​

How Erdo bids

Google’s automated bidding builds a model of who converts, and that model needs conversions to build from — roughly thirty a month for each thing that bids independently. A residential development running six search campaigns rarely gives any one of them that. One account Erdo runs produces sixty-one conversions across thirty days between its six campaigns, and between two and seventeen each: six bidders, none of them with enough evidence to bid well. So Erdo’s campaigns do not bid one at a time. It puts an organization’s campaigns on a portfolio bidding strategy — one strategy they share, whose model is built from all their conversions together. Six campaigns with sixty-one conversions between them become one bidder with sixty-one, which is a bidder that can actually learn.

​

Where you will see it

You do not configure it. New campaigns launch into the organization’s strategy, and existing ones are moved into it when Erdo proposes promoting them off click-based bidding — both as approval cards you answer, like every other change that moves money. The card names the strategy, and when the change alters the strategy’s own target rather than only the campaign in front of you, it says how many campaigns that reaches:

Retarget portfolio strategy ‘Erdo 2200-brickell — Leads’ to $85.00 CPA — affects 6 campaigns in account 8834039525

That sentence is the point of the card. A target on a shared strategy is a decision about every campaign in it, and a card that named only the new number would be asking you to approve something five campaigns wider than it described. A card that moves campaigns names them, rather than counting them, and it always states the target the campaign will actually be bid to. Joining a pool that already has a target inherits that target, so a card for a campaign joining an $85 pool says$85 even when the change itself sets nothing. In your Google Ads account the strategy appears under Tools → Shared library → Bid strategies, named after your organization.

​

What it pools, and what it never pools

It pools the conversions of your campaigns, and only yours. Some Erdo customers still share one legacy Google Ads account with other customers — a bid strategy lives at the account level there, so “pool these campaigns” and “pool two businesses” are one mistake apart. Pooling two businesses would not be a smaller version of the same benefit: it would build one model from two sets of buyers, optimise toward a cost per lead neither business is paying, and leave each one bidding on the other’s evidence. Five things keep that from happening, and none of them is a setting anyone can get wrong:

• Every call is pinned to the account your connection selected, so a call naming another account is refused before it reaches Google.
• The strategy’s name is derived from your organization and is never something a call supplies — it carries your organization’s own identifier, not just its name, so a name freed up by a closed account can never resolve back to the campaigns that used to bid through it. There is no way to name another organization’s strategy, so there is no way to reach one.
• Before anything joins a strategy, Erdo checks that the one it found is really yours — the name and the type both have to match. Anything else is refused with what it found, rather than attaching quietly or building a second pool beside the first.
• Before a campaign is moved into your strategy, Erdo checks the campaign as well as the strategy. A campaign already bidding through another Erdo customer’s strategy is refused by name, and so is a campaign the account does not hold. If that check cannot be completed, nothing is moved and nothing is retargeted — the change is reported as retryable rather than made on an unverified reading.
• A change that moves the shared target says how many campaigns it affects, on the card, before you answer.

Your campaign budgets stay individual throughout. Sharing a bid strategy asks nothing of the budgets, and a shared budget is in fact the thing Google refuses to combine with conversion bidding.

​

The learning window

A bidding change resets what Google knows. For about a fortnight afterwards the numbers describe a model rebuilding itself, not a campaign performing — and a cost per lead read inside that window is not evidence about anything. Erdo’s monitoring holds every price lever for fourteen days after any bidding change: no bid increase, no bid-ceiling raise, no budget increase on the affected campaigns or on anything sharing their strategy. That includes a brand-new campaign, which starts its bidder from nothing. The fortnight is counted from the moment the change reached Google, not from when it was proposed — a card you answer a week after it was raised starts the window on the day you answered, and a card you decline starts no window at all. Decreases, pauses, negative keywords, copy work and keyword work all continue — the hold is one-directional, and it is the same shape as the hold that applies while a campaign’s search terms are the binding problem. Each run says which proposals the window held and when it lifts, so a quiet run never has to be read as “nothing was worth doing”. Once the window closes and the pool has enough conversions to mean something, Erdo proposes a target cost per conversion — taken from what the pool actually achieved since the change, not from a number picked in advance — and tightens it by at most 15% every fortnight, and only while the lead volume holds. If the change is plainly failing rather than learning — cost per conversion more than half again what it was, on well under half the clicks — Erdo does not wait out the window. It proposes a target immediately, cites the decision it is responding to, and puts both pairs of numbers on the card so you can see why it intervened early.

​

Related