Tuning similarity percentiles to balance reach and conversion rate
Lookalike audiences are modeled on a seed audience (usually your customer list or past purchasers) and Google expands them by finding people with similar characteristics. The similarity dial is expressed as percentiles. A 1% lookalike is the tightest match, finding only the people most statistically similar to your customers. A 10% lookalike is broader, including anyone in the top decile of similarity. The tradeoff is predictable: tighter similarity means higher conversion rate but smaller reach. Broader similarity means more volume but declining conversion.
Typical conversion rate curves show a 1% lookalike performing 2-3x better than a 10% lookalike on the same campaign. But 1% might reach only 100,000 people total, while 10% reaches 2 million. If your daily budget is $5,000, you'll exhaust the 1% quickly and still have budget left, so you naturally layer in broader audiences. The question is where to stop.
Layering multiple similarity tiers into a single strategy
Best practice is running 1%, 2-3%, 4-6%, and 7-10% as separate campaigns with decreasing bids. Give 1% your highest bid cap (since it converts best) and ladder down. Monitor conversion rates across tiers for 2-4 weeks. As budget exhausts 1%, it naturally flows into 2-3%, which is still efficient. You're essentially sorting reach by quality. The contribution from 7-10% will be lower quality, but it captures incremental volume that wouldn't convert at 1-3% prices.
A common mistake is running all lookalikes under a single campaign at the same bid. This means Google's algorithm treats them equally and may allocate budget to cheaper, broader tiers first. Separate campaigns let you control the tradeoff explicitly. If conversion rate at 8-10% is half that of 1-3%, you can bid accordingly and reject impressions that don't meet your threshold. This explicit control compounds efficiency over months.