Issue #9 · June 19, 2026

The AI Feature Shipped. The PAM Interview Now Tests Whether You Can Measure It.

Paul Brown Writes Golden Data from Ogden, Utah 4 min read

Career Snapshot

Q2 is the measure quarter. Most social platforms shipped AI features in Q1 — personalized feeds, optimized ad delivery, AI-assisted tools — without finalizing how they’d be evaluated. The Product Analytics Manager hire is the attempt to answer that retroactively.

Reddit grew advertising revenue 61% year over year in Q1 2026. Active PAM reqs span the ads product and consumer product orgs. Two separate briefs. The interviews test different things.

Roblox is expanding its brand advertising business and posting Staff-level product analytics roles across advertiser measurement and user experience. The brief requires product intuition and ad performance fluency in the same seat. Most candidates arrive with one.

TikTok (US) and Pinterest have open PAM reqs, but the pace has slowed while teams decide what to measure before adding a seat to own the measurement.

Comp: Product Analytics Manager at social platforms: $175K–$240K base. Staff-level experiment-infrastructure roles at the top of that range.

Hot this week:

Product Analytics Manager — ads product. Reddit, Roblox. Active.
Staff PAM — AI feature evaluation. Emerging category at Q1 AI-launch platforms.
PAM — self-serve advertiser experience. Reddit and Snap. Distinct brief from consumer product analytics.

Reading the tape: the platforms that shipped AI most aggressively in Q1 have the most open PAM reqs in Q2. Cause and effect.

Skill Market Shifts

1. The PAM interview floor moved from A/B testing to causal inference. Most candidates are still standing on the old floor.

The shift: when AI personalization touches every session implicitly, the clean holdout breaks. Holding back a fraction of users from a personalized feed is not measuring the feature — it is measuring a degraded product. PAM JDs at Reddit, Meta, and TikTok now list “quasi-experimental methods” and “CUPED” as baseline competencies, where prior versions stopped at “hypothesis-driven experimentation.”

The evidence: Reddit’s Staff PAM postings in Q2 include “experience with observational causal methods” as a listed competency — language absent from the same role family in 2024. Meta’s product analytics JDs show the same revision. The floor moved.

The career translation: the interview question is “walk me through a measurement problem where you couldn’t randomize.” That story exists somewhere in recent work for almost every senior analyst. It is a hard story to locate on a résumé written in 2023.

2. The mid-year AI feature review is producing a new artifact type. Most PAMs have never built one.

The shift: mid-year reviews at Q1-heavy AI platforms are generating a document that didn’t exist in prior cycles — a post-hoc AI feature impact assessment. Not a usage dashboard. Not a launch metric. A structured evaluation of whether a shipped feature produced the business value that justified it. Senior PAM screens at Pinterest and Reddit now ask candidates to walk through this evaluation for a feature that shipped without a measurement plan.

The evidence: Pinterest’s Head of Product Analytics posting (May 2026) includes “AI feature evaluation frameworks” as a stated ownership area — language absent from the 2024 version of the same role family. The mid-year review surfaced the gap. The JD posted the week after.

The career translation: prepare one answer to this question before the next screen: “We shipped an AI recommendation feature six months ago. Usage is up. Revenue is flat. How do you evaluate whether it worked?” That question has three possible answers. Only one of them does not end the interview early.

3. The advertiser-facing PAM role is a separate brief. The screen sorts candidates in the first twenty minutes.

The shift: Reddit, Snap, and Roblox are posting dedicated PAM roles for the self-serve advertiser funnel — campaign creation, activation, and performance — as a distinct category from consumer product analytics. An advertiser activation analysis is not a user retention analysis with the nouns changed. The interview tests an advertiser’s workflow logic, not a consumer’s.

The evidence: Reddit posted a Product Analytics Manager, Ads Self-Serve role in Q2 2026 — a category absent from their 2024 headcount plan. Consumer-product PAM profiles are arriving for this req in volume.

The career translation: if your current role touches campaign performance data, advertiser CRM data, or ad creative analytics in any capacity, that is the document for this interview. The brief says “product analytics.” The interview is an advertiser analytics interview. Most candidates apply to one and prepare for the other.

Next Action Plan

Portfolio. Target: PAM — AI feature evaluation at Reddit or Roblox. Write a one-page AI feature impact assessment: the feature, what “value” means in measurable terms, how you distinguish adoption from impact, what the counterfactual is. That is the artifact the Q2 PAM screen is requesting. Most candidates don’t have a written version.

Networking. One message to a product analytics lead at Reddit or Roblox. Three sentences: “The Q2 AI feature review seems to be producing a new measurement problem — systems that shipped without a pre-plan, no clean holdout. How is your team approaching the retroactive evaluation?” That signals you understand the structural problem. Most outreach doesn’t.

7-day micro-skill. Learn CUPED (Deng et al., 2013 — six pages, publicly available). One practitioner application post. One hour applying the variance-reduction logic to a metric you own. By Friday: you can answer “what do you do when your experiment has high variance and limited traffic” in a way that “increase sample size” does not. The PAM methods screen is the new PAM screen.

Sources

Crafting runs weekly. The job is to be useful, not to sell. If a line in here is close to a move you're working on, the email reaches a person.

Hiring rather than hunting? Signal is the column for your side of the desk.

Email me Read more Crafting

← All Crafting issues · Drafted with Claude · Edited by Paul Brown