Agencies don’t have the luxury of slow or messy data setups. With dozens or sometimes even hundreds of clients, fragmented platforms, and tight deadlines, every inefficiency costs time. And when reporting is expected in real time, there’s no margin for error.
We asked the team behind some of the most advanced agency data stacks, Adverity’s Professional Services and Solutions Consulting teams, to share what they’ve learned from working with agencies across the globe.
So listen up! These are some of the most common best practices shared by our team, whether they’re helping clean up a bloated stack or setting up scalable infrastructure from scratch.
Here’s what they recommend…
Before you can scale, automate, or analyze, your foundational setup needs to be tight. This means making smart decisions around how you fetch, organize, and maintain your data.
Here are a few tips on how to lay out data pipelines in a way that won’t break as you add more clients and data sources into the mix.
One of the most common mistakes that agencies make is overlapping data fetches, or triggering excessive data fetches. Either case can really slow things down, especially in large accounts hosting multiple complex datasets.
Large historical data fetches, slower APIs, and datastreams hosting complex transformations can cause drag in your data pipelines. Take these into account and avoid scheduling slower fetches at the same time.
Here are a few recommendations for how agencies should schedule fetches:
You don’t need every field mapped. Only map the fields you're actively using in dashboards or reports to minimize the risk of data inflation. In the case of test fields, only map them once you’re ready to begin testing, then be sure to either delete them or fully implement them as soon as testing is complete. This one is important to avoid downstream confusion!
If you can, use an automated feature like automapping to review your data and assign unmapped metrics and dimensions to existing target fields with one click. This will make for a faster set up and will stop duplicates creeping in.
And last but not least - descriptions matter. A short explanation of what a field actually means helps everyone downstream. Especially new joiners or less technical users. I recommend this guide on creating a data dictionary from Esther, our Senior Solutions Consultant, if you want more info on this, but I will include a brief warning from Esther:
“It’s important to avoid getting caught up in semantic debates. Instead, focus on selecting standardized names for fields that will make sense to any team at any stage, from data engineers to marketing and agency teams.”
There’s no universal best practice, but there is a right structure for your agency.
“If your workspace mirrors how your teams work in real life, everything else gets easier - permissions, cloning, monitoring, scaling,” said Ben Pinkus, Professional Services Lead.
The most common setups we see within agencies include:
You can check out a more in-depth guide on choosing your agency’s data pipeline setup here.
A test stream from two years ago might still be dragging down performance. We recommend regular soft audits of your data streams to ensure all your data is recent and relevant.
Even with the best structure, bad data ruins reporting. The strongest agency stacks put quality checks in place early and monitor continuously.
Naming is infrastructure. If your setup spans dozens of clients and campaigns, you can’t rely on people just figuring it out. Taxonomy needs to be standardized, enforced, and legible both for humans and for automation.
A commonly used format is:
Datasource | Client | Report Type | Granularity
Use consistent delimiters. Pipes (|) or underscores (_) are cleanest. Avoid dashes (-) since they’re often used in date formats and can break parsing logic.
For global teams, include coded fields (like UK, BRND, CTV) and maintain a central dictionary to reduce variance. But be strategic. At granular levels like ads or creatives, over-coding slows things down. Use simple, intuitive labels that make dashboards easier to scan and segment.
As our Product Marketing Manager Dayo puts it: “It’s not length that’s the problem, it’s the readable element that counts.”
If you're operating at scale, consider automating your naming convention checks. A smart transformation layer can flag names that don’t match your standard before they reach dashboards or reports. Check out Dayo’s guide on naming conventions for a more in depth explanation of best practices.
Problems can show up in reporting long after they’ve started. The earlier you catch them, the less fire-fighting you need to do.
Ways to stay ahead:
Check out Lily’s guide for more tips on best practice for setting up data notifications.
When you have a large portfolio of clients with different data sources and people working hands on with the data, a missed naming convention can quickly snowball and create chaos. Tools with built in data quality measures like Adverity’s Data Quality Suite are great for keeping users on track by enforcing your data governance and helping you catch issues at the source before they create chaos downstream.
Once your data pipelines are in good shape and your data is reliable, scaling becomes less about effort and more about process.
This is where large agencies get their edge. With the tools like Adverity’s Scalability Suite that can perform bulk actions across multiple data streams you can:
This means you can get your North Star set up much more easily every time without the set up process being so rigid that it breaks. And it connects neatly back to how your workspaces are structured. If those are organized smartly, bulk actions become much easier to execute, and safer too.
Sometimes a client just wants to know “what we spent on Google Ads last quarter” and your dashboard doesn’t quite show it. It might not be on a dashboard. It might require custom filters. Or maybe it needs custom logic that slows everything down.
That’s where AI-driven tools can help. When you can query your data in plain language and get a clean answer back, chart, table, context included. The same goes for data transformations. Instead of writing SQL or chasing down a developer, newer AI assistants can now generate working code based on what you describe in plain English.
This doesn’t remove the need for technical oversight, but it lowers the barrier to getting started and makes iteration much faster. AI won’t fix a broken data stack. But when paired with clean structures and reliable data, it lowers the barrier to insight and drastically shortens the time it takes to get there.
Agencies thrive when their data is trustworthy, flexible, and fast. That’s only possible with a strong operational backbone. These best practices aren’t theoretical. They’re pulled from the real-life setups of high-performing agencies already doing this at scale. Use them to audit your current setup, or as a blueprint to get your next client live faster and smarter.
About Joseph Caston
Joseph Caston is Senior Director of Solutions Consulting at Adverity, with over two decades of experience in ad tech and data-driven marketing. He specializes in bridging complex technical solutions with strategic business outcomes, advising enterprise clients on everything from data governance and GDPR compliance to AI-powered analytics. Prior to Adverity, Joe spent 11 years at Sizmek, where he played a key role in technical consulting, RFP leadership, and cross-functional solution delivery across EMEA.
About Ben Pinkus
Ben Pinkus is Director of Professional Services at Adverity, where he leads a global team helping clients streamline and scale their marketing data operations. With nearly a decade of hands-on experience in analytics roles at leading agencies like Wavemaker and MEC, Ben has deep expertise in building automated reporting frameworks, managing complex data transformations, and delivering insights across enterprise marketing teams.