Incrementality Testing for B2B SaaS Without an MMM Budget

Incrementality testing measures how many demo requests, trials or pipeline dollars your marketing actually caused, by comparing people or regions that saw your ads with a matched group that didn't. A B2B SaaS team doesn't need a marketing mix model to do it. It needs a CRM outcome it trusts, a holdout it can defend and a few months of clean weekly data.

Most growth teams at Series A to C companies already have attribution, usually a GA4 property plus whatever each ad platform reports (my definitive guide to Google Analytics 4 covers that layer). Attribution tells you which touchpoints showed up before a conversion. It can't tell you which conversions would have happened anyway, and that's the number a CFO asks about when you request more budget. This guide covers that decision: when incrementality testing is worth doing at B2B volume, what it costs, what breaks and how to check that a result is real.

Key takeaways

• Google Ads Conversion Lift needs at least 1,000 observed conversions and a $5,000 minimum study budget, and LinkedIn Conversion Lift Testing needs $80,000 in cumulative budget over 30 to 90 days.
• Most Series A to C SaaS teams miss at least one of those floors on demo or SQL volume, so the practical method is a geo holdout analyzed with free, open-source tools such as Meta's GeoLift or Google's Meridian GeoX.
• Measure lift on a CRM outcome (SQLs, opportunities or pipeline) by region, not on the platform's own conversion count.
• The software is free. The real costs are held-out pipeline, analyst time and clean weekly data by region.
• A geo test that wasn't power-checked before launch often ends inconclusive, so run the power calculation on your own history first.

What is incrementality testing?

Incrementality testing is a controlled experiment that measures the outcomes your marketing caused by comparing a treated group with a holdout group that didn't get the marketing. The gap between the two is the lift. The conversions that would have come in anyway, from brand search, referrals or sales outreach, cancel out because both groups have them.

The holdout can be people or places. Platform lift studies randomize users into exposed and control groups. Geo experiments assign whole regions, such as states or metro areas, to treatment or control. In both cases the control group is your counterfactual: an estimate of what would have happened with no ads.

That's different from an A/B test of two ads. An A/B test tells you which creative wins. An incrementality test tells you whether the channel earns its budget at all.

Randomized experiments are the reference point for this work. Northwestern's Brett R. Gordon, with Robert Moakler and Florian Zettelmeyer, put the trade-off plainly in an April 2026 NBER working paper: "Randomized controlled trials (RCTs) provide the most credible estimates of advertising incrementality but are difficult to scale." For a SaaS team, difficult to scale mostly means volume, which is where the platform floors come in.

Why can't attribution answer the budget question?

Attribution can't answer the budget question because it splits credit among touchpoints that happened, and it has no view of the conversions that would have happened without the ads. Last-click, data-driven and multi-touch attribution all share that limit. They're correlational by design.

In B2B SaaS the gap is usually widest on brand search and retargeting, where the ad sits in front of someone who was already coming. Your attribution report shows a cheap cost per demo. A holdout might show most of those demos arrive anyway. Signal loss makes it worse, because tags that miss conversions undercount some channels and flatter others (I covered the tracking side in is server-side tracking worth it).

The industry has mostly stopped pretending one tool covers it. Kantar ran a 2025 study with Meta based on 1,935 interviews with measurement professionals at companies spending more than $1 million a year on digital marketing. It found advertisers use nearly four measurement solutions on average. Only 3% use MMM on its own, 51% combine it with attribution and experiments, and 55% often get conflicting results from different tools.

Google's own framing, in a November 11, 2025 announcement, splits the jobs cleanly. MMMs give the holistic view, attribution "maps the customer journey," and "Incrementality offers the precise, causal data for specific campaigns and channels." I'd treat that as the division of labor. Keep attribution for daily optimization, and use incrementality tests to check whether attribution's numbers hold up.

Can a B2B SaaS team qualify for a platform lift study?

Some can, but most Series A to C teams miss at least one floor, and it's usually the conversion count rather than the budget. Google, LinkedIn and Meta all run lift studies that randomize users for you, and each sets eligibility rules before it will run one.

Platform	Study	Published floor	What usually blocks B2B SaaS
Google Ads	Conversion Lift	At least 1,000 observed conversions, a $5,000 minimum study budget and one compatible conversion action	Demo and SQL volume, plus account access
LinkedIn Ads	Conversion Lift Testing	$80,000 cumulative budget, 30 to 90 days, Insight Tag or Conversions API	Budget
Meta	Conversion Lift in Experiments	Spend and conversion thresholds set in Ads Manager	Lead volume on a bottom-funnel event

Sources: Google Ads Help, LinkedIn Marketing Solutions Help and Meta Business Help Center.

Google lowered its barrier sharply. John Chen, Senior Director of Product Management for Ads Measurement at Google, wrote in that November 2025 announcement that "What once might have cost upwards of $100,000 for a single experiment can now be done for $5,000." Google also says methodology updates produce conclusive results up to 50% more often. Two caveats from the setup page matter, though. Conversion Lift "isn't available for all Google Ads accounts," and Google tells you to contact your account representative to use it. And conversions with supplementary data don't count toward the 1,000.

That conversion floor is the hinge for SaaS. An account that observes 80 demo requests a month would need more than a year of them to reach 1,000. The tempting workaround is to pick a soft conversion, like a pricing-page view, just to qualify. Don't. You'd get a clean answer to a question nobody on the leadership team is asking.

LinkedIn's floor is budget. If your LinkedIn spend is $15,000 a month, $80,000 is more than five months of it. And it all has to land in one 30 to 90 day window. Meta runs Conversion Lift inside Experiments in Ads Manager with its own spend and conversion eligibility, so check the current thresholds on its help page before you plan around them.

What can you test when you miss the platform floors?

When you can't run a platform lift study, run the experiment yourself at the region level. Three designs cover most B2B SaaS questions, and none of them needs the platform's permission or a conversion floor.

1. Geo holdout: Turn a channel off in a set of regions and compare their outcomes with a synthetic control built from the regions where it stayed on. This is the workhorse.
2. Heavy-up: Raise spend in test regions instead of cutting it, which tests whether more budget produces more pipeline without giving up any current pipeline.
3. Time-based on/off: Pause a channel everywhere for a set window and compare the result with your forecast. It's the cheapest design and the crudest, because anything else that changed those weeks rides along in the result.

The analysis tooling is free. Meta's GeoLift is an MIT-licensed R package built on synthetic control methods, with power calculators and market selection that pick test regions from your history. Google's Meridian GeoX is an open-source library for publisher-agnostic geo experiments. It supports holdback, go-dark and heavy-up designs and converts results into priors for an MMM later.

Synthetic control is the method underneath both. It builds a weighted blend of untreated regions that tracks your test regions before launch, then uses that blend as the counterfactual afterward. The GeoLift methodology documentation quotes economists Susan Athey and Guido Imbens calling synthetic control methods "arguably the most important innovation in the policy evaluation literature in the last 15 years."

Geo tests have real limits, and B2B adds a few. Mike Taylor, writing for Recast, named the business cost directly: "turning off an ad channel for several weeks means you lose sales in those regions you otherwise would have made." In SaaS I'd add three more. Buyers work remotely, so key the test to the lead's location rather than the company's headquarters, and use the same field on both sides. Tell sales which regions are dark, because an outbound push into the holdout erases the difference you're trying to measure. And if a few dozen accounts make up the whole pipeline, a geo split is too lumpy to read.

Which incrementality test should a B2B SaaS team run first?

Start with a geo holdout on the channel you're least sure about, usually non-brand search or paid social, and move to heavier methods only when the cheaper one stops answering your question. I call that progression the Lift Ladder.

1. Rung 1, time-based on/off: Pause one channel for a set window. Use it for direction only, when the stakes are small.
2. Rung 2, geo holdout: GeoLift or GeoX, read against CRM outcomes by region. This is where most Series A to C teams should spend their testing time.
3. Rung 3, platform lift study: Google, LinkedIn or Meta, once a channel clears that platform's floor. It randomizes users, so it's the cleanest read on that one platform.
4. Rung 4, calibrated MMM: Google's Meridian or Meta's Robyn, once you have years of weekly spend and outcome data by region, with your test results as priors.

Each rung answers a broader question with more confidence, at a higher cost in data and time. The common mistake is starting at rung 4 because it looks like the grown-up option, then spending a quarter assembling data the model can't use yet.

Which outcome should a B2B SaaS lift test measure?

Measure the test on an outcome from your CRM, read by region and week: demo requests, SQLs, opportunities or pipeline dollars. Platform conversions are the wrong yardstick because each platform only counts what it can see, and a geo test needs a business-level number that every channel feeds.

That principle isn't specific to SaaS. It's how I approached measurement for a Texas residential energy sales organization, where every enrollment source counted against the whole channel mix instead of each platform's own count.

For SaaS, the setup has three parts. Every lead needs a region field you trust, captured on the form or from enrichment. Every funnel stage needs a date, so you can count SQLs and opportunities by week. And your offline conversions should flow back to the ad platforms anyway. Google's Conversion Lift setup page recommends enhanced conversions for leads before a study.

The sales cycle creates a lag. Opportunities often trail the first touch by weeks or months, so I'd pick a leading indicator for the in-window read (demo requests or SQLs). Then schedule a second read on pipeline after your median lead-to-opportunity time has passed. Report both, and say which one the budget call rests on.

How long and how big does a B2B test need to be?

Size the test from your own history before launch, because a geo test without a power analysis often ends inconclusive even when the channel works. The GeoLift documentation makes the same point: geo-test effects are usually small, and skipping the power analysis leaves a high chance of failing to find lift that actually happened.

Three reference points help. Google's Conversion Lift setup reports study power as a certainty level from 50% to 95%. It recommends aiming for 90% and treats 50% to 90% as directional. For geo tests, Recast's Taylor described a typical design as 3 to 5 weeks of test data against a clean 4 to 8 week pretest period. LinkedIn Conversion Lift Testing requires at least 30 days.

Here's an illustration of why volume matters. It's a hypothetical company, not a client. Picture a Series B SaaS business with 200 demo requests a month across all channels, 70 of which become SQLs, and $30,000 a month in non-brand Google search. You hold out non-brand search for six weeks in regions that produce 25% of that volume.

Over those six weeks the holdout regions would normally produce roughly 70 demo requests and 24 SQLs. If non-brand search drives 20% of them, the holdout loses about 14 demos and 5 SQLs. A 5-SQL gap is easy to lose in weekly noise at about 4 SQLs a week. A 14-demo gap is far easier to see. In this case I'd read the test on demo requests, carry SQLs as a secondary metric and let the power calculator decide whether six weeks or a larger holdout is needed.

The arithmetic for the budget call is simple. The search spend withheld in those regions is about $10,400, which is 25% of $30,000 a month for six weeks. Divided by the 14 demos lost, that's an incremental cost per demo of about $740, and that's the figure this test can actually support. The same math on the 5 SQLs lost gives roughly $2,100 per incremental SQL, but it rests on a handful of SQLs, so report it with the test's interval rather than as a single number. Then divide the cost per SQL your attribution report shows by the incremental cost per SQL, and you have the channel's incrementality factor. If attribution shows $800 per SQL, the factor is about 0.38. That means roughly 38% of the SQLs attribution credits to non-brand search were incremental.

What does incrementality testing cost without an MMM?

Incrementality testing costs very little in software and a fair amount in data discipline and patience. GeoLift, GeoX, Meridian and Robyn are all free and open source. What you pay for is held-out pipeline, analyst time and weekly outcome data by region.

Method	Software cost	Data you need	What it answers
Geo holdout (GeoLift or GeoX)	Free, open source	Weekly outcomes by region, with a clean pre-period before the test	Is this channel causing pipeline right now?
Platform lift study	No license fee, study budget required	Enough volume and spend to clear the platform floor	Is this campaign causing conversions on this platform?
MMM (Meridian or Robyn)	Free, open source	Years of weekly spend and outcomes, ideally by region	How should budget split across every channel?

The MMM gap is data, not license fees. Google's Meridian documentation works through a national model with 12 media channels, six controls and eight knots. That's 26 parameters. Two years of weekly data gives 104 data points, or four per parameter. The docs call that sample "too low to estimate the model reliably." A two-year-old SaaS company whose channel mix changed every quarter doesn't have that history yet.

MMM still has a place once you do. In a 2023 Harvard Business Review article, Julian Runge of Northwestern and Harpreet Patter and Igor Skokan of Meta's Marketing Science team advised marketers to "regularly calibrate your MMM using ad experiments to make sure your measurements are accurate." That's the order I'd follow: tests first, then an MMM the tests calibrate. Kantar found 31% of large advertisers now manage MMM internally, so it's a reachable step later, not a vendor contract you need today. Once you have incrementality factors by channel, a question like the Meta vs Google budget split becomes one you can answer with your own numbers.

How do you know the test worked?

A test worked when the control tracked the test regions before launch, nothing else changed during the window and the result's interval excludes zero. Run these checks before the number goes in a budget deck.

1. Check pre-period fit. The synthetic control should track your test regions closely before launch, and if it doesn't, the lift number isn't trustworthy.
2. Confirm a clean window, with no pricing change, product launch, outbound push into test regions or other channel change during the test.
3. Read the statistics. The confidence interval should exclude zero, or Google's certainty level should reach the threshold you set.
4. Sanity-check the size. A result implying a $40 SQL from a channel that has never produced one under $1,000 needs a second look.
5. Attach a decision. Turn the result into an incremental cost per SQL, write the incrementality factor into next quarter's plan and pick the next channel to test.

Then repeat. A single test is a point-in-time read, and channel performance drifts as your market, competitors and creative change. If you want a second set of eyes on test design or the CRM plumbing behind it, that's the work I do in measurement and attribution.

Frequently asked questions

Is incrementality testing the same as A/B testing?

No. An A/B test compares two versions of something, like two ads or two landing pages, and tells you which performs better. An incrementality test compares marketing with no marketing and tells you whether a channel or campaign causes outcomes at all.

How much budget does an incrementality test need?

A geo holdout has no minimum budget, because you're withholding spend in some regions rather than buying a study. Google Conversion Lift needs a $5,000 minimum study budget plus 1,000 observed conversions, and LinkedIn Conversion Lift Testing needs $80,000 across the test campaigns.

Do I need an MMM before I run incrementality tests?

No. Tests come first. They answer channel-level questions with far less data, and their results become the priors that calibrate an MMM once you have enough weekly history to build one.

How often should a B2B SaaS team run incrementality tests?

I'd test each major channel at least once a year, and again after any big change in spend, targeting or creative. One channel per quarter is a sustainable pace for a small team, and it builds a full set of incrementality factors within a year.

Can I run an incrementality test on LinkedIn Ads below the $80,000 floor?

Yes, with a geo holdout. LinkedIn campaigns can target by location, so you can hold out a set of regions and read the difference in your CRM. You give up LinkedIn's user-level randomization, so plan on a larger holdout and run a power check first.