What Is a Marketing Agency Reporting Stack?

A marketing agency reporting stack is the set of six layers that sit between a marketing platform and the number your client reads in a report. Every agency already runs all six, because a layer is a job that has to get done rather than a product you buy.

That second sentence is worth slowing down on, because it’s the whole idea.

Take one job: getting yesterday’s Google Ads spend out of Google Ads and into a client report. Something has to do that. At a four-client agency it’s a person exporting a spreadsheet on Tuesday morning. At a forty-client agency it’s software running at 6am. Either way the job gets done, so either way the layer exists. The only thing that differs is what performs it.

So the useful question isn’t whether you have a reporting stack. You do, and you’ve had one since your first client. The question is which of your six layers is still a person, and whether that’s still working.

This article names all six layers and explains what each one does in plain language. For each one you’ll get the specific signal that tells you it’s outgrown being done by hand, and the equally specific case for leaving it alone. It also covers the data warehouse question, which is the most expensive decision in the stack and the one most often made for the wrong reason.

Swydo is client reporting software for marketing agencies. We cover four of the six layers. The two we don’t cover are storage, which we don’t offer in any form, and narrative, which should stay a person on your team. Both of those get a straight answer here anyway.

Why You Already Have All Six Layers

Every agency runs all six layers because each layer is a job that can’t be skipped. Data has to come out of the platforms. Somebody has to decide what “conversion” means. Somebody has to tell the client what happened. Skip any of those and there’s no report.

What separates a four-client shop from a forty-client one isn’t the number of layers. It’s how many of them are automated and how many are still a person with a browser tab open.

That distinction changes the question you’re asking. “Do I need a reporting stack?” has no useful answer, because you already have one. “Which of my six layers is still a person, and is that still fine?” has a very specific answer, and it’s usually different for each layer.

Some layers should stay a person forever. Telling the client what the numbers mean is the obvious one, and it’s the thing they’re actually paying for. Some layers should never have been a person to begin with. And some are fine as a person right up until the week you’re short-staffed, which tends to be the same week you changed something that broke.

Here’s a question worth answering before you read further. If you had to name who or what performs each of these six jobs at your agency, could you? Most people can name two.

The Six Layers of a Marketing Agency Reporting Stack

The six layers are connection, transport, storage, modeling, presentation, and narrative. Each one has a job, something that currently performs that job at your agency, and a specific condition that means it’s time to stop doing it by hand.

Read them in order. The order matters because a problem in an early layer shows up as a symptom in a later one. A dead connection looks like a bad month in a chart, so agencies routinely go and fix the chart.

1. Connection

The connection layer is the permission that lets your reporting see your client’s data. When you log into a client’s Google Ads account and click “allow” so a tool can read it, the platform hands back a digital key called a token. That token is the connection layer.

At most agencies, the connection layer is whoever clicked “allow” first, usually while logged into their own personal account.

This is the one layer where there’s no real argument for both sides, and not because you need to buy anything. The fix is free. Connect through a shared agency login or a role-based account instead of an individual’s personal one, and have the client own the underlying ad and analytics accounts rather than you.

Do that and you remove a whole category of failure. Tokens die for reasons that have nothing to do with your work: a client changes a password, someone switches on two-factor authentication, a permission gets revoked during a staff change on their side. When a token dies, data stops arriving, and nothing announces it.

The part worth paying for is detection rather than the connection itself. Swydo’s Data Health Check Alerts watch every integration for broken connections and expired tokens, flag the affected source under Settings, and email the person who owns that connection with a link to reconnect it. Whatever tool you use, find out how it notices a dead connection, because the alternative is a client noticing first.

The moment to fix this layer is the first time somebody leaves your agency and takes a dozen client connections with them. That’s a bad week, and it’s entirely preventable.

2. Transport

The transport layer moves data out of each platform on a schedule and into wherever you build reports. Platforms offer something called an API, which is a doorway that lets other software request data automatically instead of a human clicking export. Transport is whatever walks through that doorway on your behalf.

Four things can perform this job: a person, a script somebody on your team wrote, a connector tool, or a full reporting platform. All four are legitimate.

Leave it manual when you’re small. Under roughly eight clients, running two or three platforms each, with one person owning reports and enough breathing room to check every one before it goes out, manual exports work fine. The time it takes to set up and learn a platform can genuinely exceed the time it saves at that size. Anyone telling a six-client agency it needs automated pipelines is selling something.

Build it yourself when you have the skills on payroll. You get data shaped exactly how you want it, you pay in hours rather than subscription fees, and if there’s already a developer at your agency for another reason, the maths changes. Agencies whose actual product is measurement rather than media should own this layer. The trap is that an AI assistant will get you a working prototype in an afternoon, which makes the decision look settled before the maintenance question is asked. Our build vs buy breakdown for client reporting runs the numbers on the part that doesn’t show up in a demo.

Rent it when the upkeep starts landing on you. Platforms change how their doorways work, and each one runs its own schedule. Google Ads, Meta, and LinkedIn all retire old versions on published deadlines, and when a version retires, anything built against it stops working that day. A client running five channels means five of those schedules, forever. Rent transport and someone else tracks them.

The signal that transport has outgrown being a person is quieter than you’d expect. It’s the month you skipped the manual check because you were slammed, and nothing bad happened. That’s not reliability. That’s luck, and luck doesn’t survive forty reports a month.

3. Storage

The storage layer keeps a copy of your clients’ data after the platform stops keeping it, and holds it in a shape the platforms can’t produce on their own.

Most agencies have no storage layer at all, and most agencies are right not to. Reporting tools keep a working copy of recent data, which covers ordinary client work without you thinking about it.

Storage becomes its own layer when you need a database you control. That’s the data warehouse question, and because it’s the most expensive decision in the stack, it gets its own section further down.

4. Modeling

The modeling layer decides what a metric actually means. Which conversion action counts as a conversion. How many days after a click a sale still gets credited. Which currency the numbers are in. What goes into “blended ROAS,” which is a single return figure combining several ad platforms, and what quietly gets left out of it.

Every agency has this layer, because every agency has made these decisions. Almost none has written them down.

It works as a person when there’s only one person. One strategist builds every report, has done for three years, and the definitions stay consistent because there’s only one set of them. That’s a real situation and it functions well.

It stops working when a second person joins. Two people define the same metric differently, or you start putting a single number in front of a client that combines platforms. Either one means the definition needs to live somewhere other than a memory.

The honest advice here isn’t a tool. You make this layer explicit by writing your definitions down, and a shared document costs nothing. What software buys you is enforcement. A blended number calculated once in one place stays consistent, while the same calculation copied into three spreadsheets drifts apart within a quarter, and nobody notices until a client asks why two reports disagree.

In Swydo that enforcement is a Combined Data Sources widget, which calculates figures like total ROAS across Google, Meta, and TikTok in one place, and works inside monitoring and alerts rather than only inside reports. If you currently do this with spreadsheet formulas, data blending explains the mechanics before you commit to an approach.

One limit worth knowing before you build around it. Custom metrics can’t be used inside Combined Data Sources widgets, so a calculation you’ve defined for a single platform won’t carry through into the combined view. You’d need a manually entered figure or a separate widget for it.

5. Presentation

The presentation layer turns modeled data into something a client actually reads. A PDF, a live dashboard, a slide, an email.

Free tools are genuinely enough sometimes. A mostly-Google client roster, a manageable number of accounts, and no requirement to put your own branding on things. Paying for presentation you don’t need is one of the more common ways agencies waste money. The Looker Studio limitations that cause real trouble are rarely about what the charts can do.

Free stops being enough when rebuilding eats your time. The tipping point arrives when you spend longer assembling reports than thinking about what’s in them, or when you need branding, client-facing portals, and the ability to push a change across forty reports at once. That last one is usually the forcing issue, and scaling Looker Studio covers where the ceiling sits.

Most agencies change this layer first, because it’s the one they look at every day. It’s rarely the one costing them anything.

6. Narrative

The narrative layer is somebody saying what the numbers mean and what happens next. It’s a person, and it should stay one.

AI can write the first draft. What AI can’t do is know what matters to this particular client this particular quarter, and a summary written on top of data nobody checked will explain a downturn that never happened, in fluent and confident prose. If a connection died and spend shows as zero, an AI summary will describe a strategic pullback from paid social. Your account manager pastes it into an email, and now a broken token has become a story somebody has to walk back.

Automating the five layers above this one is what protects the time this layer needs. That’s the actual argument for a reporting stack, and it’s the same conclusion <a href=”https://www.swydo.com/blog/client-reporting-best-practices/”>client reporting best practices</a> arrives at from a different direction.

Do You Need a Data Warehouse?

A data warehouse is a database you own that stores copies of your clients’ marketing data, so that you keep that data even after the platforms delete it. Google’s version is called BigQuery. Most agencies don’t need one, and the reason has nothing to do with cost.

A warehouse doesn’t replace a layer in your stack. It adds three.

Something has to fill it, which is a pipeline you now maintain. Something has to organize the raw tables into usable metrics, which is work somebody has to do and keep doing. And you still need a presentation layer on top, because a warehouse produces database tables, not client reports. The agency that adds a warehouse hoping to escape connector upkeep has swapped one maintenance job for three.

Storage itself is cheap, and that’s exactly what makes this look like an easy decision. The real cost is the person who keeps the pipeline running, and that person is now permanent.

So the question isn’t whether warehouses are good. It’s whether the problem you have is actually a storage problem. Four times, it genuinely is.

Platform data you sell against is expiring. This is the strongest reason and the most common one. Google Search Console keeps 16 months of performance data and no more, regardless of which reporting tool sits on top of it. Older data is deleted permanently and nobody can get it back. If you pitch and defend multi-year organic growth, something has to be storing that data now. One catch worth knowing: a warehouse copy captures nothing from before the day you switch it on, so this decision is worth making early or not at all.

You need to combine marketing data with business data. Profit margin per product. Which customers stayed for two years. Sales that closed over the phone. Revenue figures living in a client’s CRM. No reporting connector reaches those, because they sit in systems that aren’t marketing platforms. If your reporting has to answer “which channel brought in the most profitable customers,” you need somewhere to put both sets of numbers side by side.

A client contractually owns the raw data. Larger clients sometimes write raw data handover into the contract. That’s a storage requirement, and no amount of dashboard quality satisfies it.

Measurement is the thing you sell. Some agencies sell analysis rather than media buying, using methods that model the effect of spend across channels or test what would have happened without a campaign. If that’s your product, you’re running an analytics business, and you should own storage and modeling deeply.

Three times the answer is no, even when it feels like yes.

Your reports keep breaking. A warehouse fixes none of the things that break reports. Dead tokens, retired platform versions, and daily request limits all happen before data ever reaches storage, and they’ll happen identically with a warehouse in place. This is the most common reason agencies reach for one, and it’s a reliability problem in a storage costume.

You want one source of truth. You can have that without owning any storage. Agree on what each metric means, calculate combined figures in one place instead of several, and write the definitions down. That’s the actual fix, and it costs nothing.

Somebody told you storage is cheap. It is. The pipeline, the organizing work, and the person maintaining both are not, and none of those show up on the first invoice.

Our own position, stated plainly: Swydo doesn’t offer a warehouse and doesn’t connect to one directly. If you conclude you need storage, we’re not that layer. The way data from a warehouse reaches Swydo today is through a Google Sheets data source, which works but isn’t graceful. Plenty of agencies run a warehouse for the two or three clients that need one and a reporting platform for the other forty, and that’s a sensible arrangement rather than a compromise.

How the Stack Changes as You Add Clients

Client count is the shorthand agencies use for how built-out their stack should be, and it’s a rough guide at best. The number that actually predicts when things break is how many broken connections you’d hear about within a day.

Work it out. Multiply your clients by the number of platforms each one runs. Twenty-five clients on four platforms each is a hundred connections. Now count how many of those you’d notice going quiet this week. The gap between those two numbers is the real measure, and it’s why two agencies with identical client counts can be in completely different shape.

Patterns still cluster by size, though.

Around five clients. Every layer is a person, and that’s correct. Fix the connection layer since it’s free, and leave everything else alone. Anything you buy at this stage is capacity you have no use for yet.

Around twenty-five clients. Transport breaks first, and it breaks as a noticing problem rather than a volume problem. You can still build every report. You can’t verify all of them in your busiest week. Modeling starts drifting at the same time, because more than one person now builds reports and nobody wrote the definitions down. That drift is a standardization problem before it’s a tooling one, and standardizing and scaling client reporting covers the template and shared-KPI work that stops two people defining the same metric two ways.

Around fifty clients. Watching for problems becomes a job of its own whether you plan for it or not. Opening dashboards one at a time stops being realistic somewhere around client fifteen, and “I’ll check when I have a minute” reliably becomes “I didn’t.” Swydo’s Metrics Overview exists for this, putting key metrics for every client on one screen so the morning check stays a two-minute glance at forty clients. The principle holds whatever tool you use: past a certain roster size you need a view that doesn’t grow when the roster does.

The sequence is more useful than the numbers. Connection, then transport, then modeling, then monitoring. Presentation slots in whenever it’s genuinely the bottleneck, which is later than most agencies assume.

Who Should Own Each Layer

Ownership of a reporting stack splits three ways between your agency, your client, and a vendor. Getting it wrong creates problems that stay invisible until a staff change or a lost account, at which point they all arrive at once.

One pattern underneath the table is worth saying out loud, because it’s the reverse of how most agencies spend their time. The two layers you should never hand to anyone else are modeling and narrative, since those are your judgment and they’re what the client is buying. The two layers most agencies pour the most hours into are transport and presentation, which are the two most worth renting.

Two specifics catch people out repeatedly.

Client-owned accounts with your agency granted access separately survive the end of the relationship in both directions. You keep working when their marketing manager quits, and they keep their history when they leave you. Agencies that hold client ad accounts in their own name think they’ve built in leverage. What they’ve built is a messy offboarding and a reason for the client to resent them on the way out.

And whoever owns storage owns the obligation that comes with it. Holding three years of a client’s data means being responsible for holding it properly, deleting it when asked, and being able to say where it lives. For most agencies that’s a good reason not to own it.

Where the Cost of a Reporting Stack Actually Sits

Reporting stacks get paid for in one of two currencies, subscription or salary, and agencies underestimate the second because it arrives without warning. A forced platform migration doesn’t send an invoice. It takes a Thursday.

Mapping cost to layer shows why the money conversation usually happens in the wrong place.

Presentation gets the most scrutiny and matters the least. Comparing tools on their charting features is a comfortable way to avoid the harder question about transport. If you’re at the stage of comparing options, selecting a marketing report automation tool is worth running against how a tool behaves when something breaks rather than against a feature checklist.

On pricing models, since they vary more than the sticker price suggests: Swydo charges by data source rather than by client, with the first 10 sources included and no limit on clients, users, or reports. A data source is one connected account, so a client running Google Ads, Meta Ads, and a Facebook Page counts as three. That model suits agencies with many light clients and suits agencies with a handful of heavy ones less well. Count your sources rather than your clients before comparing any tool’s price, because the two numbers diverge fast.

What Happens When a Layer Fails

Each layer fails in its own way, and the dangerous version is never the one where a dashboard goes blank. It’s the one where the report builds normally, every chart fills in, and a number is quietly wrong.

A dead connection usually shows up as a zero rather than an error, and a zero reads as a bad month rather than a broken pipeline. That’s why failures in the connection and transport layers get diagnosed as presentation problems, and why agencies swap reporting tools to fix something the new tool will do exactly the same way. If your reports have been going wrong in ways you can’t explain, client reporting accuracy covers the five signals that catch these failures before a client does.

Where to Start

Name who or what performs each of the six layers at your agency right now. The layers you can’t name are running on assumption, and assumption works fine until it doesn’t.

Then pick the single layer where a person is doing something a system should do, and fix only that one. For most agencies past twenty clients it’s transport, and the test is whether every report went out last month with somebody actually looking at it first.

Leave the warehouse question alone unless one of the four reasons above describes you. It’s an expensive way to solve a problem you probably don’t have.