The choice is usually framed as old versus new, or manual versus automated. Framed that way it has an obvious answer and a poor outcome.

The useful framing is about information. Every step towards automation trades control and reporting detail for reach and, sometimes, efficiency. Whether that trade is good depends entirely on whether you would be able to tell when it goes wrong.

What you are actually choosing between

A search campaign is a set of explicit instructions: these queries, these audiences, this ad, this landing page, this budget. You can see which query produced which click, and you can act on it.

Performance Max is a budget plus a set of assets and a goal. The system decides placement, format, audience and query mix across the whole inventory - search, shopping, display, video, mail, maps. You see far less of what it did.

The efficiency case for automation is real, particularly where the signal is strong and the inventory is broad. The cost is that when results degrade you have fewer levers and less evidence about which one to pull.

The four things you give up

  1. Query-level visibility. You lose the ability to see every term that triggered a click, which is how you find both the negative keywords and the product ideas. Reporting exists but is aggregated and partial.
  2. Channel separation. Search, shopping and display performance arrive combined. A campaign that looks efficient may be efficient on brand search and wasteful everywhere else, and the average conceals it.
  3. Brand and non-brand separation. This is the big one. Automation will happily spend on people who were already searching for you by name, book the conversion, and report an excellent cost per acquisition for traffic you would have had for free.
  4. Creative control. Assets are recombined into formats and placements you did not design. For most advertisers this is acceptable; for some brands it is not, and it is better to know before the campaign runs.

What has to be true before automation works

Automated bidding optimises the signal it is given. If the signal is wrong, automation makes the wrong thing happen faster and with less visibility. Three conditions, all worth checking before spending.

  • Conversions are tracked correctly and are not duplicated. A conversion counted twice teaches the system to buy more of it.
  • Conversions carry values, not just counts. Otherwise a newsletter signup is worth the same as a sales enquiry, and the cheap one wins.
  • There is enough conversion volume to learn from. Below roughly a few dozen conversions a month, automation is guessing, and a manual campaign with a human watching it will do better.

If any of those is false, the sequence is to fix measurement first. That is not a delay before the real work - it is the work that decides whether the spend does anything.

A structure that keeps the evidence

The version that tends to work is not either-or.

Keep a brand campaign separate, always, so brand traffic cannot be counted as acquisition. Keep a search campaign on your highest-intent terms, where you want the query data and the landing page control. Then run automation for discovery on top, judged on incremental volume rather than on its own reported cost per acquisition.

Then add negative keywords at the account level, so automation cannot spend on terms you have already decided are wrong. This is the single most useful lever that survives the move to automation, and it is frequently left unused.

What honest reporting looks like

Spend, qualified leads or revenue, cost per outcome, brand and non-brand separated, and what was changed this period with the reasoning. Not a platform screenshot, and not a blended cost per acquisition that mixes the two.

And no promised numbers. Auction costs are set by competitors, demand and seasonality, none of which any agency controls. What can be committed to is the work, the measurement, and telling you early when a target is not achievable in your market.