Choosing a Marketing Analytics Stack for Unit Economics
Half of marketing leaders can't defend how they measure ROI. That's not me being cynical, that's the number from a 2025 Haus survey that made me put my coffee down. Fifty percent. And these are the people who walk into the QBR and get asked, straight-faced, "so what did the $2M do?"
If you're shopping for a marketing analytics stack to fix that, here's the short answer before the long one: don't evaluate tools on their feature lists. Evaluate them on whether they can answer four unit-economics questions cleanly, in numbers a CFO won't pick apart. Everything else is decoration.
Those four questions are cohort revenue, blended CAC, payback period, and channel ROAS. A stack that answers all four is worth buying. A stack that answers three and hand-waves the fourth is the reason you'll be back here in eighteen months.
The four questions your stack has to answer
I've built this rubric the hard way, mostly by buying tools that couldn't do half of it. So let me hand you the checklist instead of the scar tissue.
| Question | What it actually needs | The failure mode |
|---|---|---|
| Cohort revenue | Revenue attributed to the acquisition month, tracked forward for 12-24 months | Tools that only show revenue by the month it landed, not by when the customer signed up |
| Blended CAC | Total sales + marketing spend / new customers, and a per-channel cut | Platform dashboards that report their own channel's CAC and ignore the rest |
| Payback period | Months to recover CAC from gross-margin-adjusted revenue | "Revenue payback" that quietly forgets margin exists |
| Channel ROAS | Return per channel that reconciles to one revenue number | Three tools reporting three different conversion totals |
Notice what's not on that list: attribution model wars, pretty funnels, AI copilots. Those are nice. But if your tool can't produce a cohort revenue curve, none of the nice stuff matters, because you can't compute lifetime value, and without lifetime value your LTV:CAC ratio is a guess with a decimal point.
Here's the thing about that ratio. The median B2B SaaS company runs about 3.2:1, according to Optifai's benchmark across 939 companies surveyed through early 2026. The 3:1 "rule" everyone quotes is real enough. But you can only get there if your stack tracks revenue by cohort. Miss that, and you'll either overspend chasing a ratio you can't see or starve a channel that was actually paying back fine.
Why cohort revenue is the load-bearing wall
Most tools show you revenue the way an accountant does: what came in this month. Useful for the P&L, useless for growth. What you need is revenue lined up by when the customer was acquired, so you can watch a January cohort mature across the year and compare it to the March cohort you paid more to get.
If a tool can't pivot revenue by acquisition cohort natively, you'll end up exporting to a spreadsheet every month. I did that for two years at a company I won't name. It worked until it didn't, which was the day someone found a formula error in row 4,000 and we had to re-defend a quarter of channel decisions. Don't be me.
The category map (and where the tools actually live)
There isn't one product that does all of this, which is the annoying part. The market is split into categories that each own a slice, and building a stack means picking one or two and gluing them together. Here's the honest map.
| Category | What it's for | Example tools | The trade-off |
|---|---|---|---|
| BI / warehouse tools | Any metric you can write SQL for, including cohort revenue and blended CAC | Looker, Metabase, Omni, dbt + your warehouse | Total flexibility, but you build every metric yourself and need someone who writes SQL |
| Product analytics | Events, funnels, retention, cohorts, user journeys | Amplitude, Mixpanel, PostHog, Kixo | Fast to answer behavioral questions; revenue and spend usually have to be piped in |
| Mobile measurement (MMP) | Install attribution, channel ROAS on mobile, deep links | AppsFlyer, Adjust, Branch | Best-in-class for paid mobile; not where you compute company-wide payback |
| Marketing attribution / MMM | Multi-touch or modeled channel contribution | Rockerbox, Northbeam, Haus | Answers "which channel," fuzzy on "what's the customer worth" |
A few notes so this table doesn't mislead you.
Product analytics tools have crept toward revenue over the last couple of years. Amplitude and Mixpanel can both hold revenue events; PostHog leans warehouse-friendly and open-source; Kixo takes the chat-first angle, where you ask for a cohort or a funnel in plain language and it generates the chart with a visible reasoning trail, which is genuinely faster for the "just show me the March cohort" moment but still needs your spend data piped in from somewhere to close the CAC loop. None of them are a substitute for a warehouse if you want blended, cross-channel truth. They're where the behavioral half of unit economics gets easy.
MMPs are a trap if you mistake them for a full stack. They're superb at what they do, which is mobile install attribution and channel ROAS. They are not where you compute company-wide payback, because they only see the channels they measure. I've watched a team declare a channel dead based on their MMP, then discover the warehouse told a completely different story once organic and web were in the same table.
The buy-versus-build fork
Every stack decision eventually hits the same fork: warehouse-native, where you own the data model and assemble tools on top of it, or turnkey, where you buy a bundle that's fast to stand up but harder to bend.
The market's been drifting one way. Per eMarketer's read of CDP Institute data, warehouse-native CDP vendors grew headcount 7.8% in the back half of 2025, roughly six times the industry average, and cloud data warehouse share of the stack climbed from 20.9% to 23.9% while packaged CDP share slid. Snowflake, BigQuery, and Databricks are quietly becoming the center of gravity.
That drift makes sense for unit economics specifically, because unit economics is fundamentally a joining problem. You're joining spend from ad platforms to signups from your app to revenue from your billing system. A warehouse is the only place all three sit in the same schema. Bundled tools tend to be excellent at one input and reliant on connectors for the rest.
But warehouse-native isn't free, and anyone who tells you otherwise is selling a warehouse. You need a data engineer, or a very brave analyst, plus dbt models plus a BI layer plus the patience to maintain all of it. Turnkey buys you a working dashboard next week at the cost of doing the math the vendor's way. My rule: if you have someone who can own the warehouse, go warehouse-native, because unit economics will keep changing and you'll want to change with it. If you don't, buy the bundle and revisit in a year. There's a fuller decision framework in the warehouse-native versus bundled breakdown over at Best Analytics Tools if you want the 2x2.
Napkin math: what payback actually costs to get wrong
Let me show you why the payback question is worth this much fuss. Round numbers, on purpose.
Say you spend $100,000 a month on a channel and it brings in 200 customers. That's a $500 CAC. Each customer pays you $50 a month, so revenue payback looks like 10 months. Fine, ship it.
Now put margin back in. Your gross margin is 70%, so each customer really contributes $35 a month, not $50. Payback is 14.3 months, not 10. Still okay, but you just moved 40% closer to the bottom quartile. First Page Sage's 2025 benchmarks put the top quartile at six months or less, the median at 16, and the bottom quartile at 24-plus. A tool that reports revenue payback instead of margin-adjusted payback just told you 10 when the truth was 14.3, and at scale that four-month gap is the difference between a channel you fund and one you cut.
Multiply the mistake. If you're running that channel at $100k/month and you misread payback for two quarters before catching it, that's $600k deployed against a number that was off by 40%. This is the single most common thing I see broken in a stack, and it's usually because someone bought a tool that optimizes for "look how much revenue we drove" rather than "here's what you actually keep."
So when you demo a tool, ask them to show you payback with gross margin applied. If the answer is "you'd export that to a spreadsheet," you've learned something.
The evaluation checklist you can hand to a vendor
Print this. Walk it into the demo. Watch which questions make the sales engineer reach for the "roadmap" slide.
- Cohort revenue, native. Can it pivot revenue by acquisition month without an export? Ask to see a 12-month cohort curve on the spot.
- Spend ingestion. Does it pull spend from every paid channel you run, or just the ones it was built for? Blended CAC dies here.
- Margin-aware payback. Can it apply gross margin to the payback calc? See the napkin math above for why this is non-negotiable.
- One revenue number. When two reports disagree on conversions, which one is right, and can everyone reconcile to it? Multiple "sources of truth" is how the QBR becomes a debate.
- Attribution honesty. Does the tool tell you its coverage limits, or pretend it sees everything? Even good multi-touch tops out at 30-60% coverage in 2026 thanks to iOS and cookie changes. A vendor claiming full-funnel certainty is either lying or hasn't read the room.
- Export and ownership. Can you get your raw data out? If the answer is friction, your unit-economics model is a hostage.
One more thing on sequencing, because people ask. Don't buy all four category tools at once. Start with the thing that answers cohort revenue and blended CAC, which is almost always the warehouse or a product analytics tool with revenue events. Get one honest number for LTV:CAC and payback before you spend a dollar on an MMP or an attribution vendor. I've seen teams buy the shiny multi-touch tool first and then discover they had no reliable revenue number to attribute to. That's building the roof before the walls.
That fifth point deserves a second. The attribution wars are mostly noise now. You will never get to 100% deterministic attribution again, and chasing it is a good way to burn a quarter. What you can do is pick one revenue number, model channel contribution against it with honest error bars, and move on. The teams I've seen defend their budgets successfully aren't the ones with the fanciest attribution. They're the ones whose numbers reconcile.
What this stack is not for, and who shouldn't buy the fancy version
If you're a five-person team spending $10k a month, do not build a warehouse-native stack. You'll spend more on the plumbing than the plumbing saves you. A spreadsheet plus one product analytics tool with revenue events will carry you further than you'd think, probably to $2-3M ARR. I've watched a seed-stage team hire a data engineer to build a modern stack they used twice a quarter. That's a hire you make when the spreadsheet breaks, not before.
And if your business is genuinely simple, one product, one channel, monthly billing, you may not need any of the four questions answered separately. Your blended CAC is your channel CAC. Buy the cheap thing.
The full rubric earns its keep when you've got multiple channels, a subscription or repeat-purchase model, and someone senior asking where the money went. That's when the difference between a stack that answers all four questions and one that answers three stops being academic and starts being your budget.
The measurement gap here is real and widening, by the way. That State of Brand finding that 56% of B2B marketers can't attribute ROI to content, with 44% unable to tie performance to business goals at all, isn't a tooling problem you can buy your way out of overnight. But a stack built around these four questions is the closest thing to an antidote I've found. If you want the spreadsheet side of it laid out, we've got a unit-economics dashboard template that pairs with any of these tools.
A few questions I get asked in demos
Do I need an MMP if I have product analytics? If you run paid mobile at any real spend, yes. Product analytics won't reconstruct install attribution the way an MMP does. If you're web-only, skip it and save the money.
Can one tool replace the whole stack? Not honestly, not yet. The warehouse plus a BI layer comes closest to answering all four questions in one place, but you'll still bolt on product analytics for behavior and an MMP for mobile. Anyone selling you a single-pane-of-glass that does everything is describing a roadmap.
How do I know if my current stack is failing? Simple test. Ask three people for last quarter's blended CAC. If you get three numbers, your stack isn't a stack, it's a pile of tools. Fix the "one revenue number" problem first, everything else after.