Capability Trumps Seniority for Project Leadership

Handshake's approach to building a successful AI data business within an established company demonstrates how to leverage existing assets to create extraordinary growth in the AI era.

Building a New Business Within an Established Company

• Maintain complete separation between the new and existing business

  • Separate engineering, design, operations, and finance teams
  • Dedicated people with no responsibilities in the core business
  • Different physical workspace (separate part of the office)
  • Separate all-hands meetings and onboarding processes
  • Different compensation structure tied to new business hurdles

• Create a distinct culture and operating cadence

  • Set clear expectations about intensity (5 days in office, weekends, 2-3am work)
  • Establish a metrics-oriented operating cadence from day one
  • Implement weekly/monthly/quarterly reporting structures
  • Maintain a flat organization where capability trumps title
  • Assign directly responsible individuals (DRIs) based on capability, not function

• Hire the right people for zero-to-one building

  • Recruit entrepreneurs who have previously started companies
  • Look for people comfortable with ambiguity
  • Be upfront about the chaotic nature of the work
  • Selectively pull top talent from the core business

• Establish a guiding philosophy

  • "Leave nothing to chance" - do whatever it takes to succeed
  • Create urgency by emphasizing the once-in-a-lifetime opportunity
  • Celebrate wins and publicly recognize impact

Leveraging Strategic Advantages in AI Data Collection

• The only true moat in human data is access to an audience

  • Competitors spend millions on acquisition while Handshake has built-in trust with 18M users
  • Decades of trust with users creates higher conversion and retention rates
  • Ability to hyper-target specific expertise (500K PhDs, 3M master's students)

• Focus on expert data as models evolve

  • Models have gotten so good that generalists are no longer needed
  • Experts can identify where models are weak or breaking
  • PhDs can create high-quality data in specialized domains

• Treat experts with respect, unlike traditional data labeling

  • Build community rather than transactional relationships
  • Provide proper training before assigning projects
  • Create cohort-based learning environments
  • Pay rates that reflect expertise ($100-200/hour for specialized knowledge)

• Ensure data quality through rigorous processes

  • Build an instructional design team and assessments team
  • Implement a post-training team to verify data quality
  • Evaluate each unit of data for potential model improvement

Scaling in the AI Era

• When demand is unlimited, execution becomes everything

  • Focus on quality first, then volume and speed
  • Build capacity to quickly test hypotheses and scale successful approaches
  • Maintain trust with frontier labs through consistent quality

• Evolve with changing data needs

  • Shift from text-based to multimodal (audio, video) data
  • Expand into trajectory data (recording complete user workflows)
  • Focus on tool use and step-by-step processes in specialized domains

• Maintain a long-term vision that connects new and existing businesses

  • Use AI data business insights to improve core matching marketplace
  • Apply learnings to transform how employers find candidates
  • Leverage AI to create step-function improvements in core business