Marketing Automation ROI: You Are Measuring the Wrong Denominator. Gartner Says Only 15% Can Prove a Return.

Every marketing automation ROI pitch quotes the same number: $5.44 back for every dollar spent.

Go find where it came from. It is a Nucleus Research brief published in March 2021 that reviewed 16 ROI case studies published between 2016 and 2020. Sixteen. Published by the vendors. No control group, no counterfactual, nothing showing what those companies would have produced without the software.

That is the most-quoted marketing automation ROI figure in the entire category. It is five years old and the sample fits in a conference room.

Now put it next to this: Gartner’s 2025 Marketing Technology Survey puts martech utilization at 49%, and finds only 15% of organizations qualify as high performers — the ones that hit their strategic goals and demonstrate positive ROI.

Both things are true at once. Marketing automation does return money. Almost nobody can prove theirs does.

The Denominator Nobody Puts in the Equation

Here is the standard marketing automation ROI calculation, the one on every vendor slide:

Attributed revenue ÷ software cost

The cost side of that equation contains exactly one item: the license. Which is, by a wide margin, the cheapest input in the system.

The expensive input is human attention. Salesforce’s State of Sales research has reps spending roughly 70% of the week on things that are not selling — admin, research, internal meetings, manual data entry. If you are carrying a rep at $85K plus commission, their selling hours are the scarce resource. Not the $1,200-a-month platform.

So when your automation stack produces 400 more MQLs this quarter, you did not generate ROI. You generated work. You increased consumption of the most expensive input you own, and you will book it as a win because the dashboard went up and to the right.

That is the whole reason ROI is unprovable for most teams. They measure the send side while paying for the decision side.

Our Thesis: ROI Is a Routing Question, Not a Software Question

Our thesis is that marketing automation returns money in exactly one way: it changes who a human talks to next, and how fast.

Everything else — sends, workflows, nurture tracks, lifecycle stages, the whole lovely apparatus — is plumbing. Useful plumbing. But plumbing does not close deals. A rep on the phone with someone who is actually in-market closes deals.

Which reframes the entire question. It is not “did the platform work.” It is: did a human make a different call this week than they would have made without the system, and was that call better?

If the answer is no, you did not buy a revenue system. You bought a newsletter machine with a reporting tab.

The Architecture We Would Build to Make ROI Measurable

Here is the architecture we would build. Not a case study — the design.

ICP list → Outbound sequence → Engagement capture → Weighted decaying score → Threshold breach → HOT LEAD ALERT with reason codes → Human call

Capture signal, not just email metrics. Opens are weak. A pricing page view three weeks after the last send is strong. So is a return visit. So is a click on one specific service page. Email engagement on its own is a thin signal; the site behavior sitting behind it is where the intent actually lives. We wire TAM mining and outbound campaigns into the same event stream as website behavior and retargeting and paid ad engagement, so one contact record accumulates everything instead of five tools each holding a fragment.

Weight the score, then decay it. A click from 90 days ago is not worth a click from yesterday. Scores that never decay quietly become a leaderboard of your most loyal newsletter readers, which is a fine thing to have and a terrible thing to call a pipeline.

Put reason codes on the alert. This is the part almost everyone skips. “Score: 87” tells a rep nothing. “Score 87 — viewed /pricing twice in 48 hours, opened 3 of the last 4, clicked the case study” tells a rep how to open the call. Our sales lead scoring system exists to produce that second sentence. The number by itself is decoration.

Write it back into the CRM. If the score lives in the email platform and the rep lives in Pipedrive or HubSpot, the system does not exist. CRM automation is the difference between a signal that is actionable and a signal that is a screenshot in a monthly deck.

The Economics: Cost Per Correct Call

Here is the metric we would put in place of cost per MQL.

Cost Per Correct Call (CPCC) = (stack cost + fully loaded rep hours consumed) ÷ number of calls that reached someone with a live buying signal

Illustrative example — run your own numbers.

Two reps at $85K base, call it $60/hour fully loaded. Each spends 20 hours a week on outbound calling. That is 40 rep-hours a week, $2,400 a week, roughly $10,400 a month in calling time alone. Add a $1,500/month automation and data stack. Total input: $11,900 a month.

Scenario A — list sorted by “last contacted.” 800 dials a month. Say 5% land on somebody with a live signal. That is 40 correct calls. CPCC = $297.50.

Scenario B — same reps, same hours, same 800 dials, list ranked by scored engagement. Say 20% land on somebody with a live signal. That is 160 correct calls. CPCC = $74.38.

Same cost. Same headcount. Same dial count. Four times the correct calls.

Notice what did not change: lead volume. Nobody bought more leads. The only variable that moved was call order.

And notice that the standard equation cannot see any of it, because software cost was identical in both scenarios. Revenue divided by license fee will report the same ROI for a system that quadrupled rep effectiveness and a system that did nothing.

What the Evidence Actually Supports, and What It Does Not

An honest read of the research, including the parts that do not flatter us.

The direction is well supported. Harvard Business Review’s analysis of 2.24 million sales leads found that firms responding within an hour were nearly seven times more likely to qualify the lead than firms that waited just 60 minutes longer. We are citing a 2011 study on purpose and flagging the date: it remains the largest published dataset on the question and nothing bigger has replaced it. Treat it as directional, not as a 2026 benchmark.

What is not supported is any specific ROI multiple. The $5.44 figure is a benefit-side average pulled from vendor-published case studies with no counterfactual. That is not a lie. It is also not evidence that your deployment returns $5.44, and anyone presenting it as a forecast is selling.

The utilization trend is worse than the headline suggests. Gartner has tracked martech capability utilization falling from 58% in 2020 to 42% in 2022 to 33% in 2023 before a partial recovery. Whatever the theoretical return is, most organizations are not collecting it. And in October 2025, Gartner reported that 45% of martech leaders say vendor-supplied AI agents have failed to deliver the business performance that was promised — which should calibrate exactly how much weight to put on the next ROI slide anybody shows you, including ours.

Where This Gets Uncomfortable

Three admissions, and the second one costs us money.

One. If you build the scoring layer and discover your database produces almost no meaningful engagement signal, the system did not fail. It handed you a diagnosis. You have a top-of-funnel problem, and no routing logic on earth fixes an audience that is not paying attention. Fix reach first — content and distribution before scoring — or you will have built a beautiful instrument pointed at silence.

Two. If you have one or two salespeople and fewer than a couple thousand contacts, you probably do not need this. A competent human holds 200 relationships in their head just fine. Prioritization software earns its keep only when the number of people producing signals exceeds what a person can track — call it a few thousand contacts plus enough campaign volume to generate real behavioral data. Below that line, buy the CRM, skip the scoring engine, and go call people.

Three. CPCC is harder to calculate than cost per lead, which is precisely why nobody uses it. You need call disposition data and you need reps who actually log outcomes. If your team will not log calls, you cannot measure this, and the logging discipline is the thing to fix before you buy any software at all.

The Playbook: Calculate Your Own CPCC This Week

You do not need us for this part. Steal it.

1. Pull 90 days of outbound calls from your CRM. Export dials, connects, and outcomes by rep.
2. Tag which calls hit a live-signal contact. Define it narrowly: opened or clicked in the last 14 days, OR viewed a pricing or services page, OR replied. Everything else is a cold dial.
3. Compute your current CPCC. Fully loaded hourly rate × outbound hours, plus stack cost, divided by live-signal calls. Write the number on a whiteboard.
4. Build a five-input score. Pricing page view, services page view, email click, repeat visit inside 14 days, form fill. Weight them 5 / 3 / 2 / 3 / 8. Halve the score at 30 days. It does not need to be clever. It needs to exist.
5. Sort next week’s call list by that score instead of by last-contact date. Change nothing else. Same reps, same hours, same script.
6. Recompute CPCC after 30 days. If it did not move, your signal inputs are wrong — not the idea.

That is a real experiment with a real control period, which is more methodological rigor than the $5.44 study ever had.

The Honest Answer on Marketing Automation ROI

Marketing automation ROI is real, badly measured, and almost never attributable to the software itself. It is attributable to whether the software changed a human decision. If you cannot point to a call that happened because of the system and would not have happened without it, you do not have ROI. You have a subscription.

Your prospects are already talking. Build something that listens. Connect the systems. Track the behavior. Score meaningful engagement and let it decay. Compress thousands of activities into a short, ranked list of humans worth calling today. Then measure the number that actually maps to revenue — cost per correct call — instead of the one your vendor put on the dashboard.

Want to build it yourself? Steal the framework above; it is the same one we run. Want it bolted onto the stack you are already paying for, with a fractional CMO running the build instead of another headcount? See what this costs → or work with Jeremy directly for 90 days.