Now you can see where the money went, and still not what it bought

A regional window and door installer moved its whole Google budget — about £14,000 a month, spread across four Search campaigns and a Shopping campaign — into a single Performance Max campaign in February. Six weeks later the account looked better than it had ever looked: cost per conversion down 31%, conversions up by half, the graph pointing the right way. The sales director asked the only question that mattered at the March meeting, and nobody in the room could answer it. Which of those conversions put a job on a van?

Answering it took an afternoon, two exports and a spreadsheet. Of the 612 conversions Google counted that quarter, 388 were enquiries that actually reached the office. 141 of those became a survey booking. 39 became signed work, worth £287,000 against £42,000 of spend. That is a good quarter by any measure a business uses. No report in the Google Ads account said so, in either direction. The account said 612, and 612 is not a number anybody can spend.

One campaign, every surface, no keywords

Performance Max is a single campaign that buys across all of Google's inventory at once — Search, Shopping, YouTube, Display, Discover, Gmail and Maps — steered by conversion goals rather than by keywords or placements. You supply assets, feeds, audience signals and a target; the system decides where each impression is bought.

This is a real bargain, and it is worth stating fairly before criticising it. Buying seven surfaces separately means seven budgets, seven bid strategies and seven sets of assumptions about which one deserves the next pound — assumptions made monthly by a person who cannot see the cross-channel effect. Performance Max makes that decision continuously, per auction, using signals no advertiser has. Accounts that consolidate usually do see the cost per conversion fall, and it is not a trick.

The bargain has a price, and the price is epistemic. You have handed over the allocation decision, and with it the ability to answer questions in the form which channel produced this. For three years you could not even ask where the money went; the campaign reported as one thing because it was one thing.

• No keywords means no search terms in the shape you are used to, and no keyword-level bid.
• No placements means an impression on YouTube and an impression in Gmail arrive in the same total.
• No per-channel budget means the split is an outcome you read, not an input you set.
• One conversion column means every goal you enabled is being optimised toward at once, weighted by the values you gave them — including the values you never got round to setting.

Performance Max does not hide where your money went out of malice. It hides it because you sold the allocation decision in exchange for a lower cost per conversion, and reporting followed the decision.

The channel report arrived, and it answers a different question

The channel performance report, rolled out through 2026, splits a Performance Max campaign across seven named channels with impressions, clicks, interactions, conversions, conversion value and cost. It closes the where-did-it-go gap completely, and it leaves the what-did-it-buy gap exactly where it was.

Take the improvement seriously. An advertiser can now see that 62% of a campaign's spend went to Search, that YouTube took 19% and produced conversions at four times the cost, that Discover was quietly absorbing a tenth of the budget. Those are actionable facts, and for three years they were unavailable at any price. Agencies that spent 2023 and 2024 arguing about a black box were arguing about something real, and it got fixed.

But a channel split is a split of the same number the installer could not use. Conversion value in that report is the value of the conversions Google counted, at the values the account assigned, subject to the windows and the modelling that produced them. Splitting a number seven ways does not make it a revenue number. It makes it seven numbers with the same provenance.

Channel transparency solved the 2023 complaint. The 2026 complaint is that a conversion is still whatever you told Google to count, and most accounts told it something they no longer remember choosing.

Search themes steer, they do not point

Search themes are phrases you give Performance Max to widen what it will bid on. They are additive: they add queries the system might not have predicted from your assets and landing pages, and they subtract nothing. Exclusion is a separate mechanism entirely.

The practical consequence catches people who came from Search campaigns, where the keyword list was both the accelerator and the brake. Here the brake is a different pedal: negative keywords and brand exclusions. An account that carefully curates fifteen search themes and never touches negatives has tuned the accelerator and left the brake alone, and it will keep paying for the branded searches it was going to win anyway.

Branded traffic is the specific trap, because it is where the flattering numbers come from. A campaign that hoovers up people already searching your company name will show a magnificent cost per conversion and a return on ad spend that makes the channel look like a licence to print money. Those customers were coming. Brand exclusions exist so you can find out what the campaign does without them, and the month you switch them on is the month the real numbers appear — lower, and worth having.

The bidding is exactly as good as the conversion you feed it

Performance Max optimises toward the conversion actions you marked as primary, weighted by the values you gave them. Every distortion in that definition is amplified, because the system will faithfully buy more of whatever you said was good.

Open the conversion actions in any account that has been running for three years and read the list as if you were seeing it for the first time. There is usually a form submission, which is real. There is often a phone-number click, which is a person tapping a number and not necessarily a person speaking to anybody. There is sometimes a PDF download, a chat-widget open, a scroll depth, a visit to the contact page. Each was added by somebody with a reason, and collectively they define what the bidding believes success is.

The window installer's account counted four things, and two of them were the tap-to-call and the contact-page view. Those two produced the majority of the 612. The campaign had been dutifully buying more of the audience that taps numbers and reads contact pages, which overlaps with, but is not the same as, the audience that pays for eleven windows.

Values make it worse before they make it better. An account that sets every lead to a flat £50 tells the system that all leads are interchangeable; if commercial enquiries close at eight times the value of domestic ones, the bidding has been told explicitly not to care. Value-based bidding is the right answer, and it only works if the values come from somewhere true — which means they come from the closed deals, which means somebody has to do the reconciliation first. The bidding cannot bootstrap itself out of a definition you invented in a meeting.

Smart bidding is a very fast optimiser pointed at whatever you called a conversion. Get the definition wrong and it will make the wrong thing happen more efficiently every week.

Offline conversions are the fix, and they come with a clock

Uploading closed deals back into Google Ads is the intended solution, and it genuinely improves bidding. It is also bounded by hard import deadlines: ninety days from the last click for a GCLID-based upload, sixty-three for enhanced conversions for leads.

For an e-commerce purchase, ninety days is luxurious. For anything sold by a person it is the middle of the process. A conservatory quoted in April and signed in August is a straightforward, well-run sale with a delighted customer at the end of it, and it is four months old — the deal that best proves the campaign works is precisely the deal the campaign will never be told about. Every business with a considered purchase runs a version of this: the faster half of your revenue teaches the bidding, and the slower, larger half silently does not.

1. Decide what a real conversion is, in the language your sales team uses, and mark only those as primary.
2. Capture the click identifier at the point of enquiry, or set up enhanced conversions for leads so the match runs on hashed customer data instead.
3. Upload on a schedule — daily is the norm — and assign the real value, not a placeholder.
4. Accept that anything closing beyond the window will never appear in the platform, and report it somewhere else.
5. Reconcile everything, in full, outside the ad account, so the sales the window excluded still count for the channel that earned them.

That last step is what CloseRev does. It takes a closed-sales export from whatever records your revenue and a lead export with its sources, normalises the phone numbers and email addresses, matches at the confidence each piece of evidence actually supports, and reports revenue by channel — with the share it could not match shown as Direct / Unknown rather than quietly spread around. No tag, no pixel, no window: a sale that closed in August against an April enquiry is credited to the April enquiry, because that is what happened.

What to put in the monthly report

Report the channel split from Google for allocation questions, and revenue per channel from your own records for value questions. Keeping the two apart is what stops a good campaign being cancelled and a bad one being doubled.

The fourth row is the one people want to delete, and it is the one that makes the other three trustworthy. A report that attributes 100% of revenue to channels is a report that has guessed somewhere. Saying that 34% of revenue could not be matched to any recorded lead is an invitation to go and find out why — usually it is a receptionist writing names into a book, or a sales system where the source field is optional, and both are fixable in a fortnight.

The window installer kept Performance Max. They also switched off the tap-to-call and contact-page conversions, set commercial enquiries to their real average value, turned on brand exclusions, and started reconciling closed jobs against lead sources monthly. The Google account now reports about 180 conversions a quarter instead of 612, and cost per conversion looks four times worse. Revenue attributed to the channel went up, because they can finally see it. The graph pointing the wrong way was the price of the report being true.

A campaign you cannot steer per channel is not a problem if you can still tell what it sold. Google fixed the first half of the black box in 2026; the second half was always yours to fix, and it lives in your own exports.

Questions people actually ask

What does the Performance Max channel performance report actually show?

It breaks a campaign's impressions, clicks, interactions, conversions, conversion value and cost across seven channels — Google Search, the Display Network, YouTube, Discover, Maps, Gmail and Search partners — and lets you segment by channel, ad format, conversion category and conversion action. It tells you where the budget went. It does not tell you which of those conversions became money, because Google only knows about the ones your site or your uploads told it about.

Can you control how much Performance Max spends on each channel?

No. Allocation is decided in real time by the bidding, and Google's own documentation describes a channel showing no spend as the budget being prioritised for more efficient channels elsewhere. You influence the split through assets, feeds, audience signals and exclusions, and you read the outcome afterwards in the channel report. There is no per-channel budget to set.

Do search themes work like keywords?

They rank like phrase and broad match keywords for prioritisation, but they are additive rather than restrictive: a search theme adds queries Performance Max might not have predicted from your assets, feeds and landing pages, and it removes nothing. Up to fifty per asset group. If you want traffic excluded, that is negative keywords and brand exclusions, not a theme you left out.

How long do you have to import an offline conversion into Google Ads?

Ninety days from the associated last click for a conversion uploaded with a GCLID, and sixty-three days for enhanced conversions for leads, which match on hashed customer data instead. Uploads later than that are not imported at all. Any sales cycle longer than three months therefore closes outside the window, and the deal that proves the campaign worked never reaches the account that ran it.

Why does Performance Max report more conversions than the CRM has leads?

Usually because the conversion actions being counted are not all sales-relevant — page views, chat opens, PDF downloads and phone-number clicks counted alongside real enquiries — and because some of the count is modelled rather than observed. A conversion in Google Ads is whatever you told it to count, weighted by whatever value you gave it, and most accounts told it something looser than they remember.

How do you prove Performance Max return on ad spend if the platform cannot?

Reconcile outside it. Export the closed sales from the system that records revenue, export the leads with their source, match them on phone number and email address, and report revenue by channel with the unmatched share shown honestly. That number carries no attribution window, no modelling and no dependence on a tag firing, and it is the only one a finance review survives.

Found this useful?

• Share on X
• Share on LinkedIn
• Share on Reddit
• Share on Email