A search account of any size produces more information each week than anyone can read. A few hundred keywords, the search terms that actually triggered them, the ads competing inside each group, impression share lost to budget against impression share lost to rank, quality score and its three components, auction insights, and all of it changing against last week and the week before. A person can review a slice of that properly. Reviewing the whole of it every week, across several accounts, is beyond what the working day allows, which is why most accounts get looked at when something goes wrong rather than before it does.

The comparison we keep coming back to is the flight deck. Most of a commercial flight runs on automation, and no one considers this a problem, because the pilot is looking at instruments that report what the automation is doing and what conditions it is doing it in. The automation flies. The instruments make the automation legible. The pilot decides when to intervene, and retains the authority to do so at any point.

That is the shape of how we run Google Ads. Automation does the work no human can do at that volume, which is reading everything the account produced this week and surfacing what changed. The account manager reads the instruments and decides what to do.

What gets collected

Every week a script pulls the account down into a structured export: search terms over seven and thirty days, keyword inventory, campaign performance against the prior period, ads, assets, landing pages, negative keyword lists, conversion lag, disapprovals, settings, and the full change history. Nothing is summarised at this stage. The export is the raw material.

In parallel, lead outcomes come back from the CRM. Each paid lead carries its status and, where it was rejected, the reason a salesperson rejected it. Where the CRM connects to the campaign and ad group directly, that connection is preserved. Where it does not, the click identifier carries the join.

Those two sources form the basis of the weekly review, alongside the analyses Google Ads produces itself, including auction insights, recommendations and diagnostics. One describes what the account did. The other describes what the business received.

What the analysis produces

The export and the lead data are read together against the client's own context and against our playbook for the account. What comes out is a weekly reading rather than a report: week on week movement and what caused it, impression share held against share lost to budget and share lost to rank, whether the search terms that spent money bear any relation to what the client sells, which component of quality score is weak on the keywords that matter, how audiences are performing, and which paid leads were disqualified and why.

Google's own recommendations are assessed here too, on their merits. Some are worth applying. Many are optimised for Google's outcome rather than the client's, and recommendations are never applied automatically on accounts we manage.

The output arrives as an email each week, alongside a queue of specific actions and an optimisation log that records every change with the reason behind it and a date to check its effect. The log is the part clients tend to find most useful six months in, because it is the only record that explains why the account looks the way it does.

Where the human sits

The account manager receives the analysis and decides. Which negatives to add, which bids to move, whether an ad group needs splitting, whether a landing page is the problem rather than the keyword, and whether something surfaced this week is a trend or a fortnight of noise. Every change to the account is made by a person who can explain it.

This matters more in long-cycle B2B than elsewhere. An account producing fifteen enquiries a month does not generate enough data for a machine to distinguish a real pattern from a run of coincidence, and a confident automated decision made on that volume is usually wrong. Judgement is doing the work that data cannot do yet.

It also means we can say what we would not let automation do. Smart bidding needs a steady flow of conversions to learn from, and below that threshold we set bids by hand. Auto-applied recommendations stay off, because on more than one inherited account we have found Google quietly adding broad match keywords nobody chose. The pilot analogy holds in this respect as well: the automation is trusted within the conditions it was designed for, and overridden outside them.

What this is not

Programmatic advertising uses automation in a different way and toward a different end. There, the technology is buying media in real time across exchanges, covering display, video, connected TV, audio and digital out of home, making bidding decisions per impression at a speed and scale no person could match, with the human role sitting at the level of strategy, audience definition and budget. The machine is transacting.

What we are describing does no buying and makes no decisions. It reads what the account did, compares it against what the business received, and presents the result to the person responsible for the account. The buying is still Google's auction, and the decisions are still ours.

Why it is built this way

The alternative is the model most accounts run on: a monthly check, a report assembled from whatever the platform put in the conversions column, and changes made in response to whatever the client asked about. That model works while nothing is drifting. It fails silently when something is, which is the situation we find in most accounts we inherit.

Reading everything every week is the only way to catch a search term that started spending in March, an ad whose relevance dropped after a landing page change, or a conversion action that stopped firing three weeks ago. None of those announce themselves. They show up in the numbers as a gradual softening that any month-to-month comparison will absorb without comment.