reads the dashboard, ignores the obvious move, and makes the right call anyway. We turned David's Meta back on. Blended ROAS climbed to 5.7. Branded search returned. The platforms are interested parties, each selling you a version of reality that flatters its own bill. Your blended ROAS is the only number none of them can spin. And the judgment to trust it, even when one channel looks ugly, is the difference between scaling a business and accidentally dismantling one.