Marketing Operations KPIs Are Not Campaign KPIs
If you search this question you will mostly be handed a list: cost per acquisition, return on ad spend, conversion rate, impressions, clicks, leads. Those are real numbers and none of them answer the question.
Want this built for you?
Animus Engine installs the marketing brain and the engine that runs on it. A human hand on every gate.
They tell you whether a campaign performed. They do not tell you whether the system that produces campaigns is working — whether work gets specified, shipped when promised, measured after it ships, corrected when it is wrong, and stopped when it is not worth doing.
The Three Layers, and Why Mixing Them Hides the Failure
There are three separate things being measured, and most dashboards stack them in one view:
- Business outcomes — revenue, qualified demand, conversion, retention. What the business got.
- Channel performance — impressions, clicks, cost per lead. How a given channel behaved.
- Operating system health — whether work is specified, shipped, measured, corrected and stopped, with traceable ownership and cost.
Mixing them is not untidy, it is diagnostically fatal. When results are flat, the three layers give three different culprits and you cannot tell which you have. A campaign that underperformed is a channel problem. A campaign that was never measured after launch is an operations problem. A campaign that shipped six weeks late into a seasonal window is an operations problem wearing a channel problem’s clothes.
You cannot fix the third with a better ad. That is the entire reason this category of measure exists.
Tell Us the Task That Eats Your Week
Name one marketing job you’d hand off tomorrow. We’ll send back how we’d run it with AI — the steps, the tools, and where a human stays in charge. A real reply from Chris within one business day.
Start With the Decision, Not the Dashboard
Before choosing any measure, apply one test.
A measure earns its place only if you can name the action it changes. Not “it tells us how we’re doing” — the specific thing you will do differently when it moves. If nobody can state that in a sentence, the number is decoration, and it will cost you the attention of everyone who reads it.
Two consequences follow, and both are more important than which measures you pick:
Set the cadence and the owner before the target. A number nobody looks at on a schedule is not a KPI, and one that everybody owns is owned by nobody. Monthly is right for most small businesses; weekly invites noise-chasing.
Expect the number to be gamed, and decide whether you mind. Every measure creates an incentive — this is Goodhart’s law, in the phrasing usually attributed to Marilyn Strathern: “when a measure becomes a target, it ceases to be a good measure.” The question is not whether it will be optimised, it is whether optimising it produces the behaviour you wanted. The table below names the gaming risk for each measure, because that is the part a list of KPIs never tells you.
A Small Marketing-Operations Scorecard
Six candidate measures. Choose three. A small business running six operational KPIs is doing measurement instead of marketing — and three that trigger real decisions beat six that produce a monthly report nobody acts on.
Six Measures and the Decisions They Trigger
| Measure | What it reveals | Minimum evidence | Decision it triggers | Gaming risk |
|---|---|---|---|---|
| Flow — did specified work ship when promised? | Whether commitments are real or aspirational | A dated record per planned item with a promised and an actual ship date | Persistently late → cut committed volume or add capacity. Never late → you are under-committing | Promise later dates; ship trivial items to keep the count |
| Quality — how much shipped work came back for correction? | Whether the brief and review stage are working | A count of items returned for rework, and the stage that caught it | High rework → fix the brief, not the writer. Zero rework → nobody is actually reviewing | Stop recording small corrections; loosen what counts as “returned” |
| Measurement coverage — did shipped work produce a usable record? | Whether you will be able to answer “did it work?” later | Per shipped item: was a record created, yes or no | Coverage falling → stop shipping until recording is fixed; unmeasured work cannot be judged | Create empty records that technically exist |
| Cost visibility — can spend be tied to the work it bought? | Whether the budget is legible or a lump | Provider spend recorded against the run or item that incurred it | Untraceable spend → fix the record before renewing anything | Allocate everything to a single bucket labelled “marketing” |
| Decision latency — how long did a known issue sit undecided? | Whether the bottleneck is doing or deciding | A timestamp for when an issue was raised and when it was resolved | Long latency → the constraint is a person’s calendar, not the team’s capacity | Close items as “decided” without deciding |
| Stop discipline — did low-value work leave the queue? | Whether the system can subtract, not only add | A record of items removed or retired, with a reason | Nothing ever stopped → the queue is growing by construction and will eventually fail on volume | Retire things that were never really going to happen |
Read the “decision it triggers” column as the real content. Each row has a bidirectional trigger, and the flattering direction is usually the informative one: zero rework almost never means the work is flawless, it means the review step is not happening.
On thresholds: this page deliberately gives none. A published benchmark from a company selling marketing software is not evidence about your business, and we have no dataset that would license a universal number. Set your first target from your own last quarter, then move it.
What a Record Can and Cannot Prove
We can be concrete here, because we run this on ourselves.
The system we operate writes one row per run: what ran, when, the outcome, a link to the work produced, the measured units and cost, and a summary of what needed human review. Provider spend is mirrored separately so it can be traced to the work that incurred it. The marketing operations hub sets out that record’s anatomy and the outcome vocabulary it uses — including the one most systems lack, a value for nothing needed doing, which is a genuine result and not a gap.
What matters for this page is narrower: the scorecard above is computed from that record, not from a separate dashboard. Flow comes from the dates. Quality comes from the rework counts. Measurement coverage comes from whether a row exists at all. Cost visibility comes from the spend mirror. The split is not ours: DORA’s software delivery metrics separate throughput — how quickly a change reaches production — from instability, the share of deployments that fail or have to be redone.
Two measures in that table are honest about their own limits, and you should treat them the same way in your own system:
Decision latency needs two timestamps, and most record designs only have one. A record with a single date tells you when something ran, not how long an issue waited for a verdict. If your log has one date field, decision latency is a concept you are tracking by hand, not a number you can compute — and you should say so on the dashboard rather than estimating it.
Stop discipline needs a record of removals. Systems record what shipped. Very few record what was killed and why, because nothing forces them to. If you cannot answer “what left the queue last quarter,” that is a schema gap, not a performance result, and adding one field fixes it.
Process Health Is Not Proof That Marketing Worked
This is the load-bearing caveat and it cuts against our own interest.
A system can score perfectly on all six measures and still produce marketing nobody responds to. Shipping on time, with full measurement coverage and traceable cost, means the machine is running. It does not mean the machine is pointed anywhere useful.
So these measures belong alongside business outcomes, never instead of them. Anyone — us included — who shows you a healthy process scorecard as evidence that marketing is working is answering a question you did not ask. The operating measures tell you whether you can trust your own reporting and whether problems get fixed. Whether the marketing is any good remains a separate judgement, informed by layer one.
That judgement stays human. Collection and calculation can be automated safely; deciding what a number means, and what to do about it, is on the list of work we argue should keep a named human owner — the reasoning is on our AI in marketing page, and it applies to your own dashboard as much as to ours.
Use the Scorecard to Choose the Next Fix
The point of separating the layers is that each failure has a different repair.
A flow problem — work is late or does not ship — is usually capacity or commitment, not effort. The fix is to commit to less, or resource more. Adding a tool here almost always makes it worse, because you have added a thing to maintain to a system already short on time.
A quality problem — work returns for correction — is nearly always upstream. Rework concentrated at one stage points at the brief that fed it. Blaming the person at the end of the chain is the most common wrong answer in marketing operations.
A measurement problem — shipped work with no record — is the one to fix first, because it silently disables every other diagnosis. You cannot tell a flow problem from a quality problem in unrecorded work.
A latency problem — issues known and unresolved — points at ownership, not capacity. Somebody needs the authority to decide, or a standing slot where decisions get made.
A stop-discipline problem — nothing ever leaves the queue — points at scope. It is the failure that looks healthiest for longest, because a growing queue reads as ambition right up until it reads as a backlog.
And there is a point at which more measurement is itself the problem. If you have run three cycles, the numbers point the same way each time, and the decision still has not been made, stop measuring and decide. Another month of data is not what is missing.
Where This Sits in a Small Marketing-Operations System
This scorecard is the measurement layer of a larger thing. If nobody currently owns this work, the hub is the place to start — it covers what the function is and what it costs to leave unowned. If the question is who does it and at what size, the team-structure page covers staffing, ownership and the do-it-yourself-versus-hand-it-over call.
Build the workflow after the decisions are defined, not before. Automating a process whose decision points nobody has named produces a faster version of an unclear system — and the measures above will faithfully report that it is running well.
What to Do This Month
- Write down the three layers and sort your current report into them. Most of it will be layer two.
- Pick three measures from the table. Not six.
- For each, name the action it triggers when it moves, in one sentence. If you cannot, pick a different one.
- Name the owner and the review date before you set a target.
- Check whether your record can actually produce each number. Where it cannot, add the field or mark the measure as manual — do not estimate it.
- Review once. Change one thing. Then review again.