Measurement & Attribution — straight answers.

Multi-signal triangulation, because no single attribution source survives 2026's privacy environment intact. Our stack:

1. MMM-lite weekly modeling — geo holdouts on campaigns above $40K reveal incremental lift vs paid social baseline
2. Post-purchase surveys — "where did you hear about us" embedded in checkout flow; recovers 15-30% of unattributed conversions
3. Branded search lift — query volume on brand+product terms during and 14 days post-campaign
4. Affiliate/coupon redemption — last-touch but useful as conversion floor
5. MMP postbacks (for apps) — AppsFlyer/Adjust with SAN integrations where supported
6. Direct/organic lift attribution — sessions during campaign window minus 30-day pre-baseline

Single-source ROI numbers are misleading by design. A campaign showing 0.4x ROAS on last-click typically shows 2.1-3.8x when MMM and survey data are layered in. The brands losing money on creator campaigns are usually the ones measuring with last-click only and killing campaigns that were actually working.

Last-click captures 25-45% of true creator-driven conversions, depending on vertical. For high-consideration purchases (fintech, SaaS, auto, premium beauty), it can be as low as 15%. Using it as your sole measurement framework systematically undervalues influencer marketing by 2-4x.

Where last-click still has utility:

• Floor measurement: if last-click ROAS is positive, true ROAS is almost certainly profitable
• Creator-to-creator comparison within a single campaign (relative ranking holds even when absolute numbers undercount)
• Coupon/affiliate-driven beauty and fashion (purchase intent is immediate, attribution gap is smaller)

Where it actively misleads:

• Long consideration cycles (auto, fintech, B2B)
• Multi-touch journeys where YouTube discovery seeds branded search 7-21 days later
• Cross-device journeys (saw on TV via YouTube app, bought on desktop)

Pair last-click with branded search lift and post-purchase surveys at minimum. Anything less is flying blind.

A model output giving you the expected cost-per-acquisition for a specific creator + offer + geo combination, generated before you sign the deal. Inputs include:

• Creator's historical view velocity and audience composition
• Past performance of similar creators in your vertical
• Geo-adjusted conversion baselines
• Seasonality factor for the campaign window
• Your offer's expected CTR and landing-page CVR

Output is a CPA range (e.g., "$42-68 install CPA, midpoint $54") with a confidence interval. For verticals we have deep data on, our ranges hit within ±25% of actuals about 78% of the time. For thin-data verticals or first-of-kind creator categories, ranges are wider and we say so.

Practical use: brands run it before approving creator picks, kill deals where predicted CPA exceeds payback threshold, and renegotiate pricing when our forecast diverges from the creator's quoted rate. Replaces "the agency thinks this creator is good" with a number you can plan against.

Aggregate accuracy across our 1,400+ delivered placements:

• Predicted CPA within ±25% of actual: 78% of campaigns
• Predicted CPA within ±50% of actual: 92%
• Predicted view delivery within ±20%: 84%

Accuracy varies by data density. For verticals where we have 100+ comparable placements (mobile games, beauty, fintech apps), we hit the ±25% band on 82-87% of forecasts. For thinner verticals (B2B SaaS, automotive specifically, regulated health), we hit ±25% on 60-70% and we communicate wider confidence intervals upfront.

What we get wrong: brand-new creators with no prior sponsorship data, sudden algorithm shifts (mid-2025 YouTube Shorts monetization change broke our model for ~6 weeks), and offers with creative quality far above or below the brief expectation. We publish accuracy retrospectives quarterly to clients on retainer — the model is wrong sometimes and you should know when.

Cross-site cookie deprecation finished rolling out in late 2025, and the Privacy Sandbox APIs that were supposed to replace them ended up partially shelved. Practical impact on creator attribution:

• Web-to-web last-click: lost another 20-30% of accuracy on top of the iOS 14.5 hit
• View-through attribution: essentially dead outside walled gardens
• Cross-device: functionally impossible without first-party identity graphs

What still works:

1. First-party data capture — email/phone at conversion, then back-match to campaign window
2. Server-side conversion APIs with affiliate/promo codes as the identifier
3. Geo-based MMM — running creator campaigns in test markets vs holdout markets
4. Modeled conversions from Meta/Google when creator content is whitelisted
5. Post-purchase surveys — went from "nice to have" to essential

The brands handling this well moved attribution budget into MMM tooling and survey infrastructure in 2024-2025. The ones still trying to do pixel-based last-click are flying blind and don't fully realize it yet.

Both, with different roles.

UTMs: Free, universal, but only capture click-through. Use them on every creator link as a baseline. Standardize parameters across campaigns — utm_source=creator_handle, utm_medium=youtube_integration, utm_campaign=q2_launch. Without strict naming conventions, your analytics become a mess inside 90 days.

Affiliate/promo codes: Spoken codes ("use code SARAH20") capture conversions where the user never clicked, which is 30-60% of creator-driven purchases. Essential for verbal-pitch content like YouTube and podcasts. Pair with a 15-25% discount or value-add to drive code redemption.

Short links (Linktree, Bitly, branded shorteners): Don't rely on them for attribution — use them for cleaner audience experience. The underlying UTM does the measurement work.

For apps, replace UTMs with MMP deep links (AppsFlyer OneLink, Adjust). For web, layer UTMs + promo codes + post-purchase survey. The combination triangulates to within 80-90% of true creator attribution. Either tool alone tops out around 40-55%.

Three layers, scaled to budget:

Sub-$50K campaigns: Branded search lift in Google Trends + Search Console, measured 30 days pre vs 30 days post. Cheap, directional, works.

$50-200K campaigns: Add a panel-based brand lift study (Lucid, Cint, Attest) running pre/post on awareness, consideration, and purchase intent. Cost: $4-12K. Sample size 800-1,500 per market gets statistically valid reads on 3-5 point shifts.

$200K+ campaigns: Full MMM contribution analysis, plus pre/post brand tracker waves with control geos. Cost: $15-40K. Reveals which creator tiers/formats drove which brand metrics.

What not to use: EMV ("earned media value") is a self-reported vanity number with no consistent methodology. Engagement-rate "brand lift" claims from creator agencies are post-hoc storytelling. If a vendor offers brand-lift measurement without a control group or a survey methodology, you're paying for a slide, not data.

EMV (Earned Media Value) is a calculated estimate of "what equivalent paid media would have cost." It's almost always inflated and rarely tied to revenue. Common formula: impressions x assumed CPM x engagement multiplier. The assumed CPM and multiplier are usually picked to make the number look big.

Real differences:

• EMV measures hypothetical advertising cost. ROI measures realized revenue per dollar spent
• EMV treats every impression as equally valuable. ROI weights conversions
• EMV has no standardized methodology — two agencies will give you 3x different EMV for the same campaign. ROI ties to your P&L

When EMV is useful: comparing earned-media volume across campaigns using a consistent internal formula. When it's not: justifying campaign spend to a CFO. We've seen brands report 7x EMV on campaigns that delivered 0.6x ROAS, then renew the program based on EMV alone. Don't be that brand. Ask for revenue numbers, not media-value numbers.

Depends on purchase cycle and content format, but practical waiting periods:

• Mobile games (install + D7 ROAS): 14 days minimum, 30 days for confident read
• DTC/beauty/fashion (single purchase): 21-30 days
• Fintech (funded account, KYC): 45-60 days
• B2B SaaS / high-consideration: 60-90 days
• Auto / large purchases: 90-180 days

Why you can't judge faster: YouTube long-form continues earning views for 30-90 days post-upload, accumulating 40-70% of total conversions after week one. Killing a YouTube campaign at day 7 based on initial CPA is one of the most common — and expensive — mistakes we see brands make.

TikTok and Reels have faster decay curves (80%+ of conversions in first 14 days), so you can judge those quicker. Podcast sponsorships are slowest — half the conversion volume can land 60-120 days after release because of catalog listening.

Set the read window in your campaign brief. Don't change it mid-flight because results look bad on day 5.

Honestly — barely. At $10K, sample size on most measurement methods is too thin for confident incrementality reads. What you can do:

Works at $10K:

• Geo holdout if you can isolate one city/state vs a matched control
• Promo code redemption (gives you a conversion floor)
• Post-purchase survey if you already have 200+ daily orders to sample from
• Branded search lift, directional only

Doesn't work at $10K:

• Panel brand-lift studies (need $50K+ campaign to justify $5K+ study cost)
• Full MMM contribution (needs 6+ months of data)
• Multi-cell creative testing with statistical significance

Honest framing for $10K: treat it as a learning budget, not a measurement budget. You're testing whether a creator/format pairing produces sane unit economics, not generating boardroom-grade incrementality data. If the test signals positive, scale to $40-60K and measure incrementality there. Brands that demand bulletproof incrementality reads from $10K tests are asking for math that doesn't exist.