Pinterest Scales Open-Source AI for AI-Powered Discovery and UCAN KPI Clarity in Q2 2026

Pinterest’s open-source AI use and the best KPIs for UCAN engagement

1) How Pinterest leverages open-source AI models within its broader AI strategy (today)

Model-agnostic deployment: “right model for the task”

Pinterest says it uses a model-agnostic approach—deploying in-house compact fit-for-purpose models, open-source models that are post-trained on its own data, and in limited cases selected closed third-party models. ¹

Open source is integrated into user-facing and monetization-facing AI

In user experience, Pinterest describes its Pinterest Assistant conversational layer as being enabled by an approach that combines proprietary signals with cost-efficient open-source models, specifically noting that it can run securely in Pinterest’s own environment. ²
More broadly, management describes AI as “at the core of everything we do”—including the user experience, the ads platform, and internal operations. ³

Why open source matters economically: cost, customization, control

Management attributes three advantages to open-source adoption: cost, customization, and control. ¹

Cost advantage (quantified):

• For Pinterest’s use cases with open models, it reports cost per transaction at less than 8% of the cost of comparable closed proprietary models. ¹
• Pinterest also characterizes this as “less than 1/10 of the cost” versus comparable closed proprietary models. ⁴

Customization advantage (post-training on Pinterest data):

• Pinterest emphasizes that open models become “more effective” because it can post-train on its own data (i.e., performance is in “your own environment,” not just on benchmarks). ¹
• It also states that when it leverages open source models (including for Pinterest Assistant), it sees superior performance for its use cases vs closed third-party models, because it can post-train open models on its unique data. ⁵

Control and security advantage:

• Pinterest highlights the ability to run open models in a secure environment on its own cloud infrastructure, citing the ability to do so without concerns like “phoning home” or exposing the company’s most important data to third parties. ⁴
• It ties this explicitly to the “combination of proprietary signals with… open source models” enabling secure deployment. ²

How this fits the strategic goal: AI-powered visual search and shopping, compounding engagement

Pinterest frames its broader AI strategy as converting user intent into an increasingly “actionable” shopping and discovery journey:

• Management says it has transformed Pinterest into an AI-powered shopping assistant, where every pin a user sees is personalized and served by AI. ⁶
• Pinterest reports completing migration to GPU serving more than 2 years ago, which “materially improved relevance and personalization using larger models and more data.” ⁶
• It describes its “powerful feedback loop” linking user interactions to training signals from its Taste Graph, using scale metrics such as 80+ billion monthly searches and 16+ billion boards created. ⁶

Within this framework, open-source models are positioned as a way to scale AI capabilities efficiently:

• Pinterest connects its early open-source adoption to scaling AI efficiently to serve 640 million users, while also expanding margins and generating significant free cash flow (as described by management). ¹

2) Is MAU the best KPI for UCAN engagement, or are other metrics more informative?

What Pinterest does say about engagement measurement

Pinterest does not claim MAU is the only engagement KPI. Instead, management says it internally looks at “various baskets of metrics” to measure engagement across the platform. ⁷

What UCAN engagement signals they emphasized instead of MAU alone

For UCAN specifically, management highlights that engagement-related actions outpaced user growth in Q2:

• In UCAN, “searches and boards created continued to grow faster than users in Q2.” ⁷

Management also suggests that confidence about UCAN resonance comes from strength across multiple engagement measures:

• It states that across a number of metrics, it sees strength that gives it confidence it is resonating with users in UCAN, “in particular, with Gen Z.” ⁷

Their broader engagement narrative includes repeatable intent-to-action behavior

Although the excerpt doesn’t provide UCAN-by-metric dashboards, the overall AI narrative supports why “actions” are likely more diagnostic than MAU alone:

• Pinterest describes moving users through shopping journey stages (from inspiration to decision) and says Pinterest Assistant is designed to help users answer later-stage research questions “entirely on Pinterest.” ⁸
• It describes AI-driven personalization as already at scale (e.g., “every pin… personalized and served by AI”). ⁶
  Given that, engagement depth (searches, boards created, assistant interactions, etc.) is arguably more informative about whether AI is improving user value than MAU level.

Conclusion on KPI choice for UCAN engagement

Based on management’s own statements in the excerpt:

• MAU is a useful top-line scale indicator (Pinterest reports ending Q2 with 640 million global MAUs, growing 11% and reaching “another record high”). ⁹
• However, for UCAN engagement quality, Pinterest explicitly indicates it relies on multiple engagement metrics and that in Q2 searches and boards created grew faster than users in UCAN. ⁷

So, MAU alone is unlikely to be the most informative KPI for UCAN engagement. The more decision-relevant indicators (per management’s own “baskets of metrics”) include action-based engagement such as searches and boards created, because they track whether user intent is translating into platform behaviors. ⁷

3) Bottom line

• Open-source AI in Pinterest’s strategy is not just “using open models”—it’s a model-agnostic deployment strategy where Pinterest post-trains open models on its data for performance, runs them securely in its own environment, and uses them to scale AI efficiently (with management reporting <8% cost per transaction vs comparable closed proprietary models). ¹⁴⁵
• For UCAN engagement, Pinterest signals that MAU is not sufficient by itself. Management says it uses multiple engagement metrics, and specifically points to searches and boards created growing faster than users in UCAN in Q2—suggesting these are more informative than MAU for assessing engagement depth and resonance. ⁷⁹