Quality-Volume-Speed Triangle

Model builders care about three things in order: quality first, volume at scale, then speed

Garrett Lord
Inside the expert network training every frontier AI model

Quality-Volume-Speed Triangle

"They care about three things. They care about quality first and foremost. And then, the other huge problem you have is volume. And then, the other thing I would say model builders care about is speed." - Garrett Lord

What It Is

When serving AI frontier labs or any demanding B2B customer with stringent requirements, buyers prioritize three attributes in a specific hierarchy: Quality, Volume, and Speed. Understanding this ordering helps providers focus investments correctly and avoid the trap of optimizing for the wrong dimension.

Quality is non-negotiable and must be solved first—bad data actively harms models, like teaching a student incorrect information. Only after quality is established does volume matter—the ability to produce large quantities of high-quality output. Speed is the final priority—how quickly you can deliver that volume.

How It Works

Quality First: In AI training data, a single low-quality data point can degrade model performance. Model builders are training systems that will make critical decisions; they cannot accept garbage. If you don't have quality, nothing else matters. You won't retain customers, and you'll damage your reputation.

Volume Second: Once quality is proven, the next challenge is scale. One perfect physics problem doesn't move the needle. Customers need thousands of high-quality examples in advanced domains like chemistry, mathematics, and physics. Can you actually produce the quantity required while maintaining quality standards?

Speed Third: When quality and volume are established, speed becomes the differentiator. AI labs run multiple experiments simultaneously with different hypotheses. When one hypothesis shows promise, they want to rapidly expand data collection in that area—potentially within days. The provider who can quickly pivot and scale wins the next contract.

How to Apply It

  1. Solve quality before scaling - Don't chase volume until you've proven you can deliver quality consistently. Build internal QA processes, post-training evaluation, and customer feedback loops.

  2. Invest in scalable quality - Once quality is established, build systems that maintain quality at scale: training programs, community learning, peer review, automated checks.

  3. Build speed into your architecture - Design your operations to rapidly reallocate resources when customer priorities shift. Experiments succeed or fail in days; you need to pivot at that pace.

  4. Communicate in this order - When selling, lead with quality credentials, then demonstrate scale capability, then highlight speed. Match the customer's priority hierarchy.

When to Use It

  • When prioritizing product investments for demanding B2B customers
  • When building operations for data, content, or service businesses
  • When your customers are running scientific or experimental processes
  • When deciding whether to optimize for throughput vs. quality

Example

Garrett Lord describes how Handshake approaches AI data delivery:

  1. Quality: Hire from top physics programs (Stanford, Berkeley, MIT), build internal post-training teams that evaluate every unit of data, rent GPUs to verify data quality
  2. Volume: Leverage existing network of 500,000 PhDs and 3 million master's students to scale production once quality is proven
  3. Speed: Structure operations to respond within days when labs want to expand promising hypotheses

The hierarchy is strict: they don't try to be fastest at producing low-quality data. Speed only matters after quality and volume are established.

Source

  • Guest: Garrett Lord
  • Episode: "Inside the expert network training every frontier AI model"
  • Key Discussion: (00:20:30) - Explaining what AI lab customers care about and in what order
  • YouTube: Watch on YouTube

Related Frameworks

Back to Growth