The Marketing Engineering Audit: 20 Questions That Show Where Your Growth System Breaks

Most companies hire the builder before they know what to build

A job ad is written before anyone knows what the job will find. That is fine for a sales rep. It is expensive for a marketing engineer.

Figma's posting for a Marketing Engineer, AI Deployment says success is measured three ways: agents live, hours saved and revenue driven. Every one of those is a change from a starting point, so the company needs the starting point before the hire walks in.

That starting point is what this audit gives you. Twenty questions, one point each, across the systems described in what a marketing engineer is. One afternoon, three people, no software.

It will show you where your growth system breaks. Not where it feels weak. Where it breaks, with proof.

Why an audit comes before a hire, a vendor or a tool

The failure numbers on AI projects are loud, and they point at the same cause. Fortune's write-up of an MIT report says about 5 percent of AI pilots reach rapid revenue gains, while the rest stall with little or no measurable effect on profit and loss.

Gartner predicts that over 40 percent of agentic AI projects will be canceled by the end of 2027, and two of the three reasons it gives are unclear business value and costs that keep rising. Both get worse when nobody wrote down the number before the build.

Read those as predictions and surveys, not laws. They still describe a pattern any owner has seen: a tool bought for a problem nobody measured.

An audit is the cheap fix. It names the number, the gap and the system closest to revenue, so the first 90 days have a target, which is the plan laid out in The First 90 Days With a Marketing Engineer.

The MIT report's lead author made the same point from the other side. The young companies doing well, he told Fortune, pick one pain point and execute well.

One pain point. That is what the score is for.

The Five-Minute Proof

Ask a marketing team whether they track lead source and you will hear yes. Ask them to open five leads and show it, and the room gets quiet.

So the audit has one rule. A yes counts only if someone can open the proof within five minutes: a file, a report, a log, a dated memo.

"We have that somewhere" scores zero. "Sam does that" scores zero, unless Sam's work is written down where the next person can find it.

The rule is strict on purpose. Engineers measure how fragile a team is with the bus factor, the number of people who could leave before nobody knows how something works, and any question where the answer lives in one person's head has a bus factor of one.

We call this the Five-Minute Proof. It turns an opinion survey into an inventory.

The 20 questions, by system

The questions run in the order you should fix them. Four foundations first, because without them no other system can be measured, then the six systems.

Those six come from Greg Isenberg's episode on the marketing engineer role: customer truth, founder content, outbound signals, creative testing, AI search visibility and a weekly growth cockpit. We split them into two or three questions each, chosen so the answer is a thing you can open.

Copy the list into a sheet. Each line gets a 0 or a 1, a link to the proof, and the name of whoever owns it.

Be harsh with yourself. A half-yes is a no, and a no today is the most useful line on the page.

Foundations: four questions that decide whether anything else can be measured

Questions 1 and 2 are tracking. If a lead arrives without its source, you cannot say which ad worked, and if a closed deal never flows back, the ad platforms optimize for form fills instead of customers.

Google's offline conversion import lets you track what happens offline, such as a signed contract, after someone clicks an ad, and Meta's Conversions API does the same job from your server. The setup, step by step, is in server-side conversion tracking.

Then check the path in between. A Google Analytics funnel exploration shows each step a visitor takes and where they drop off, once the steps are defined as events.

Question 3 is speed. Harvard Business Review researchers sent test leads to 2,241 US companies and timed the replies. 586 answered within five minutes, and 512 never answered at all.

The same authors found that firms trying to reach a lead within an hour were nearly seven times as likely to qualify it as firms that waited even an hour longer. The study is from 2011, and the gap it found is the kind a script closes.

Question 4 is memory. If your customer quotes, rules and past tests live in a dozen chats and one person's head, every new agent and every new hire starts from zero. The fix is one folder, described in The Growth Repo.

The six systems: what a real yes looks like

Customer truth, questions 5 to 7. A yes is a dated weekly summary where every line carries a quote or a count, and an ad that reuses a customer's own words. How to build the weekly memo is in the customer truth system.

Founder content, questions 8 and 9. A yes is one recording turned into several pieces, ranked afterward by a number that touches pipeline. Likes alone do not count.

Outbound signals, questions 10 to 12. A yes means every first email had a reason, a person approved the sequence, and the team counts replies that became meetings.

Count sends instead and you get more sends. That is Goodhart's law: when a measure becomes a target, it stops being a good measure.

Creative testing, questions 13 to 15. A yes is a finish line written before launch, tests that compare different buyer reasons, and a log that keeps the losers. Without the losers, the log only tells you what you hoped.

AI search, questions 16 and 17. A yes is a fixed list of buyer questions run on a schedule, with a count of the sources cited. Google says there are no special optimizations needed to appear in its AI Overviews, so the work is knowing what gets cited for your questions, not hunting for a trick.

Growth cockpit, questions 18 to 20. A yes is four Monday memos in a row, numbers a script can reproduce, and a written owner and metric per system. Search numbers such as clicks and impressions come straight from the Search Console performance report, and the memo format is in The Growth Cockpit.

Scoring: what a 6 out of 20 means

Add the points. The total is out of 20, and the bands below say what to build first.

Here is an illustrative company at 6. It has lead source in the CRM, a shared folder, a test plan before launch, a results log, a list of AI questions, and an owner list.

It has no closed-deal feedback, no five-minute reply, no weekly customer summary, and no Monday memo. Its ads are optimizing for form fills, and nobody reads what buyers say.

That company does not need six systems. It needs question 2 fixed this month, because every other system will be judged against revenue it cannot yet see.

Then it builds one system: the lowest-scoring group whose number it already tracks. For most companies at 6, that is customer truth or the cockpit.

A low score is not bad news. It is a short list.

"This is an agency lead magnet with extra steps"

Partly true, and we would rather say it first. Our engagements start with an audit, and we wrote about why audits make good lead magnets in Customized Audits as Lead Magnets.

Here is the part that is not a funnel. The 20 questions work without us, there is no form on this page, and every question can be answered by your own team in an afternoon.

If you score 16 or more, you do not need an outside team for this. Spend the money on your next experiment.

If you want a paid first step that is smaller than a full engagement, the structure we prefer is in Quick Win Offers. Either way, the score is yours to keep.

Run it this afternoon: a 90-minute plan

Minutes 0 to 10. Get three people in a room: whoever owns marketing, whoever owns sales, and whoever touches the CRM most. Share the sheet with the 20 lines.

Minutes 10 to 70. Go line by line. For each one, the owner opens the proof on screen or the line scores zero, and no line gets more than three minutes.

Minutes 70 to 80. Add the points and read your band. Circle the foundation questions that scored zero.

Minutes 80 to 90. Pick one thing: the zero foundation that touches revenue, or the weakest system whose number you already track. Write its owner and its number on the sheet.

Put the date on the sheet and run it again in 90 days. The change in the total is your first honest measure of whether the work is working.

When the audit points at the plumbing, lead routing and CRM fields, that work sits under CRM and RevOps. When you want the deeper version, where we read your actual accounts, calls and data, that is weeks one and two of our marketing engineering engagements, followed by one system built and measured before the next.

What to read next, and why

Traffic Is Never the Problem: A Conversion-First Audit [audit]. The companion check for the offer chain. Run it if question 2 or 3 scored zero.

How to Evaluate a Marketing Engineer's Work [scorecard]. What to track once a system exists. This audit tells you what to build; that page tells you if it worked.

How We Run a Marketing Agency as Marketing Engineers [proof]. The same rules run on our own business, with the checks and the numbers.

So the simple takeaway is this: a yes you cannot show in five minutes is a no, and the first no tied to revenue is your first build.