Pinterest — Marketing Data Science — Product Sense & Metrics
Role context: Data Scientist, Marketing · Est. study time: 35 min · 4 practice questions
How to prepare for this role
This is B2B marketing analytics, not consumer product DS. The "customer" is the advertiser, and the subject is Pinterest's own marketing spend to acquire and grow advertisers. It's judged on marketing/business sense, choosing the right measurement method, causal reasoning about marketing lift, and communicating ROI — not on ranking or feed metrics.
The day-to-day is measuring and optimizing the marketing that grows Pinterest's advertiser base: analyze the advertiser funnel (acquisition, conversion, retention, churn), quantify marketing ROI across channels (paid media, email, in-product, webinars, events), build propensity / forecasting / LTV models, develop attribution, and measure incremental impact with causal inference and experiments — then recommend where the budget should go. The instinct that separates strong candidates: pushing past "marketing-influenced" revenue to what marketing actually caused.
Where to spend your prep time:
- Marketing product sense and metrics (this article) — the advertiser funnel, the ROI metrics (CAC, LTV, LTV:CAC, ROAS), and why incremental beats influenced.
- Measurement methods and causal inference — MMM vs multi-touch attribution vs incrementality, geo-holdout design, LTV under churn. See the A/B Test & Causal Inference section.
- SQL and communication — be fluent in SQL/Hive, and practice turning a measurement into a budget recommendation for leadership.
The through-line: frame the vague question into a metric and a measurement design, anchor on incremental ROI, and recommend a budget. That matters more than any single model.
What Pinterest Marketing DS actually is
Pinterest is an ad-supported visual-discovery platform (~631M monthly users, ~$4.2B ad revenue in 2025). The money comes from advertisers paying to reach users with high commerce intent. This role sits in the marketing org, whose job is to acquire, grow, and retain those advertisers, and to prove and improve the ROI of the marketing spend that does it.
Two facts define the work:
- The advertiser base is two-tiered. A smaller set of large brands drives much of the spend, while a fast-growing SMB long tail (many advertisers, low individual spend, higher churn) is a growth focus, alongside better measurement to win performance budgets. Marketing analytics has to serve both a high-value/low-volume segment and a high-volume/high-churn one.
- Marketing ROI is a causal problem, not a reporting one. Whether a webinar, an email, or a paid campaign actually caused an advertiser to sign up or spend more is not visible from correlations. It takes careful attribution and, ultimately, incrementality experiments. The whole role is about measuring marketing impact honestly, then moving budget toward what truly works.
The advertiser funnel
At each step a DS asks: which channel moved the advertiser here, at what cost, and would they have moved anyway. The end of the funnel matters as much as the top: an SMB advertiser who activates once and churns after a month has very different value from one who scales and stays.
The channels and the money flow
The levers — paid media, email/CRM, in-product, organic social, webinars, e-learning, in-person events — each have different cost, cycle length, and measurability. The DS's job is to estimate incremental revenue per channel and reallocate the budget toward the channels (and segments) with the best marginal return, not the best last-touch credit.
Metrics by layer
North Star (efficient growth):
- Marketing-incremental advertiser revenue, at a healthy LTV:CAC — revenue truly caused by marketing, acquired at an efficient cost.
Acquisition and activation:
- New advertisers and CAC (cost to acquire an advertiser) by channel; first-campaign / activation rate and time-to-first-spend.
Value and retention:
- Advertiser LTV (lifetime ad spend), ARPA, payback period; retention / churn rate and net revenue retention (upsell).
Channel ROI and attribution:
- ROAS / marketing ROI and incremental revenue per channel; marketing-influenced vs marketing-incremental revenue.
Guardrails:
- Blended CAC creep, over-attribution, and spend on advertisers who would have converted anyway.
The senior point in one line: "marketing-influenced" revenue is a vanity metric — the number that matters is incremental revenue at a healthy LTV:CAC and payback.
The data you'd look at daily
The unit here is an advertiser, not a Pinner, and the tables reflect that. Three do most of the work (example columns and rows):
1. Advertiser lifecycle table — one row per advertiser, updated daily:
| advertiser_id | tier | first_touch | signup_date | first_spend | spend_30d | spend_90d | status |
|---|---|---|---|---|---|---|---|
| a_10441 | smb | paid_social | 2026-04-02 | 2026-04-05 | $840 | $2,310 | active |
| a_10598 | smb | webinar | 2026-04-11 | — | $0 | $0 | never_activated |
| a_09877 | brand | events | 2025-11-20 | 2025-12-02 | $61,000 | $204,000 | active |
| a_10233 | smb | 2026-02-14 | 2026-02-16 | $410 | $410 | churned |
Two columns carry most of the meaning. first_touch is the honest limit of what you can attribute without an experiment — it records what the advertiser touched, not what caused them to sign up, and the difference between those two things is the whole role. And the gap between signup_date and first_spend is where advertisers are quietly lost: row two signed up and never ran a campaign, which costs the same to acquire as one who did.
2. Channel spend and outcome — one row per channel × week:
| channel | spend | new_advertisers | activated | rev_90d_influenced | rev_90d_incremental | cac | marginal_roi |
|---|---|---|---|---|---|---|---|
| paid_social | $410,000 | 1,240 | 690 | $3,900,000 | $1,120,000 | $331 | 1.9 |
| $22,000 | 310 | 240 | $1,400,000 | $310,000 | $71 | 4.1 | |
| webinars | $95,000 | 180 | 78 | $980,000 | $120,000 | $528 | 0.8 |
| events | $620,000 | 44 | 41 | $5,200,000 | $2,800,000 | $14,090 | 2.2 |
The two revenue columns are the point of this table. Influenced is three to eight times larger than incremental in every row, and only one of them can move a budget. Webinars look respectable on influenced revenue and have a marginal ROI below 1, meaning the next dollar spent there loses money. Events look terrifying on CAC and are one of the best rows here, because 41 of 44 activate and they are brand advertisers.
3. Cohort retention and LTV — one row per signup cohort × tier:
| cohort | tier | advertisers | m1_retained | m3_retained | m6_retained | ltv_to_date | payback_months |
|---|---|---|---|---|---|---|---|
| 2025-10 | smb | 2,940 | 61% | 34% | 22% | $1,180 | 7.4 |
| 2025-10 | brand | 71 | 96% | 93% | 90% | $412,000 | 1.9 |
| 2026-01 | smb | 3,510 | 64% | 38% | — | $690 | — |
| 2026-01 | brand | 66 | 97% | — | — | $214,000 | — |
Never report one blended LTV across these rows. An SMB advertiser is worth about $1,200 and most are gone within six months; a brand advertiser is worth several hundred thousand and almost none leave. A single average would be a number that describes no advertiser who actually exists, and every budget decision made from it would be wrong for both groups.
What the dashboards look like at Pinterest Marketing
Two views matter most.
The health dashboard — the weekly view for the marketing org:
| Metric | This week | WoW | vs target | Status |
|---|---|---|---|---|
| Marketing-incremental revenue (North Star) | $4.35M | +2.1% | on track | OK |
| LTV : CAC (blended) | 3.4 | −0.2 | below | Watch |
| New advertisers | 1,774 | +4.0% | on track | OK |
| Activation rate (first campaign) | 55.6% | −1.8pp | below | Alert |
| SMB 90-day retention | 38% | +1pp | on track | OK |
| Blended CAC | $631 | +6.2% | below | Watch |
| Payback period (SMB) | 7.4 mo | +0.3 | below | Watch |
Your job is to explain every Watch or Alert cell. Note that new advertisers are up while activation is down and CAC is up — acquiring more of the wrong advertisers looks like growth on the top line and shows up here as three separate warnings.
A trend tile sits alongside it, and it is the one that settles budget arguments:
Beyond those two views, a Marketing DS reads the business the way it actually runs:
- Incremental, not influenced. ROI is reported as incremental revenue (calibrated by holdout/geo experiments), with influenced revenue flagged as an upper bound.
- By channel and by segment. CAC, LTV, and ROAS are split by channel and by advertiser tier (brand vs SMB), because the economics differ sharply.
- Marginal, not average. Budget decisions use marginal ROI and diminishing-returns curves from MMM, not average ROAS.
- Cohort LTV. LTV is read on cohorts with survival/retention curves, because SMB churn makes a flat multiple misleading.
Every read ends in a decision: fund a channel more, cut it, or run an incrementality test to find out.
Quick check
Who is the customer whose behaviour this role is measuring?
A channel report shows a webinar series 'influenced' $12M in advertiser spend. What is the problem with reporting that number?
Practice Questions
What metrics tell you Pinterest's advertiser marketing is working?
Metrics frameworkMediumYou own analytics for the marketing org that grows Pinterest's advertiser base. What metric set tells you the marketing is working, and why each?
Before you reveal: say your answer out loud, as if you were in the real interview — get your reasoning across clearly first. There is no single correct answer: reading what the interviewer is really after and defending your own thinking is what makes an answer strong.
Anchor on efficient, caused growth, and read the whole funnel plus the economics:
- Primary (efficient growth): marketing-incremental advertiser revenue at a healthy LTV:CAC. The point of marketing is revenue it actually caused, acquired at a cost that pays back — not gross new advertisers or influenced revenue.
- Acquisition efficiency: new advertisers and CAC by channel, so you can see which channels acquire cheaply.
- Value: advertiser LTV (lifetime ad spend) and payback period, split by tier — a cheap SMB acquisition that churns in a month can be worse than a pricier brand acquisition that stays.
- Retention: advertiser churn and net revenue retention, since keeping and growing spend is usually cheaper than acquiring.
- Guardrail: are we spending marketing dollars on advertisers who would have converted anyway (low incrementality), and is blended CAC creeping up?
One-line framing: incremental revenue at a healthy LTV:CAC as the North Star, CAC and activation for acquisition, cohort LTV and payback for value, churn for retention, and incrementality as the guardrail.
Why measure incremental revenue instead of marketing-influenced revenue?
Metric designMediumMarketing reports that campaigns "influenced" $50M of advertiser revenue last quarter. Why is that the wrong number to optimize, and what should you use instead?
Because "influenced" credits marketing for revenue it didn't necessarily cause. If an advertiser saw an email, attended a webinar, and then signed up, every one of those touches claims influence — but many of those advertisers would have signed up anyway (they were already searching for Pinterest ads, or a brand was always going to buy). Influenced revenue double-counts across channels and includes people at the bottom of the funnel who needed no persuasion, so optimizing it rewards touching soon-to-convert advertisers rather than actually creating conversions.
The right number is incremental revenue — the revenue that exists because of the marketing and wouldn't have happened otherwise. You establish it with a counterfactual: a holdout or geo experiment where comparable advertisers/markets get no (or less) marketing, and the lift over that control is the true causal effect. Then judge efficiency by LTV:CAC and payback, using incremental revenue in the numerator. This mirrors the general principle that a proxy optimizes for exactly what it measures: "influenced" measures proximity to a conversion, not causation, so you anchor on the caused, incremental outcome. The senior signal is refusing to run marketing on influenced-revenue vanity metrics.
Design an advertiser LTV metric when SMBs churn fast
Metric designHardPinterest wants an advertiser LTV metric to weight acquisition and budget by value. Many advertisers, especially SMBs, spend once or twice and churn. Design an LTV metric that doesn't fool you.
The hard part is that LTV is a prediction about the future under heavy, uneven churn — a flat "average spend × average lifetime" massively overstates the fast-churning long tail and is dominated by a few big brands. Build it up:
- Cohort + survival, not a flat multiple. Estimate LTV from retention/survival curves per cohort: the probability an advertiser is still spending in month t, times expected spend if active, summed (discounted) over the horizon. This naturally down-weights advertisers likely to churn.
- Segment first. Brand and SMB advertisers have completely different retention and spend, so fit LTV by tier (and by acquisition channel), then aggregate. One blended LTV hides that the SMB tail is low-value and high-churn.
- Predicted, then validated. Use a model (survival / propensity-to-retain, or a probabilistic LTV model) on early signals (activation depth, first-90-day spend, product adoption), and validate it by checking predicted LTV against realized spend on matured cohorts.
- A finite, honest horizon. Cap the horizon (e.g., 12-24 months) and discount, rather than extrapolating to infinity, so a couple of retained whales don't create a fantasy number.
So the metric is cohort-and-tier survival-based predicted LTV over a finite discounted horizon, validated on matured cohorts. The trap is average-spend × average-lifetime, which over-counts churny SMBs and is skewed by a few large brands. Then use LTV to weight CAC (LTV:CAC by segment) so acquisition is judged on value, not volume.
Acquisition vs retention — where should the marketing budget go?
Trade-offHardLeadership wants advertiser growth. The team can spend the next marketing dollar acquiring new advertisers or retaining/growing existing ones. Which do you optimize, and how do you decide?
Name the tension: new-advertiser growth versus retaining and expanding existing advertisers. Acquisition grows the logo count and looks like "growth," but new advertisers (especially SMBs) are expensive to acquire and churn fast, while retention/upsell of existing advertisers is usually cheaper per dollar of revenue and compounds. The wrong move is to chase whichever the org is culturally biased toward.
How I'd decide:
- Compare marginal ROI, not average. The real question is where the next dollar returns more: the marginal incremental revenue per dollar of acquisition vs retention spend, accounting for diminishing returns in each (from MMM/experiments). If acquisition channels are saturated (rising marginal CAC) while retention has headroom, shift to retention, and vice versa.
- Weight by LTV and payback. Acquisition is only worth it if acquired advertisers' LTV:CAC and payback are healthy; if the SMB tail churns before payback, that acquisition dollar is worse than a retention dollar that lifts net revenue retention.
- Segment. The answer differs by tier: retaining/expanding brands may have the best marginal return; acquisition may be worth it in an underpenetrated segment or new market where there's untapped demand.
- Test the uncertain bets. Where marginal ROI is unclear, run an incrementality experiment (hold out a segment from a retention program, or a geo from an acquisition push) rather than argue from attribution.
The recommendation is a marginal, LTV-weighted, segment-specific allocation, validated by experiments where uncertain — not a blanket "acquire more." This is the budget-allocation instinct: fund the highest marginal incremental ROI, wherever in the funnel it is.