---
title: "The Ad Compute Trade-Off: Allocate by Marginal Utility"
description: "Ad ranking is a compute trade-off: more resources per request can improve relevance, but GPU capacity and latency are finite. Learn how to allocate by marginal utility instead of uniformly."
---

# The Ad Compute Trade-Off: Allocate by Marginal Utility

**Type:** [Trade-offs](https://yomu.fyi/visual-explainers/category/trade-offs.md)

![A left-to-right trade-off diagram for ad ranking compute: a benefit panel for better relevance, a cost panel for GPU capacity and latency, and a condition panel for marginal-utility allocation, with a decision bar to route compute where it pays off.](https://yomufyi.s3.us-east-1.amazonaws.com/visual-explainers/c91a282c-8638-435f-a54a-241eadb46e77/image-a3e0fbef-8121-45f4-8cb0-d8c752c8d5d4.webp)

> Ad ranking is a compute trade-off: more resources per request can improve relevance, but GPU capacity and latency are finite. Learn how to allocate by marginal utility instead of uniformly.

## Explanation

Ad systems face a trade-off: more compute per request can improve ranking, but GPU capacity and latency are finite. Meta's Intelligent Delivery uses a Resource Core Model to predict per-request marginal utility, then a hierarchy of decision engines allocates across region, hardware, and host. The lesson is to measure where extra compute actually pays off, not to spend uniformly. Demand signals and campaign controls add conditions: fragmented search/social interest and volatile shopping behavior mean allocation must be revisited as conditions change.

## FAQ

### Why not give every ad request more compute?

Meta's Intelligent Delivery notes that the marginal impact of compute varies across requests, models, and chips. Uniform allocation would waste capacity on requests where extra compute barely changes ranking quality. The system instead targets the system-wide optimum.

### How does the system decide where ad compute should go?

Intelligent Delivery uses a Resource Core Model with 6B sparse parameters to predict per-request marginal utility. A hierarchy of decision engines then allocates across region, hardware, and within-host levels, while agentic AI detects and responds to system shifts.

### What operational controls keep high-volume ad delivery from overrunning?

A campaign-style orchestration layer can apply queue polling, batching, and circuit breakers before starting each workflow. Zillow's notification pipeline uses this pattern at 100 million daily notifications, and reusable schema-driven content modules keep variants consistent.

**Tags:** [A/B Testing](https://yomu.fyi/topic/a-b-testing.md), [Ads](https://yomu.fyi/topic/ads.md), [Advertising](https://yomu.fyi/topic/advertising.md), [Analytics](https://yomu.fyi/topic/analytics.md), [Audience Intelligence](https://yomu.fyi/topic/audience-intelligence.md)

## Related reading

### [How Zillow runs 100 million daily notifications on AWS Step Functions](https://yomu.fyi/post/how-zillow-runs-100-million-daily-notifications-on-aws-step-functions.md)
- Company: [AWS](https://yomu.fyi/company/aws.md)
- Author: Nishita Kheur
- Published: Nov 11, 2026
- Type: Problem & solution

Zillow sends over 100 million daily notifications using a centralized, configuration-driven orchestration layer called Campaign Orchestrator. It runs each notification as a short-lived workflow on AWS Step Functions Express Workflows, chosen for being serverless and lower cost than alternatives. A router applies campaign controls like queue polling, batching, and circuit breakers before starting each workflow. The workflows call reusable, schema-driven content modules from the Universal Content Generator library, with JSONata expressions validated via the TestState API. The new architecture cut build effort from 32 to 9 weeks and lowered ongoing maintenance from 8 to 2 hours per week.


### [Intelligent Delivery: Making Every Unit of Compute Serve Better Ads](https://yomu.fyi/post/intelligent-delivery-making-every-unit-of-compute-serve-better-ads.md)
- Company: [Meta](https://yomu.fyi/company/meta.md)
- Author: Zhaojun Zhang
- Published: Nov 11, 2026
- Type: Problem & solution

Ads ranking models are growing toward LLM scale faster than hardware can keep up, and the marginal impact of compute varies across requests, models, and chips. Meta built Intelligent Delivery, an ML-driven system that allocates resources across the serving stack toward the system-wide optimum. It uses a Resource Core Model with 6B sparse parameters to predict per-request marginal utility, a hierarchy of decision engines for region, hardware, and within-host allocation, and agentic AI to continuously detect and respond to system shifts. Since 2017 through the first half of 2026, Intelligent Delivery has contributed an additional ~10% incremental ranking performance on top of model and system advancements. The system shortens the optimization loop from months to weeks, with AI agents now contributing to about 80% of engineering effort on the most automated mechanism.


### [Your Algorithm: Giving People More Ways to Personalize Facebook Recommendations](https://yomu.fyi/post/your-algorithm-giving-people-more-ways-to-personalize-facebook-recomme.md)
- Company: [Meta](https://yomu.fyi/company/meta.md)
- Author: Jiarui Wang
- Published: Nov 10, 2026
- Type: Explainer

Traditional Facebook recommendation systems learn from implicit engagement signals, which adapt slowly and capture little nuance. Your algorithm is a suite of AI-powered tools that lets people state their preferences in natural language to personalize Feed and Reels. The end-to-end system includes profile generation, preference management, ranking, and delivery optimization, with supplementary tail loads and cache invalidation for responsiveness. For interests added to ‘Interested In,’ four dedicated retrieval sources find matching content and a control layer spaces out recommendations and caps daily contribution, while for ‘Not Interested In,’ layered filtering uses embedding comparison, demotion rules, an LLM reranker, and device cache invalidation. LLMs generate the initial profile and Lately Summary, validate each edit for coherence and duplicates, and provide a final reranking check against ‘Not Interested In’ interests, with a separate LLM acting as a judge for evaluation.


### [Maxxing Is Evolving: Search and Social Tell the Story](https://yomu.fyi/post/maxxing-is-evolving-search-and-social-tell-the-story.md)
- Company: [Brandwatch](https://yomu.fyi/company/brandwatch.md)
- Author: Ksenia Newton
- Published: Nov 10, 2026
- Type: Explainer

Between 2021 and 2026, ‘maxxing’ evolved from niche slang to Dictionary.com’s Word of the Year, starting with looksmaxxing and expanding to any area of optimization. Brands saw contradictory signals: social mentions jumped 372% year over year while search interest dropped 55.7% over three months. The post argues that single-channel analysis misses the fragmentation and cultural shifts driving these trends. Whimsymaxxing, a counter-trend of decorating objects with rhinestones, saw a 169% increase in mentions and drove commercial demand for bedazzling supplies. Slowmaxxing, which rejects optimization culture entirely, grew 4,452% year over year, showing that some of the search decline reflected conscious rejection rather than fading interest.


### [What Consumers Want This Holiday Shopping Season](https://yomu.fyi/post/what-consumers-want-this-holiday-shopping-season.md)
- Company: [Brandwatch](https://yomu.fyi/company/brandwatch.md)
- Author: Written by Helen Booth
- Published: Nov 10, 2026
- Type: Explainer

Consumers are showing early holiday shopping anxiety, with mentions up 5% in August 2025 versus the prior year and a peak of 56,402 conversations in early October. Economic pressure drives strategic behavior: Google searches for discounts rose 11% year over year, and buy-now-pay-later mentions jumped 73% from August to October 2025. Shoppers crave in-store festive experiences, seeking sensory magic over purely transactional online shopping. DIY gifting is surging, with 21% more unique authors discussing homemade gifts and 91% positive sentiment. Brands should reduce promotional noise, offer flexible payment options, and create immersive physical or digital experiences to build trust and engagement.
