RESEARCH REPORTS
Research Reports
Deep dives into ads, AI and operations for founder-led ecommerce brands — researched, sourced, and reviewed. A new report each week.
October 4, 2026
Paid social went in-house in 2026, but the savings did not
Brands are pulling paid social buying in-house faster than any other function, but the 2026 data says they are no longer doing it to save money. We put the ANA, Gartner and published agency fee numbers against a $50K/month budget and show where the real cost sits.
Read reportOctober 4, 2026
Why Recharge Stores Double-Count Conversions (And How to Fix It)
Recharge stores report duplicate conversions for two reasons: the migration off checkout.liquid left two tracking layers firing at once, and first, recurring and prepaid charges get counted differently in each system. Here is the inventory-first fix, the deduplication key that actually works, and what we could not verify.
Read reportOctober 4, 2026
Agents in Production: What the 5% Actually Do Differently
MIT's much-quoted 95% figure is the inverse of a 60/20/5 funnel, not a verdict on agents. We read the funnel, McKinsey's flat EBIT line, Gartner's cancellation forecast and Sierra's pass^k reliability metric, then turn them into three things an ecommerce ops team can finish in 30 days.
Read reportOctober 4, 2026
What to Ask Before You Sign an AI Engineering Engagement
Article 50 disclosure went live on 2 August 2026 while the high-risk deadlines slipped to 2027, and the FTC is still charging AI capability claims two years into Operation AI Comply. Here is what to ask a vendor before signing, and the four contract clauses that actually cost money.
Read reportOctober 4, 2026
AI Ad Creative, Honestly Ranked: What Still Needs a Human
Meta says its GEM model lifted Instagram ad conversions 5% in Q2, and 4 million advertisers now use its generative creative tools. Icon raised $9.2M, paid $12M for a domain, then quietly wound down after selling human-made UGC at $399 a pack.
Read reportOctober 4, 2026
Grounding checks for AI ad copy: what actually holds up
A September 2026 NAD decision turned on whether an AI system used as a measuring instrument had been validated for the claim it supported. At the same time, two preprints put uncomfortable numbers on how well LLM judges agree with human raters.
Read reportSeptember 28, 2026
AI Phone Agents: The $0.05 Minute That Costs $0.12
Vapi's own calculator turns a $0.05 headline rate into $0.082-$0.129 per minute once the model and the phone line are added. Here is how to price a Shopify-connected voice agent on your own call volume, and the four things that decide buy versus build.
Read reportSeptember 28, 2026
Meta's Default-On Generative Creative Broke Your A/B Tests
Meta swapped the image model behind Advantage+ creative mid-quarter, and the opt-out surface is now per-feature and per-creative. Here is what is verified, what is not, and three things to do this month if you spend $50K+/month on paid social.
Read reportSeptember 28, 2026
Creative Testing at $50K+/Month: What to Measure Instead of CTR
Meta says its own system expects the number of ad creatives to grow significantly, which changes what a buyer should judge and how fast. Here are the 2026 hook and hold-rate bands, a kill rule with an impression floor, and the sizing a holdout test actually needs.
Read reportSeptember 28, 2026
Server-Side Tracking for Shopify Plus: What It Fixes, What It Costs
Shopify's web pixel sandbox, Safari's restrictions and Meta's quality scoring broke tracking setups that worked two years ago. Here is what server-side tagging fixes, what it costs at real request volume, and the three checks to run this month.
Read reportSeptember 28, 2026
What Breaks the Week Your Agency Leaves
Meta released Graph API v26.0 on July 29, 2026, and Google blocked offline conversion uploads in the Google Ads API on June 15, 2026. Both changes hit hardest in the week after an agency leaves, when nobody owns the dataset, the tokens, or the deduplication logic.
Read report