Bid to a target you can defend, then leave the machine alone

Once your tracking is honest, the smartest thing a lean founder can do is set a target cost-per-acquisition the business can defend, hand the bidding to Google's algorithm, and stop touching it.

What smart bidding is

Smart bidding is Google's machine-learning system that sets each individual bid in real time, using signals you cannot see or react to manually — device, time, location, query nuance, audience. You give it a goal, either a target CPA (what you will pay for a conversion) or a target ROAS (the revenue return you want), and it optimises every auction toward it. For a solo operator this is not a convenience, it is a genuine edge, because no human can adjust bids per auction at that granularity.

Why "defendable" is the key word

The target you set must come from your actual unit economics, not a hopeful number. Work backwards: if a customer is worth, say, 6,000 EUR over their lifetime and one in five qualified leads closes, then a qualified lead is worth roughly 1,200 EUR, and you can defend paying a few hundred for one. Set the target too low and the algorithm starves itself of volume; set it on a vanity metric like cheap clicks and it optimises for the wrong thing. A defendable target is one you can justify with the numbers behind it.

The exact steps

1. Gather enough conversion data first. Smart bidding needs signal; aim for at least 15 to 30 conversions in 30 days before switching from manual or maximise-clicks.
2. Calculate your defendable target CPA from lifetime value and close rate, then set Target CPA (or Target ROAS if you track revenue values).
3. Feed it the right conversions. It only learns from what you count, so make sure the qualified-pipeline conversions from the previous chapter are the ones it optimises toward.
4. Give it room. Leave changes alone for at least two to three weeks after a switch; the algorithm enters a learning phase and reacts badly to constant edits.
5. Adjust the target, not the bids. When you want more or fewer leads, move the target CPA gently (10 to 20 percent) and wait, rather than reverting to manual bidding.

Worked example

A 16-person B2B fintech ran manual bids and spent the founder's evenings nudging them. Leads cost about 280 EUR and the numbers were noisy. They calculated a defendable target — a closed customer was worth 9,000 EUR, close rate one in six, so a lead justified up to 250 EUR — set Target CPA at 220 EUR, fed it CRM-qualified conversions, and stopped touching it. After a three-week learning phase, cost per qualified lead settled at 190 EUR and monthly qualified leads rose from 8 to 14, with the founder reclaiming the evenings.

Pitfalls

• Switching to smart bidding with no conversion data. It cannot learn from nothing; build up signal first.
• Fiddling daily. Constant edits reset the learning phase and keep performance unstable.
• A fantasy target. A CPA your economics cannot support either starves volume or loses money on every lead.

Handoff

The machine now runs the auctions, but an account is never finished. Treat the account as a system you improve, never a thing you set up — that is next.