Attribution, Branded Search Overlap, and 22 Deals a Quarter
Tessera Industrial makes vibration-monitoring sensors for manufacturing plants. Deals average $60,000 and the company closes about 22 per quarter, split between direct sales and distributors. Marketing runs three channels: Google Ads at $18,000 a month, LinkedIn at $9,000 a month, and a monthly webinar that costs $4,000 in promotion. The team's last-touch dashboard credits Google Ads with 68% of sourced pipeline and LinkedIn with 9%. On that basis the VP of sales wants to move 80% of the budget into Google Ads next quarter. A separate review of won deals shows LinkedIn was the first touch in 41% of them, and webinars appear somewhere in the path of 55% of closed-won revenue. Google's own conversion tracking shows a 2.1% form conversion rate, but 44% of Google-attributed deals also had a branded search click before the form fill.
1. Explain what is distorting the last-touch report, and which model you would replace it with given the deal volume.
Read the worked answer
Two distortions are stacking.
First, last-touch gives 100% of the credit to the final click, which structurally favors demand capture over demand creation. LinkedIn and the webinar generate awareness and interest weeks before anyone searches; Google then intercepts the branded query and takes the credit. That is the 44% figure: nearly half of "Google-sourced" deals had a branded search - someone looking for Tessera by name - immediately before converting. Branded search is a demand-harvesting signal, not a demand-generating one.
Second, the 41% first-touch figure for LinkedIn is not proof LinkedIn is the hero either. It is the mirror image of the same error. Both numbers describe the same multi-touch reality from opposite ends.
What to use: a position-based (40/20/40) or W-shaped model - 40% to first touch, 40% to the touch that creates the opportunity, 20% spread across the middle - with the webinar treated as a milestone touch rather than a click. Also report sourced and influenced pipeline side by side so no single model gets to be "the truth."
The honest caveat: with 22 closed deals a quarter, any weighting scheme is a heuristic, not a measurement. A 40/20/40 model will produce confident-looking percentages from a sample where one deal moves the number by several points. Use it to direct budget, not to compute ROI to one decimal place.
2. Design an experiment that produces credible evidence within two quarters, given those volumes.
Read the worked answer
Accept the constraint first: you cannot A/B test at the deal level. Twenty-two deals a quarter means a two-arm test would need several quarters to detect anything short of an enormous effect. So test at a level with more units.
Option A - matched geo holdout (primary recommendation). Pick six comparable sales regions by historical pipeline, size, and industry mix. Randomly withhold LinkedIn spend in three for one quarter; keep Google and webinars constant everywhere. Compare pipeline created per region against the matched control. This measures budget effect, not just attribution, and it is the only design here that can show causation.
Option B - account-level holdout. If deal volume is larger inside a named-account list, split the target account list in half, suppress LinkedIn to one half, and measure meetings booked and opportunities created per 100 accounts. Still small, but the unit count is higher than closed deals.
Use leading indicators as the primary read. Marketing-qualified accounts, demo requests, webinar registration-to-meeting conversion, and branded search volume in holdout regions. These have 10–50x the event count of closed deals and move within a quarter.
Pre-register: the metric, the minimum detectable effect (plan for 20%+ relative, not 5%), and the decision rule. With this sample size, an honest negative result is worth more than a false positive.
3. The CFO wants one ROI number per channel. What do you present, and how do you handle the 80% reallocation proposal?
Read the worked answer
Present ranges with sample sizes, never a single point estimate.
A slide per channel with: spend, opportunities created, closed-won deals touched, cost per opportunity, cost per closed-won deal, and the number of deals behind each figure. For LinkedIn that might read "$27k/quarter → 9 closed-won deals touched → $3k per closed deal, n=9." The n is the honesty.
Show the branded-search overlap explicitly. A simple statement - "44% of deals credited to Google included a branded search immediately prior; in those cases Google captured demand created elsewhere" - reframes the 68% as a measurement artifact rather than a performance result.
Refuse the 80% shift, offer a 20% staged shift with a checkpoint. Reasons to give the CFO:
- The evidence for Google's 68% is the exact metric under dispute; reallocating 80% based on it is doubling down on a measurement error.
- Branded search spend has a ceiling: it scales with existing demand, so pouring budget in raises CPCs on queries you would win anyway.
- A staged shift is reversible and produces the experiment that settles the question. Propose moving 20% of Google budget into the LinkedIn holdout regions for one quarter, with a defined success threshold and a review date.
Say what is unmeasured. Webinar influence, distributor-sourced deals, and offline word of mouth are not in any of these numbers. Naming the blind spots builds more credibility with a CFO than a clean-looking dashboard.