Local optimization breaks the system

Do not index

The Predictive Path

Course 2: Revenue as a system

Lesson 2: Local optimization breaks the system

Local optimization breaks the system

Why every function can improve its own number while the company gets worse — and why the compensation plan is the reason.

Every function can hit its own target and the company can still lose. Marketing drives more leads, sales closes them faster, support cuts its cost per ticket — each a clean local win, each sending a bill to a stage no one was watching. The parts improve and the whole gets worse, and nobody in the room is wrong. This lesson is about why optimizing a piece of a system so reliably damages the system it belongs to.

Every number up, the company down

Here is a year that happens often enough to describe exactly. Each function improves the number it is measured on:

Marketing improves cost per opportunity by a fifth and delivers more pipeline than ever.

Sales improves win rate and closes the year above quota.

Customer success drives net retention up with a strong expansion motion.

Finance tightens the close and narrows forecast variance.

Every one of those is real. Every quarterly review is a good one. And the company grows slower than the year before, at a thinner margin, with cash arriving later than planned.

Nobody is lying. The leadership team is left with a question the review structure cannot answer, because the review asks each function about its own number and every one of those answers is good: how does a year in which everyone improved produce a company that did not?

The answer is not a hidden underperformer. It is that the parts of a revenue system are connected by levers that cascade, and the compensation plan pays each function for its end of a lever while the rest of the cascade lands somewhere else, unowned and unattributed.

A lever, and what a cascade is

Course 1 established that revenue is a system whose outcomes live in the interactions between its parts. A lever is the operational form of that. A lever is any control a function can pull inside its own boundary: marketing's channel mix and acquisition spend, sales's discount latitude and contract terms, customer success's expansion targeting, finance's hiring pace. Pulling a lever moves that function's own number — that is what it is for, and that is the part the function can see.

A cascade is what happens next. The lever does not stop at the function's boundary, because the parts are connected. It moves the function's own number first, then moves the next thing downstream, then the next, until the effect reaches the company's results — often in a different function, often quarters later. The function pulling the lever sees the first step. It does not see the cascade, because no function owns the cascade. Functions own levers; the system owns where the levers land.

There is a reason the far end is unobservable at the moment the lever is pulled, and it is the object the previous lesson established. The company commits to the customer's entire lifecycle at the point of entry — and to every dimension that lifecycle will eventually settle: retention, expansion, cost-to-serve, margin, the timing of cash. But those dimensions are committed before they are observable. Cost-to-serve and second-year margin cannot be read in the quarter the deal closes; depending on the contract structure and the complexity of onboarding, they only resolve several quarters later, or not until year two. So the lever is pulled against a far end that does not yet exist to be measured. The blindness in the cascade that follows is not only organizational — it is a built-in lag between the moment a dimension is committed and the moment it can be seen.

Trace one operationally, step by step. Sales is behind on the bookings number with a quarter to close it.

If sales widens discount latitude and loosens contract terms to bring deals in, then bookings recover and the quota number is hit. This is the step sales sees, and it is the step sales is paid on.

Then the deals that closed did so on shorter terms and softer structures, which moves cash later and thins the second-year margin on that cohort. This step lands in finance, two quarters on.

Then finance sees cash arriving behind plan and margin below model, and explains it as a pricing problem or a collections problem — local explanations, because the cause is on the far side of a boundary finance cannot see across.

Then the renewal base that next year's plan assumed at a given value comes in under it, because the cohort was acquired on terms that did not hold. The forecast misses on a line nobody was watching, and the miss is attributed to whichever function is standing nearest the damage.

Four steps. The function that pulled the lever saw step one and was compensated for it. Steps two through four happened in other functions, were explained locally, and were never connected back to the discount latitude that caused them. The cascade ran in full view and was recorded nowhere as one thing, because no instrument was pointed at the place it happened and no person was accountable for the place it happened. That is local optimization breaking the global outcome — not through error, but through a lever whose near end is owned and whose far end is not.

The compensation plan is the engine

The cascade is the mechanism. The reason it runs is the compensation plan, and this is the load-bearing point of the lesson.

Local compensation pays a function for the near end of its lever — the step it can see, the number it is measured on. The cascade damage lands at the far end, in another function, unattributed. So the plan is not a neutral measurement of performance. It is an active instruction to pull the lever, and it keeps issuing that instruction regardless of what the far end does, because the far end is not in the plan.

Make it concrete. Say the company's actual targets are high net revenue retention and durable growth — the board wants a base that compounds, not just logos that close. Now look at how sales is paid: on bookings, with no selectivity for which customers will retain and expand. The two are not aligned. They are not even pointed the same direction.

If sales is paid purely on bookings, then the rational move under that plan is to close every winnable deal, including the low-fit, low-retention ones — because the plan pays for the close and is silent on what the customer does afterward.

Then sales hits quota and is compensated in full. The plan worked exactly as written.

Then the base fills with customers who churn or fail to expand, and net revenue retention — the company's actual target — comes in below plan.

Then the company has paid, in full, for the partial destruction of the thing it said it wanted.

Nobody cheated. Sales did precisely what the plan rewarded. The failure is that the set of local compensation plans, taken together, does not add up to the company's targets — and frequently competes with them. Marketing is paid for cheap pipeline, which rewards broad low-fit segments. Sales is paid for bookings, which rewards closing them. Customer success is paid for gross retention, which rewards saving accounts that should never have been sold. Each plan is locally sensible. Assembled, they instruct the organization to do something other than what the company is trying to do — and they pay for it on delivery.

The same cascade, pulled hard: after the round

The operational version is the steady-state case. The dangerous version is when a single lever gets pulled hard against the cascade — and the clearest instance is the quarters after a funding round.

A round changes one input: capital is no longer the binding constraint, growth is. Every function's lever is now pulled in the same direction, hard, at once.

If the mandate becomes growth and acquisition spend is increased sharply, then pipeline rises — but the marginal pipeline comes from looser segments, because the tight ones were already being worked.

Then sales scales headcount fast to convert the larger pipeline, and win rate holds only by widening discount and softening terms across a much larger deal count — so the cascade from the operational case now runs at several times the volume.

Then customer success inherits a cohort that is both larger and lower-fit, and its cost-to-serve rises faster than its expansion, so the expansion motion that looks strong is largely paying for the cost the cohort created.

Then twelve to eighteen months on, net revenue retention and margin are materially below the plan the round was raised against — not because any function failed its number, but because every function's lever was pulled hard in the same direction and the cascade compounded across all of them at once.

This is the same lever and the same cascade as the operational case. What changed is amplitude and synchronization: every lever pulled hard, in the same direction, at the same time, with the compensation plans all rewarding the near end. A company that does not see the cascade experiences this as "we raised, we grew, and somehow the unit economics fell apart." A company that does see it treats the round as a decision to run the whole cascade harder, and reads the far end before pulling the near one — the same cascade, understood end-to-end instead of one step at a time. The lever did not change. The visibility into where it lands did, and that difference is the whole difference between growth that compounds and growth that has to be unwound.

What this costs before it shows

The single bad year is the visible case. The common case is slower. Most organizations do not get one year where everything diverges at once — they get a long sequence of quarters where each lever moves the cascade by a small amount: a point of margin, a few weeks of cash timing, a slight rise in the cost of the average new customer. None is large enough to investigate. Each is inside the noise of a normal quarter. Every individual move is defended successfully, because at the scale of one quarter it genuinely is too small to matter.

They compound, because the compensation plan that drives them does not change between quarters. A small consistent drift in the same direction accumulates into a structural position no one chose and no one can point to the decision behind. By the time the aggregate forces attention, it is two years of drift with no single owner and no single quarter to undo. The organization goes looking for the quarter it went wrong and there is no such quarter — there is only the structure, paying for the near end of every lever, quarter after quarter.

The first cost is in confidence, ahead of the numbers. A team that cannot find the decision behind a drift stops trusting that the next decision will behave any better, and a team that does not trust its own levers stops pulling them deliberately — the precise opposite of steering a system.

What this lesson does and does not claim

This lesson establishes the failure and its engine: in a system, optimizing each part against its own number is not a path to optimizing the whole, because the levers cascade and the compensation plans pay for the near end while the far end lands unowned. It is structural and expected, and pushing the people harder makes it worse, because effort applied through this structure is aimed, by design, at the near end of every lever.

The resolution is visible from here even though its mechanism is the next lesson's. If the problem is that local compensation rewards the near end of a cascade that the company does not see, the answer is not better people or more effort. It is to make the cascade visible and to set the local incentives against the company's actual targets rather than against isolated local numbers — so that pulling a lever is rewarded for where it lands, not only for where it starts. Stated as a posture that is the whole of it: growth in a system is something you configure so the parts are set against the whole, not something you accelerate part by part. How that configuration actually works — what is set, by whom, against what — is the next lesson.

Next up

If pushing harder on the parts backfires, the real question is how the whole is configured.

→ Continue to Growth is a configuration

———

This article is part of The Predictive Path

By Niko Laine, SaaS CFO

→ About the author

→ LinkedIn

You can read it on its own or explore the full curriculum.

→ View The Predictive Path

→ Explore courses