But everybody did their job...

So when growth slows, we go looking for the department responsible. Pipeline’s weak, that’s a marketing problem. Deals won’t close, sales problem. Customers don’t stick, product problem. Delivery’s slow, engineering problem. It’s a reasonable way to think, and I’ve watched it be wrong more often than right.

The problem is often not where the bad number appears. It’s upstream, in the seam, or one floor up from where everyone’s staring.

AI is about to make that harder to see, not easier. As producing things gets nearly free, the value doesn’t disappear. It moves to judgment: the ability to read the whole system instead of one department’s slice of it, and the willingness to hold a problem across boundaries no single department owns.

I’ve spent most of my career moving between engineering, product, marketing, operations, and the leadership seat above all of them. The thing you notice after enough of that is that the expensive problems, the ones that actually cost you a year, rarely live cleanly inside any one box. They live in the gaps. A marketing problem that’s really a product problem. A product problem that started in a sales call. An engineering velocity problem that started with a leadership decision nobody wrote down. A retention problem that was set in motion before the customer ever paid you.

Every department can be doing a perfectly reasonable job while the company as a whole produces a bad result.

We built companies around lines that problems don’t respect

Departments exist for good reasons. Specialization works. An engineering leader understands things about architecture, testing, and delivery that you shouldn’t expect a marketer to understand, and the marketer understands positioning and persuasion in a way most engineers don’t. That’s fine. That’s the point of them.

The trouble starts when we assume the org chart also describes where problems begin and end. It doesn’t.

Say engineering velocity has dropped. The natural move is to look inside engineering. Maybe the codebase got tangled. Maybe there aren’t enough tests. Maybe people are stepping on each other, or the technical debt finally came due. All plausible, and worth checking. But what if engineering is slow because the requirements keep moving? Or because priorities aren’t clear? Or because the company decided to build a complicated answer to a problem that had a simple one? Now engineering’s dashboard shows a delivery problem, and the actual cause is sitting one floor up, upstream, where nobody’s looking.

At a travel SaaS company I worked with, we kept hearing a version of the same thing from sales: nobody will buy this without AI. So we did the logical thing. We invested in AI. We built recommendation technology and put real engineering effort behind it.

Nobody bought it anyway.

AI had never been the reason they weren’t buying. We had accepted a sales diagnosis as a product requirement without getting far enough underneath it. Engineering delivered what was asked for. Product could point to the new capability. Sales had the feature it said it needed. Everybody did their job, and we had solved the wrong problem.

It runs the other way too. An engineering team can optimize its own process beautifully and create enormous friction for implementation, support, and sales in the doing. Locally efficient. Globally terrible. That’s exactly the kind of thing a departmental metric is built not to show you.

Every department sees the company through its own window

This isn’t because functional leaders are incompetent or territorial. Usually it’s the opposite. They’re good at what they do, and they diagnose with the tools they’re good with.

An engineer who becomes an engineering director can look at a delivery system and feel a subtle problem in QA or CI almost in their body, before they can explain it. Ask that same person to spot a subtle positioning problem and it gets a lot harder. The marketing leader has the exact inverse problem. Nobody’s failing. Everybody’s reaching for the instrument they trust.

So when a genuinely cross-functional problem lands in a room with five executives, you can get five completely reasonable diagnoses. Marketing needs better campaigns. Sales needs better leads. Product needs better positioning. Engineering needs clearer requirements. Customer success needs a better product. And here’s the part that makes these things so hard: everyone in the room is right. Each one is describing a real thing they can see through their own window. None of them is describing the whole.

Marketing and sales aren’t even really a handoff

We draw the customer journey as a funnel. Marketing makes demand, hands it to sales, sales closes, onboarding takes it, success keeps them. Nice clean arrows. Real companies almost never work that way.

Take marketing and sales. I’m not even sure “handoff” is the right word. Marketing is building awareness, running campaigns, trying to pull the right people into the top of something. Sales is talking to live prospects right now, finding out what they actually care about, trying to close. Those are two systems running at the same time, and you’re hoping they point in roughly the same direction. You are not guaranteed that.

Marketing might still be fishing for the buyer that worked eighteen months ago while sales has quietly figured out that a different buyer is the one signing. Sales might be selling a promise product doesn’t think is strategically important. Product might have changed underneath the story without anyone changing the story. Customer success might know precisely why people leave, and that knowledge might never once touch the other three. The problem isn’t inside any of those functions. It’s the distance between them. That’s the thing nobody owns.

AI drops a room full of interns under every department

That was already hard. AI makes it harder, and in a way that’s easy to miss.

The way I’ve started picturing it: AI puts a room full of fast, cheap, eager interns underneath almost every department. They’re capable. They produce an astonishing amount of work. And they don’t actually know what’s true this quarter.

A marketing team probably has years of strategy docs, personas, campaign plans, decks, transcripts, and old content sitting in its systems. Ask an AI to build a new campaign and it will synthesize all of that history, beautifully. Which is the problem. Something that mattered a year ago may not matter now. The product moved. The buyer moved. Sales learned something. A competitor did something. But the model doesn’t know which parts of the company’s own history are dead unless somebody did the work of telling it. So the brand-new plan comes out looking a lot like last year’s plan. Not because anyone copied it. Because the machine reconstructed it from the past and handed it back looking new.

The obvious mistakes get caught. It’s the plausible ones that get through. The slightly stale positioning. The feature still described as the headline. The customer profile that drifted a few degrees. The claim that was true enough to survive review two years ago, now getting copied into fifty pieces of content. And then those outputs become inputs. Summarized, rewritten, turned into ads and emails, fed into the next model. The organization starts producing information faster than it can check whether that information still describes the organization. We used to worry about alignment drift as a slow, human thing. Now we can automate it.

“Pretty good” quietly changes the economics of quality

There’s a second AI effect I don’t hear talked about enough.

When making something was expensive, we inspected it. When making something takes thirty seconds, the pull is to fire and forget. And most AI output is pretty good, which is precisely what makes it dangerous. Terrible output gets rejected at the door. Pretty-good output ships. It’s good enough that nobody stops.

This is where “pretty good” and “nobody’s checking the seams” meet, and it gets expensive fast.

None of this started with AI. I once looked at an agency’s dashboard reporting 11.7x return on ad spend. Pretty good. Nobody upstairs was questioning it, because why would you question a number that good. But when I traced that 11.7x back to the actual orders it was crediting, the campaign had produced $918 in real sales. The multiple was counting conversions the ads had touched but not caused, so it looked like a fortune and reconciled to almost nothing. The dashboard wasn’t lying exactly. It was measuring something that had quietly come loose from money, and everyone had been reading it for months.

AI doesn’t create this problem. It changes its scale. The same thing is now happening across engineering, marketing, sales ops, support, internal analysis, all of it at once. We’re cranking up the volume of last-mile work while quietly dialing down the human attention on each piece of it. So the bottleneck moves. Generation gets cheap. Judgment gets expensive. And quality assurance stops being a thing each department does to its own output and becomes a problem that belongs to the whole company, which means it belongs to nobody yet.

Treat the company as one system, not an org chart

One way to find these problems is to stop starting with departments and start with the flow of value instead.

Follow the customer all the way through the company. Someone finds you. They form an expectation. They talk to sales. They buy. They onboard. They use the thing. They get support. They decide whether they got what they thought they were buying. They renew, expand, or leave. The question isn’t whether each of those steps looks good on its own. It’s whether they connect, whether the customer can actually get from one end to the other without falling through a gap between two teams.

Now lay the data under that journey instead of reporting each step separately. Funnel data, product usage, sales conversations, support tickets, retention, revenue, and time. Read it across the department lines, because that’s where it usually isn’t read. Service designers have done a version of this for years: map the experience, map the machinery under it, and pay unusual attention to the seams. I think that’s becoming one of the more useful ways to diagnose growth. The interesting questions are all about the seams. Where does the work leave someone’s hands? Where does ownership change? Where does a number quietly change its definition? Where does the customer’s expectation shift? Where does something get summarized instead of actually passed on? Where does one team’s output become another team’s input? That’s where organizations lose information, and time, and money, without anyone posting a bad number.

Be suspicious when every dashboard is green

One of the stranger things you can walk into is a company where every department can prove it’s winning and the company is still missing the number.

Marketing hit its lead target. Sales hit its activity target. Engineering’s velocity is up. Product shipped the roadmap. Success is inside its SLA. Revenue hasn’t moved.

When you’re there, arguing harder about any single one of those numbers won’t help you, because none of them is lying. The question isn’t whether the numbers are good. It’s whether they connect. Are we marketing to the people sales can actually close? Are we building the thing those people care about? Are they using what we built? Does using it produce the outcome sales promised? Does that outcome turn into renewal or expansion? Are we even tracking the same customer as they move across the company, or does the customer get a new identity every time they cross a border?

Measurement should cross departmental lines because value crosses them. If it doesn’t, you can build a company where everyone hits their target and the business still loses.

Cross-functional problems need closed-loop ownership

Which raises an awkward question. Who owns a problem that needs marketing, product, engineering, and sales to all change something at once?

I don’t think there’s a universal title for it. Sometimes it’s revenue operations. Sometimes product. Sometimes the CRO, sometimes the CEO. And honestly, sometimes it’s just the one person who noticed and refused to let go of it. The title matters less than the loop. Somebody has to own the problem from the moment it’s spotted to the moment the company can show it’s actually resolved. That person needs enough visibility to follow it across departments, enough judgment to notice when the problem quietly changed shape on them, and enough authority to make sure something actually happens.

Without that, the problem just gets passed around the building, and every ticket is closed while the customer still has the same problem. Everyone did their job, and the loop was never held.

The org chart is not the system

We’re very good at optimizing the things we can name. Marketing, sales, engineering, product, success. We hire leaders for them, give them budgets, build dashboards for them, set their goals. Then we wonder why a problem that crosses four of those boxes is so hard to kill.

Growth doesn’t happen inside the boxes. It happens in whether the boxes connect. That was true before AI. But AI speeds up how fast information gets created, transformed, and acted on. It lets departments move faster while talking to each other less. It can preserve yesterday’s assumptions while dressing them up as new. And it makes “pretty good” cheap enough to put almost everywhere. Which brings the whole thing back to where I started: the piece that doesn’t get cheaper is judgment, the ability to look across the whole system without being trapped inside one department’s reading of it. And that is exactly the piece that lives between the boxes, where nobody’s looking.

Sometimes marketing really does just need better marketing. Sometimes engineering really does just need to ship faster. But when every team looks like it’s doing its job and the company still isn’t getting the result it wants, I’ve learned to stop staring at the departments.

I look at the space between them.