How to Track Marketing Spend Across Channels (Without a Warehouse)

Ask a marketing leader what they spent by channel last month and you’ll usually get one of two answers: a spreadsheet someone is quietly afraid of, or a roadmap slide for a data warehouse project that’s three months out. We help clients answer this question every month, and the honest finding is that both defaults are wrong. The right method is smaller than the warehouse and more disciplined than the spreadsheet — and it runs on CSV exports, a ledger, and a couple of rules about identity.

The trap: fragile spreadsheet or three-month warehouse project

The spreadsheet fails slowly. It starts as a clean tab with a row per campaign, and then entropy does its work: someone pastes April’s re-exported Meta numbers over the originals, nobody reconciles why the quarter total moved, the Apollo subscription that renews on a personal card never makes it in at all. Six months later the spreadsheet still produces a number, but nobody in the room would defend it. The numbers aren’t wrong because anyone was careless — they’re wrong because the spreadsheet has no rules about what a row is, so every update is a small act of silent history rewriting.

The warehouse fails expensively. Fivetran plus a warehouse plus a BI tool is a legitimate stack — for a different problem. If the question is “what did we spend by channel last month,” standing up connectors, modeling layers, and dashboards is three months of work to replace a fifteen-minute monthly export. And here’s the part the warehouse pitch leaves out: the long tail of marketing spend — Apollo, Clay, the conference sponsorship, the agency retainer, the niche newsletter placement — has no connector. There is no Fivetran source for “the booth we paid for in March.” So even after the project ships, the warehouse holds the ad platforms and someone still maintains a side spreadsheet for everything else. You’ve built the expensive system and kept the fragile one.

The insight: three kinds of spend, three different methods

The reason both defaults fail is that they treat marketing spend as one kind of data. It’s three:

1. Ad platform spend (Google, Meta, LinkedIn). High-volume, campaign-level, changes daily — and every platform exports a clean campaign-level CSV in about two minutes. You don’t need an API integration to get this data. The export is the integration. It’s the same numbers the API would give you, on a cadence that matches how often you actually need them.
2. Recurring tool and data vendors (Apollo, Clay, ZoomInfo, your email platform, the agency retainer). These don’t need exports at all — they need a ledger: vendor, monthly amount, channel, start date, end date if known. This is fifteen minutes of configuration and then it’s permanently solved. The ledger generates a row per vendor per month until you tell it to stop.
3. One-offs (events, sponsorships, one-time placements). Dated entries with an invoice reference. There will be a handful per quarter; the only failure mode is forgetting them, which the coverage check below catches.

Most spend-tracking pain comes from forcing one of these shapes through a method built for another — hand-typing ad spend like it’s a ledger entry, or trying to “integrate” a conference invoice.

The method, step by step

1. Define the channel taxonomy first

Before importing a single row, write down the explicit list of channels: for most B2B teams that’s something like paid-search, paid-social, events, outbound-data, and content. The exact list matters less than two properties: it’s written down, and every dollar maps to exactly one channel. If two people can reasonably file the same campaign under different channels, your rollup is an opinion poll. This taxonomy is also what makes CAC by channel computable later — spend by channel is the numerator, and it has to mean the same thing every month.

2. Run the monthly export ritual

In the first few days of each month, download the standard campaign report CSV from each ad platform — same report, same settings, same date grain, every month. Consistency is the whole trick: a report whose columns shift month to month is a new data source every month. Put the files somewhere versioned and dated. The ritual takes ten minutes for three platforms.

3. Normalize to canonical columns

Every export gets reshaped to the same five columns: source, campaign, period, amount, currency. Google calls it “Cost,” Meta calls it “Amount spent,” LinkedIn calls it “Total spent” — after normalization you don’t care. Five columns is enough to answer the monthly question; resist the urge to carry forty columns of platform metrics into your spend table. Those belong in the platforms.

4. Map campaigns to channels with explicit rules

Mapping should be rules, not judgment calls: a source-level default (everything from Google Ads is paid-search unless overridden) plus campaign-name overrides (campaigns matching *-retargeting-* go to paid-social, the brand-video campaign goes to content). This is where campaign naming discipline pays off directly — if your names follow a convention, the rules are short and stable. If they don’t, start with UTM and naming hygiene and the mapping problem mostly disappears.

The critical rule: unmapped spend goes into a visible unmapped bucket — never silently guessed into a channel. A rollup with an unmapped line item is honest and self-correcting; someone sees it and writes the missing rule. A rollup where ambiguous spend was quietly filed somewhere plausible is wrong in a way nobody can detect. This is the same principle that makes CRM cleanup stick: explicit rules, visible exceptions, no silent fixes.

5. Expand the ledger

Each month, the ledger emits its rows: Apollo, $X, outbound-data. Clay, $Y, outbound-data. Agency retainer, $Z, its channel. Add any one-offs that landed this month with their invoice reference. This is the step that makes your spend picture complete rather than just paid-media-complete — and it’s the step no connector will ever do for you.

6. Review the rollup with two trust checks

Before anyone reports the numbers, run two checks:

• Coverage: did every expected source report this month? If LinkedIn is missing, that’s almost never zero spend — it’s a forgotten export. A coverage check turns “we forgot a platform” from a quarter-end surprise into a two-minute fix.
• Restatements: does any re-exported period differ from what you previously recorded? Platforms restate past periods routinely — late-closing conversion windows, invalid-traffic clawbacks. When it happens, record the before and after explicitly. Don’t paste over.

Why row identity matters

This is the subtle piece, and it’s the difference between a spend table people trust and a spreadsheet people argue with.

Give every row an identity of (source, campaign, period) — not the amount, and not “whatever row it landed on.” Now re-importing a corrected export has a well-defined meaning: rows whose identity already exists get their amount updated, with a visible trail showing the old and new values; genuinely new rows get inserted. Without stable identity, a re-import either duplicates everything (so totals inflate) or someone deletes-and-replaces (so history silently rewrites). With it, “Meta restated April down $1,840” is a recorded fact you can point to when the quarter total shifts — instead of an argument about who touched the spreadsheet.

This one design decision is most of what people think they need a warehouse for.

When a warehouse is the right answer

To be honest about the boundary: the exports-and-ledger method answers “what did we spend by channel, monthly,” and that’s the question most teams actually have. A warehouse earns its keep when you need daily-grain analysis, want to blend spend with product or revenue data, are aggregating more than eight to ten sources, and — the load-bearing condition — have a data team that owns the pipelines. If all four are true, build it. Even then, keep the ledger running alongside it, because Fivetran will never have an Apollo connector worth using, and the long tail doesn’t stop existing just because the paid channels got pipelines.

What this looks like in practice

This method — canonical CSV ingestion, a recurring ledger, explicit channel mapping with an unmapped bucket, and restatement detection on stable row identity — works in any tool that respects the rules: explicit taxonomy, same exports every month, visible unmapped spend, and rows whose identity survives a correction. A spreadsheet does it; a small script does it better. And if you want the CRM side of the picture held to the same standard, our open-source toolkit audits source completeness and pipeline hygiene against HubSpot and Salesforce with evidence behind every finding.