Why Your Agency Can’t Fix Data Loss, and What To Do Instead
About ten years ago I was on a tennis court, trying very hard not to think about work. It had been a difficult day on an analytics project for a Fortune 100 brand. Tennis was my reset. And during a break between sets I did the thing I tell myself never to do: I looked at my phone.
There was an email from the managing director of IT on the client side. The entire message was one character:
?
Below it was a forwarded thread from one of their agencies. A long and detailed list of grievances, the gist of which was that the analytics didn’t look right and that we had apparently caused a traffic collapse on a billion-dollar brand’s website.
I went home earlier than planned.
By late evening, my team and I had the answer. A new set of filters had started excluding bot and spam traffic that had been inflating the numbers for months. The “collapse” was the reporting finally telling the truth. The traffic hadn’t dropped. The fake traffic had.
Here’s what has stayed with me: nobody in that email chain was lying, and nobody was incompetent. The agency saw a number fall off a cliff and did what a good partner does – raised the alarm loudly. They just had no way to tell the difference between a site that was broken and measurement that had been fixed. Both of those look identical without context.
Why Accurate Measurement Can Look Broken
True issues and corrected measurement produce the same cratering graph. Consider how many ways a number can move without a single thing being wrong with the business:
- Bot filtering gets enabled and months of inflated sessions disappear overnight.
- Consent Mode goes live and measured conversions fall while actual conversions don’t move at all.
- Cross-domain tracking is fixed and your self-referrals vanish while “direct” traffic shifts dramatically.
- The attribution model changes from last-click to data-driven and every channel’s credit moves at once.
- Transaction deduplication is corrected and revenue “falls” to the number that was always real.
And then the genuinely alarming version, which looks exactly the same: a redirect gets added during a routine site release that quietly strips campaign parameters, and paid traffic starts landing in the reports as direct.
Two identical line charts, each showing the same sudden drop. One is labeled measurement fixed, the other site broken.
To a marketer or analyst, there’s nothing to differentiate this very real and urgent problem from a routine tracking update. Knowing the true story isn’t something most are equipped with. They have to be able to go beyond reporting, behind the dashboards, and diagnose the root cause.
Agencies aren’t Designed to Address Data Issues
Over the years, versions of this have happened on a regular basis. I’m genuinely in favour of working with partners to get to the bottom of an issue, that’s the job. But these experiences have led me to a conclusion that isn’t about talent: different partners should be trusted with different things.
Agencies are extraordinary at what they’re built to do: generating demand, building brands, driving traffic, creating conversations that didn’t exist before. The question isn’t whether they’re good. It’s whether the structure they operate in can produce reliable measurement. Here’s why it usually can’t.
1. The math of agency staffing
The agency industry is overwhelmingly made up of small firms. Promethean Research’s 2026 Digital Agency Industry Report found that 87% of North American digital agencies employ fewer than 50 people, with the average agency in their sample at 31 employees and $4.43M in revenue. Their analysis of 3,172 positions across 1,228 digital agencies found that production employees make up roughly two-thirds of the typical agency workforce, rising to about three-quarters once project management is included.
- Let’s run that math for a 31-person agency.reative, media, content = ~20 people
- Account management, strategy, operations, new business and finance: ~10 people
A 31-person agency
- ~20Creative, media, content
- ~10Account management, strategy, operations, new business and finance
- ~1Left for measurement
There’s just not much room for dedicated measurement specialist, let alone a team.
But headcount isn’t even the biggest constraint for most agencies. Analytics troubleshooting typically requires an engineering skillset, not just analysis expertise.. The person who can interpret a report is rarely the same person who can debug a tag container, trace a server-side request, or work with a client’s IT organisation to get a release reviewed. The combination of engineering and analytics skillsets is what resolves (and prevents) measurement issues, and it’s one agencies are least likely to carry.
2. Channel reporting and measurement are different jobs
Pulling numbers out of Meta and comparing them to Reddit is a reporting exercise. You take each platform, normalise the columns, and build a view of relative performance. That’s legitimate, valuable work.
It’s a fundamentally different discipline from making sure a click is correctly attributed across domains and devices, that a conversion fires once rather than three times, that consent state travels from the browser to the ad platform, or that your media mix model is being fed something true.
The IAB’s State of Data 2026 survey found that up to 75% of the buy-side say leading advanced measurement approaches underperform on rigour, timeliness, trust and efficiency. A joint CIMM and 4As study published in March 2026, based on a survey of 197 senior marketers, reached a similar conclusion: confidence in media and marketing measurement has not advanced at the same pace as measurement capability itself, with 43% citing cross-platform gaps. And Forrester points at the cause: data issues remain the top marketing measurement challenge, with too many unconnected data sources and inconsistent levels of quality among them holding marketers back.
3. When you have many agencies, who scores the game?
Most large brands don’t have an agency, but rather a roster. The World Federation of Advertisers found that the dominant operating model is “multiple agencies managed individually by marketing,” cited by 90% of respondents, and that in 2022 most brands worked with between 1 and 25 agencies across all marketing disciplines.
And brands know the arrangement isn’t working. WFA’s research found marketers rated their roster setup just 5.7 out of 10 for being fit for purpose, with 60% looking to reduce the number of agencies they work with.
Now imagine a brand agency, a performance agency, and an SEO agency each presenting a different number for the same campaign. Each is using the data closest to them, and each number is defensible from where they sit. Someone has to decide which one is true.
Here is the structural problem: you want your scorekeeper to not be on one of the teams. It isn’t a question of honesty – Nobody should be asked to produce the scorecard that evaluates their own performance. This is especially true when multiple agencies have a seat at the table, it’s bound to create conflicts and confusion.
4. The metrics that matter most live outside marketing systems
Impressions and clicks live in ad platforms. Revenue, margin and customer lifetime value live in your CRM, your order system, your data warehouse. Measuring what actually matters means connecting those worlds, which is integration engineering, not campaign analysis. It involves systems most agencies have no access to and no mandate to touch.
5. Analytics gets treated as a launch task, not a running system
Tracking gets configured once at the start of a campaign and then left alone. It becomes an administrative item on a launch checklist rather than something anyone owns.
The trouble is that your website isn’t static.
- It ships releases
- Vendors push updates
- Consent tools change
- Someone adds a redirect
- A developer renames an event
Every one of those is a chance for measurement to drift and break, but none of them generate an alert or raise an issue.
6. Standards only work if someone verifies them
This is the one that underpins everything else. Adverity’s research with CMOs found that 45% of the marketing data used for business decisions is incomplete, inaccurate or outdated. They also stated that not one CMO surveyed considered their data more than 75% reliable with 43% believing that less than half of their marketing data merits confidence.
Forrester’s 2024 Marketing Survey puts it more bluntly: 64% of B2B marketing leaders don’t trust their company’s marketing measurement for decision-making.
Most would point to this being a tooling issue, but really it’s a governance problem. You should be setting the standards for your agencies, not adapting to whatever each of them happens to do. Verification has to span every platform and every partner, because the whole point is to measure against one definition rather than against their own.
What to do instead
None of this means handling analytics badly is inevitable, or that agencies can’t be part of the answer. Some do this genuinely well. But if you’re a brand of any scale, four things are worth getting right:
- Separate the scorekeeper from the players. Whoever validates measurement shouldn’t be compensated on the performance that measurement reports. This single structural choice resolves most source-of-truth arguments before they start.
- Write your standards down, and own them. Which events must fire, what parameters are required, how consent must behave, what counts as a critical failure. If this document doesn’t exist, every partner is using their own definition, and they’re all technically right.
- Monitor continuously, not periodically. An annual audit tells you the state of things on one day. Your site changes weekly. The gap between those two facts is where data loss lives, and it compounds quietly.
- Build the diagnostic capability. In-house or with a partner whose job it is, but someone needs to be able to tell the difference between a site that broke and a measurement that got fixed, within hours rather than weeks.
The agency that emailed my client a decade ago wasn’t wrong to raise the alarm, and had every right to. They just had no way to answer the question they were raising and neither, at that moment, did anyone else in the thread.
That’s the gap worth closing.
Can You Tell a Fixed Measurement From a Broken Site?
InfoTrust Suite monitors your analytics and tagging continuously, so you hear about a break when it happens, not weeks later.
Explore InfoTrust Suite
