Why contribution is the hardest part of a case to state
The portfolio guide asks every case to be precise about ownership, to show results with a comparison and to be explicit about what you cannot attribute to your work alone. The career ladders guide asks for outcomes described honestly in a promotion case. Both are right, and both leave the hardest part to the writer: deciding how much of a result is yours and saying so in a sentence.
Three things make that difficult. Marketing results are produced by many people at once, so almost no outcome belongs to one person. Results move for reasons nobody in marketing controls, such as prices, seasons, stock, competitors and the news, so a change in the number is not evidence that the work caused it. And the figures most marketers see first are reported by the platforms they work in, which describe what the platform recorded, not what the business gained.
There is also a reason to state contribution in business terms rather than channel terms. The finance reporting hub notes that the Chartered Institute of Marketing frames reporting as accountability: marketing showing its contribution in terms the whole business recognises. The same applies to an individual. A contribution stated in orders, customers kept or revenue from the business's own records means something to anyone who reads it; a contribution stated only in a platform's own measures means something mainly to people who work in that platform.
The skills hub treats evidence as one of the four stages of building a skill. This guide is about the sentence at the centre of that evidence, wherever it ends up.
Your part and the team's part
Start by writing down everyone who touched the result and what each of them did. A campaign might involve a strategist, a copywriter, a designer, a channel specialist, an analyst, a developer, an agency and a manager who argued for the budget. Your part is one line on that list, and the list is what lets you describe it precisely.
The KPI tree helps place that line. A tree splits an outcome into the drivers that produce it and the inputs a team can act on. Find the input you moved: an email's click rate, a landing page's conversion rate, the share of at-risk customers who saw an offer, the speed of a report. Then follow the branch upward to the outcome. Your claim is strongest at the level you moved and weakens as it climbs, because every level above you is also moved by other branches and other people. "I raised the share of trial users who completed onboarding" is a claim about your input; "I grew revenue" is a claim about the whole tree.
A useful test is the verb. Words such as designed, wrote, built, ran, analysed, tested and recommended describe your actions and can be checked. Words such as led, owned and drove describe your role in a group and need support: who else was involved, what you decided and what they decided. If you would need to explain away the verb when asked about it, choose a smaller one. Where an AI tool produced part of the work, the list of who touched the result gains a line for the tool, and the tutorial on describing AI-assisted work shows how to split its part from yours and a reviewer's and how to word the claim.
The result and what would have happened anyway
Every result has a counterfactual: what the number would have been without the work. You rarely observe it, but the claim depends on it. Three kinds of comparison stand in for it, and each supports a different strength of claim.
The strongest is a controlled comparison. A Holdout GroupA randomly chosen part of an audience kept from exposure to measure what exposure caused. is a randomly chosen part of an audience kept from exposure so that the difference between exposed and held-out groups shows what exposure caused. The incrementality testing guide explains how such tests are designed and read, including the point that AttributionAssigning conversion credit to touchpoints. describes which touchpoints were present when people converted but cannot say which conversions would have happened anyway, while an IncrementalityConversions caused by marketing. test can. The comparison of geo holdout tests and conversion lift studies sets out two designs and who builds the control in each. If your work was tested this way, you can say what it caused, within the range the test produced.
The next is a comparison against a BaselineAn estimate of what a metric would have done without a given change, campaign or promotion, against which the effect is measured. or another period. The guide to comparing periods covers choosing the comparison that answers the question, aligning calendars and separating growth from seasonality. A before-and-after comparison shows that something changed while your work ran; it does not on its own show that your work changed it.
The weakest is no comparison at all: a single figure, such as an open rate or a cost per lead, reported without anything to read it against. That supports a description of the work, not a claim about its effect.
Two further checks apply to any comparison. The first is whether the number is large enough to read. The guide to reading small numbers explains why small counts jump around and how to put a range around a figure; a formal test of whether a difference exceeds what chance would produce is a Statistical SignificanceA test result unlikely enough under 'no real difference' to pass a threshold chosen in advance. question. The second is what else happened at the same time. The guide to explaining a change in a marketing metric sets out a method for matching a change to an event and testing the explanation against data it did not come from. Run your own result through it before you claim it: list every other change in the period, from prices to a competitor's campaign, and decide which of them could have produced the same movement.
Platform figures and the business's own numbers
Advertising, email and social platforms report conversions, revenue and returns according to their own counting rules. Those figures are useful for running the channel, but they are not the business's results. The incrementality testing guide notes that every attribution model credits touchpoints that were present on paths people would have taken anyway, which is why a platform's figure and the incremental figure can differ widely.
The business's own records are the orders system, the finance ledger, the customer database and the sales pipeline. A contribution stated in those terms is the one the CIM's accountability framing points to, and it is the one a reader outside marketing can check against what they already know. The comparison of ROI and ROAS for finance reporting explains why the two returns answer different questions: return on ad spend relates revenue to media and suits running a channel, while return on investment relates profit to all costs and suits finance.
In practice, keep both and label them. Report the business measure as the result, name its source and its comparison, and use the platform figure only to explain what happened inside the channel. Never present a platform-reported return as if it were the business's return, and never mix the two in one sentence without saying which is which.
| Kind of figure | Where it comes from | What it can support | What it cannot support |
|---|---|---|---|
| Platform-reported | The ad, email or social platform's own counting | How the channel behaved; changes inside the channel | What the business gained, or what would have happened anyway |
| Analytics-attributed | The analytics tool's attribution model | Which touchpoints were present on converting paths | Which conversions the work caused |
| Business records | Orders, finance, customer and pipeline systems | What the business recorded in the period | Why it changed, without a comparison |
| Controlled comparison | A holdout, geo or lift test | What the tested work caused, within a range | The effect at another spend, period or audience |
Wording that matches the evidence
Once the three separations are made, choose the wording the evidence supports. Think of it as a ladder: each rung needs more evidence than the one below, and the honest claim is the highest rung you can support, not the highest one you can write.
| Evidence you hold | Wording it supports | How it reads |
|---|---|---|
| Your work was tested against a holdout or control | Caused, produced, added, within the test's range | "The offer I designed cut cancellations among exposed subscribers by an estimated amount, against a holdout" |
| You did a defined part of work whose result was measured against a baseline | Contributed to, as part of a team | "As part of the lifecycle team, I built the win-back sequence; returning customers rose against the same period last year" |
| The measure moved while your work ran, with no comparison that separates other causes | Coincided with, during, while | "Enquiries rose during the quarter in which I rebuilt the landing pages" |
| The work was done but not measured | Describe the work and what you would measure | "I rebuilt the landing pages; no comparison was set up, and I would now measure enquiries against a baseline" |
The bottom rungs are not failures. Most work is never tested, and a sentence that says exactly what is known is stronger evidence of judgement than one that claims more than the evidence holds. When asked about a claim, as the interview preparation tutorial prepares you to be, you should be able to name the comparison, the source of each figure, the period and what else changed, without looking anything up. If you cannot, step down a rung.
Ranges belong in the claim when you have them. A test result is a range of plausible effects rather than a single figure, and quoting the midpoint alone overstates how much is known. "An estimated effect between one figure and another" is longer than a single number and far easier to defend.
Keeping a record you can defend later
Claims are written months or years after the work, when the details have faded and the colleagues who knew them have moved on. A short private record kept at the time is what makes the later claim accurate. For each piece of work worth a case, note:
- What you did and what others did, with names of roles rather than people if you will share it.
- The measure, its definition and its source, including which system it came from.
- The baseline and the comparison, with the dates of both periods.
- Any test, with its design, the holdout and the result as a range. If the team wrote it up with the experiment results report template, keep a note of where the report lives and what it concluded.
- What else changed in the period, from prices and promotions to tracking changes.
- What is confidential, and any permission you were given to describe the work.
Keep the real figures separate from any version you share. The portfolio guide's section on confidentiality and permission covers anonymising by sector and size, replacing absolute figures with relative ones and asking early for permission; what you may disclose is governed by your contract and your employer's policy, not by this guide. The certifications and self-study guide suggests ending each study session with a written note; the same habit applied to live work produces this record as a side effect.
A lifecycle marketer states her part in a retention result (illustrative)
Nyxabel writes up a fall in cancellations
Nyxabel is a lifecycle marketer at an invented meal subscription business. Over one quarter, monthly cancellations fall from 6.1% of active subscribers to 4.8%. Her manager's summary says the team "reduced churn by 1.3 points". Nyxabel designed a save offer shown to subscribers who started to cancel, and the team kept a random 10% of those subscribers as a holdout who saw no offer. In the same month the offer launched, the business announced a price freeze for existing subscribers.
She separates the parts. The holdout shows cancellations among exposed subscribers 0.7 points lower than among the held-out group, with a plausible range of 0.3 to 1.1 points; the rest of the fall happened in both groups, so it coincided with the price freeze and the season rather than the offer. The platform's dashboard credits the offer with 2,400 "saved" subscribers, but the subscription system shows that 1,150 of those cancelled within two months, so she reports the figure from the subscription system instead. On the KPI tree, the input she moved is the cancellation rate among people who start to cancel; the copy came from a colleague and the build from the developer.
Her private record keeps all of this. Her portfolio version reads: "Designed and tested a save offer for subscribers starting to cancel. Against a random holdout, it lowered cancellations among them by an estimated 0.7 points (range 0.3 to 1.1) in a quarter when a company-wide price freeze also reduced cancellations. Copy by a colleague; build by the development team." It does not say she reduced churn.
The example shows the shape rather than the figures: the team's headline is split into the part a controlled comparison attributes to her work and the part that coincided with other events, the business's records replace the platform's count, and the wording stays on the rung the evidence supports.
Common mistakes
- Claiming the outcome at the top of the tree. "Grew revenue" when you moved one input invites the question of who moved the rest. Claim the input you moved and show the branch it feeds.
- Reading a before-and-after as cause. A change during your work is not a change caused by your work. Name the comparison and what else happened in the period.
- Reporting platform figures as business results. Platform conversions and returns describe the platform's counting. Use the business's own records for the result.
- Quoting a midpoint without its range. A test result is a range. Give it, or say the result was uncertain.
- Borrowing the team's verbs. Led, owned and drove need support. If someone else made the decision, say what you did instead.
- Leaving out what else changed. Prices, promotions, seasons and tracking changes all move numbers. A claim that ignores them will not survive a question about them.
- Rebuilding the record from memory. Figures recalled months later drift upward. Note the measure, source, comparison and your part while the work is fresh.
- Breaching confidentiality to make a claim stronger. Relative figures and an anonymised description carry most of the meaning. Check what you may share before you share it.
