"What's our ROI?" has five questions inside it

A CEO once asked me for my north star in paid advertising, and I took about five tries to give it to him.

He runs a B2B SaaS platform that's past its Series B, doing eight figures in annual revenue, with a plan to multiply that over the next couple of years. We were talking through how they'd get there, the channels that had stalled and needed rebuilding, and where the data should come from. Then he leaned in with what was clearly the real question: if you were running Google Ads for us, what number would you manage toward?

I started talking about portfolios. In a complex B2B sale with a buying committee, I said, you don't judge a channel on its own any more than you'd judge one stock in a fifty-name portfolio. You look at how the whole mix performs over time.

He asked again, a little narrower. Is there a north star in paid?

Revenue, I said. Are we adding paying accounts efficiently enough that the channel isn't losing money.

He asked how I'd measure "efficiently enough." I said customer acquisition cost against lifetime value, with some nuance about attribution. He asked one more time, and said outright that he was testing me, for the metric. Then he gave up waiting and answered it himself.

"That's return on ad spend, right? ROAS?"

"Yes," I said. "With an asterisk."

The thinking was right and the order was wrong

I'll own the delivery. He wanted a number and then the caveat, and I gave him the caveat four times before I let a number out. If you've ever asked someone a yes-or-no question and gotten a framework back, you know how that feels from the other chair. I was that guy for about six minutes.

The direction held up, though, and the asterisk is the part worth writing down. A lot of leadership teams and early-stage founders end up managing to the instrument instead of the result, and "what's our ROI?" is where it usually shows up.

Revenue is the result, everything else is an instrument

The north star for marketing, sales, and the operations underneath them is revenue. More precisely, it's revenue you can acquire at a cost the business can live with, given how long customers stay.

Everything else is an instrument. Cost per click, click volume, form fills, MQLs, demo bookings, even channel-level ROAS. They're useful, and I look at all of them, because they move faster than revenue and they tell you where something is leaking. They are also very easy to improve without improving the business. Cost per click goes down when you buy cheaper clicks. MQLs go up when you lower the bar for what counts as one. Nobody in the room is lying, and the bank account still doesn't care.

Founders reach for indicators because they're legible. They update daily, they fit on a slide, and somebody owns each one. Revenue in a long sales cycle updates slowly, gets credit from five teams at once, and arrives months after the spend that caused it. So the fast number quietly becomes the goal, and the slow number becomes something you check at board meetings.

The five questions inside "what's our ROI?"

"What's our ROI on paid?" sounds like a single number. On a long-cycle, committee-driven B2B sale it rests on at least five assumptions, and each one has its own confidence level.

1. Which touches get credit. The answer depends entirely on your attribution model. Last click gives everything to the search ad someone clicked right before booking a demo. First touch gives everything to whatever introduced them. Weighted models split the difference by rules somebody chose. Same spend, same customers, very different ROI depending on the rule.

2. Whether the window covers the cycle. If your sales cycle runs six to nine months and your attribution window is thirty days, the model is measuring the end of the journey and calling it the whole thing. Most of the work that made the deal possible happened outside the frame.

3. Who you're actually tracking. A buying committee might be five or six people. The person who clicked your ad is often the researcher, and the person who signed may never have touched a single tracked link. Your ROI is built on whichever person happened to leave a cookie.

4. What you can't see at all. Conference conversations, a podcast in the car, a peer recommendation in a private Slack group, a connected TV ad, the answer an AI assistant gave when someone asked who solves this problem. None of these carry a tracking parameter, and in a lot of B2B deals they do the heaviest lifting.

5. How long the customer stays. ROI on acquisition only makes sense against lifetime value. If customers stay five to ten years, a channel can lose money on the first contract and still be the best investment you make. If they churn in year one, a channel that looks profitable may not be.

So when someone asks "what's our ROI?", the honest version of the answer is a number plus a confidence level, and the confidence level is mostly a statement about your attribution model.

What channel-level ROAS does when nobody checks it

Here's the failure I was trying to describe in that conversation, in the wrong order.

A team looks at channel-level ROAS and sees that search is crushing it. The demo came through a tagged search click, so search gets the credit. Brand spend, the connected TV and the sponsorships and the content, shows almost nothing, because nobody clicks a TV ad and then fills out a form with a tracking code on it.

The logical move is to cut brand and pour the money into search. For a month or two, the dashboard looks great.

Then total pipeline drops a quarter later. Search volume starts sliding too, because fewer people were searching for you by name. The brand spend had been creating the demand that search was harvesting, and the model couldn't see it, so the team defunded the part of the machine that was invisible to the instrument.

That's why I said "with an asterisk." ROAS is a fine number at the level of the whole business. Measured channel by channel, in a sale where most of the journey is invisible, it can steer you confidently in the wrong direction.

How to give the crisp answer without lying

The CEO was right to push, and I've thought about what I should have said in the first ten seconds. Something like this:

Revenue, efficiently acquired. I'd manage to CAC payback against lifetime value, steer week to week on pipeline per dollar, and check it against win rate. And I'll tell you how much I trust the attribution behind every one of those numbers.

Lead with the number, then attach the confidence. That keeps the answer crisp without pretending the measurement is better than it is.

The other thing worth saying out loud: you never have fully accurate attribution. Nobody does. The job is shrinking the gap between what you think is working and what you can confidently defend. If the gap is big, that's your first project. Get the tagging and the CRM fields in place so you can at least trace where deals came from, ask buyers directly where they first heard of you, set a baseline, and measure the change over a month, a quarter, and a year.

Three questions to ask the next time someone asks for ROI

You don't need a new attribution tool to stop managing to the instrument. You need a few follow-up questions.

Which model produced this number, and would a different model change the decision? If last-click and first-touch point you in opposite directions, the number isn't ready to drive a budget move by itself.

Does the measurement window cover the sales cycle? If deals take nine months and you're measuring thirty days, say so on the slide.

What happened to total pipeline the last time we moved money? The portfolio view is the check on the channel view. If you shifted budget and the channel number improved while total pipeline fell, believe the pipeline.

None of these are hard. They just turn "what's our ROI?" back into the question it actually is.

The number and the asterisk

I'd still give that CEO the same answer today, just in the right order. Revenue is the north star. ROAS is a fine instrument at the level of the whole business. And every ROI number in a long B2B sale comes with an asterisk that says how much you trust the model underneath it.

The founders who get into trouble usually have the right number. They lost the asterisk somewhere between the analyst's spreadsheet and the board deck, and it turns out the asterisk was carrying most of the weight.