I sat on a monthly review call in June where the agency had prepared 41 slides.
Reach was up. Impressions were up. There was a word cloud. Somewhere around slide 30 the founder asked how many of last month’s 62 leads his sales team had actually phoned, and the room went quiet in that particular way where everybody suddenly needs to check something on another screen.
Nobody knew. Not the agency, not the founder, not the sales head who was sitting right there.
Fourteen months of retainer, and the single most basic question in the business had never been asked out loud.
Here’s my honest read on this, and it’s less satisfying than the usual take. That report isn’t evidence, and the reason isn’t that your agency is lazy or dishonest. The platforms genuinely stopped showing them things. A real part of every number on that PDF is now an estimate produced by Google’s own models, and neither you nor the person presenting can pull the estimate apart from the measurement.
Which changes what you should be testing them on. Stop grading your agency on whether the numbers went up. Grade them on whether they’ll tell you what they cannot see.
The report got prettier exactly as it got less knowable
Two examples, both from Google’s own documentation rather than from me being cynical about agencies.
Take Performance Max. You can pull a placement report showing where your ads appeared. Google’s help documentation says plainly that this report is not intended for evaluating campaign performance, and that it does not include data such as clicks or conversions. So you can see that your ad ran on some parked domain in the display network. You cannot see whether anything happened when it did.
Then there’s modelled conversions. When a user declines cookies, Google models the conversion rather than losing it, and Google’s own blog has put the recovery at more than 70% of ad-click-to-conversion journeys that consent choices would otherwise have destroyed. Genuinely useful. The catch is that you can’t segment observed conversions from modelled ones inside Google Ads. They’re added together and presented as a single figure.
Sit with that one, because it’s the whole argument.
When your agency reports 62 conversions last month, some unknown share of those 62 is a model’s best guess, and NOBODY in that meeting can tell you which ones.
I’m not saying the modelling is wrong. It’s probably closer to the truth than throwing the unconsented journeys away would be. But a number you cannot take apart is a reading, not a proof, and it should be presented that way.
Most reports present it as a proof.
The agency that explains everything is the one to worry about
So what does a good agency actually sound like in 2026?
Less certain than a bad one.
An agency with a tidy explanation for every wiggle in the graph isn’t analysing your account, it’s narrating it. Any of us can tell a convincing story about a line going up once we already know it went up. That costs nothing and proves nothing.
The tell is what they do when a number moves and there’s no clean reason for it. The operators I rate say things like “leads dropped about a fifth, I’ve got two theories, I’m testing the cheaper one this week and I’ll know by Friday.” The ones coasting say “we observed a seasonal dip and have optimised accordingly,” which is a sentence built to survive being quoted back at them.
Certainty got cheap at precisely the moment measurement got expensive. That’s worth remembering the next time somebody sounds very sure on a call.
Four questions for your next review
Each of these can be answered in one sentence by somebody doing real work, and cannot be answered at all by somebody who isn’t.
- “Of last month’s conversions, how many were observed and how many modelled, and if you can’t split them, what do you check instead?” The right answer admits the split isn’t available in Ads and then describes the workaround. Something like: I reconcile against your CRM monthly, last month Ads claimed 62 and your CRM has 51 with names attached, here’s where I think the gap comes from.
- “What did you switch off last month?” An account that only ever receives additions is not being managed, it’s being decorated. Every genuine month of optimisation contains a subtraction somewhere. Negative keywords added, an audience paused, a creative killed, a placement excluded.
- “If I cut the budget by 30% tomorrow, what goes first and what breaks?” This forces a prioritisation they should already carry in their head. Hesitation here means nobody has ever ranked your channels by what they’re actually worth.
- “How many of the leads we sent got called, and what did your contact in sales say about them?” If your agency has never spoken to whoever answers the phone, they are optimising towards form fills. Form fills are not customers.
Question four is the one that ends engagements. I’ve watched it land twice this year, and both times the silence told everyone in the room what the previous eleven reports hadn’t.
About that Rs 25,000 retainer
Let’s do the arithmetic that nobody does out loud.
Rs 25,000 a month, after the agency’s own salaries, tools, office in Koramangala or Andheri and margin, buys you somewhere around three to five hours of somebody’s attention. Not senior attention. At that price the person on your account is junior, carrying eight other accounts, and working from a template.
A monthly PDF is roughly what those economics support. That part isn’t a scandal, it’s maths.
The rubbish part is the description. Selling three hours a month as “full service digital marketing across SEO, social, paid and content” sets up a relationship that has to be theatre, because no honest version of it fits in the budget. Plenty of the people selling it know this perfectly well.
If Rs 25,000 is genuinely what you have, the better spend is one channel done properly by one person who answers your WhatsApp, or a fixed-scope audit that tells you where the money is leaking before you commit to another year. Both beat a retainer that funds a report.
What I’d actually do this week
Pull your last three monthly reports and put them side by side. Not to check whether the numbers improved. To check whether the same four metrics appear in all three, or whether the metrics quietly change each month depending on which ones happened to look good.
That second pattern is the most reliable signal of an agency managing your perception rather than your account, and it takes about ten minutes to spot.
The agencies still standing in two years will be the ones comfortable saying “we can’t see that, here’s what we do know instead.” Measurement is getting foggier, not clearer, and the ones still selling certainty are selling the one thing the platforms have stopped supplying.
If you want a straight second opinion on whether yours is earning the money, an audit is where I’d start, and the alignment notes cover what to look for. If you’ve already made up your mind about replacing them, I’ve written separately about choosing the next one without getting sold to. Either way, ask question four first. I’m not for hire [I’m contracted full time] but if you want to think it through out loud, drop me a line on email, WhatsApp or LinkedIn and we can have a quick chat.