CHAPTERS
- 0:05 – 1:05
Outset’s core idea: AI-led interviews at survey-scale depth
Harj introduces Aaron Cannon and Outset’s mission: modernizing customer research. Aaron explains how Outset uses AI to conduct real, back-and-forth interviews that combine the speed/scale of surveys with the depth of qualitative conversations.
- •Traditional research tradeoff: surveys are shallow; human interviews are slow and expensive
- •Outset’s innovation: AI leads live conversations with real participants
- •Goal: achieve interview-level insight without interview-level cost/time
- •Positioning as a new kind of customer research workflow
- 1:05 – 1:18
Scale and traction: millions of AI interviews with major enterprises
Aaron describes Outset’s current scale and customer profile. The platform is already used by large global brands and has executed millions of AI-led interviews.
- •Enterprise adoption from well-known brands (e.g., Google, Microsoft, Nestlé)
- •Millions of interviews completed using AI interviewers
- •Outset becoming a key mechanism for organizations to learn from users at scale
- 1:18 – 2:55
Being too early: launching a category before buyers had a mental model
The conversation rewinds to YC W23, when AI interviews were novel and hard to place within existing research practices. Despite strong interest and many demos, conversions were low because organizations didn’t yet know how to deploy the concept.
- •2023 AI landscape: many teams still learning basic LLM concepts
- •High curiosity produced many demos, but little budgeted demand
- •Early-stage symptom of category creation: attention without adoption
- •Founder insight: novelty alone doesn’t create an internal buyer roadmap
- 2:55 – 3:32
The first product version: conversational survey + CSV export
Aaron details the initial Outset build: a simple chat-style interview experience driven by early GPT models. The synthesis layer was minimal—results were delivered via a basic CSV export.
- •Wrapped GPT in a chat interface to ask and adapt questions
- •Participants responded like a survey, but with conversational flow
- •Early synthesis was rudimentary (CSV export)
- •Framing: “interactive” or “conversational” survey
- 3:32 – 6:11
Why enterprises didn’t buy: no budget line item and no category fit
Harj probes why demos didn’t convert. Aaron explains that enterprise procurement depends on existing line items or strong validation to justify a new one, and Outset didn’t map cleanly to either quant survey tools or qual interview budgets.
- •Enterprises buy against predefined budget categories
- •Outset didn’t fit “quant survey” or “qual interview/focus group” buckets
- •Incumbents (e.g., Qualtrics/Medallia) owned existing spend
- •Core friction: behavior change plus new budget justification
- 6:11 – 8:15
Market education playbook: case studies, validation, and forward-deployed time
Aaron shares how Outset got early traction through enterprise validation, starting with Weight Watchers. They heavily discounted the first deal to secure a case study and invested significant hands-on time to create a successful blueprint for adoption.
- •First customer: Weight Watchers via founder network connection
- •Strategy: prioritize a case study over near-term revenue
- •Early pricing included deep discount to win validation
- •Forward-deployed effort to find a repeatable successful use case
- 8:15 – 9:41
Model improvements as a compounding advantage: better interviews and better synthesis
Outset benefits directly from rapidly improving foundation models. Aaron frames customer insights as an “uncapped” demand problem—better models don’t just automate work; they unlock more insight, more use cases, and more business impact.
- •Customer research demand is effectively unlimited; more insight drives growth
- •Better models improve both interviewing quality and synthesis depth
- •Outset evolved from early models (e.g., DaVinci-era) to frontier model stacks
- •Sales motion: growth story—accelerated insights accelerate the business
- 9:41 – 12:57
Breakthrough insight: participants share more with AI than with humans
A concrete Weight Watchers example illustrates the unique value of AI interviewing. Outset enabled 100 deep interviews in hours (vs. months) and elicited unusually candid responses due to reduced fear of judgment.
- •100 interviews completed within hours—impossible with traditional methods
- •AI can unlock “ground truth” via perceived anonymity and less social pressure
- •Not only cheaper/faster—qualitatively different data becomes accessible
- •Implications for research: richer motivation and sentiment discovery
- 12:57 – 14:19
Tailwinds arrive: from “experimental bet” to market consensus in 2025
Aaron explains the company’s recent growth inflection as category adoption caught up. Outset’s early market education meant that once consensus formed, the company was already top-of-mind and associated with the new category.
- •Long period of skepticism required persistent education and proof
- •2025 marked a shift from tentative exploration to mainstream acceptance
- •Category creation advantage: build mindshare before the inflection point
- •Result: rapid growth over the last 18 months
- 14:19 – 15:24
Next capabilities: voice, vision, and real-time co-design in interviews
The product evolves beyond text chat into multimodal research. Aaron describes voice-based interviewing, computer vision for contextual cues, and “co-design” where AI iterates designs live based on participant feedback.
- •Shift from explaining LLMs in 2023 to widespread AI fluency now
- •Multimodal interviewing: voice plus visual intelligence
- •Contextual probing: screen/face/environment signals can inform follow-ups
- •Co-design: AI changes designs in real time and gathers immediate reaction
- 15:24 – 16:36
Outset Digital Twins: turning interviews into customer simulations
Aaron announces Digital Twins and the Simulations Lab: creating interactive representations of real users to predict behavior. The thesis is that businesses ultimately want to simulate the future, and twins make that accessible continuously.
- •New launch: “Outset Digital Twins” as part of a Simulations Lab
- •Goal: simulate what users would say/do, not just collect past feedback
- •Research as prediction: forecasting adoption, purchase, and response
- •Twins complement ongoing research by extending insights into simulation
- 16:36 – 19:39
How twins are built and why they matter: grounding interviews → persona core → org-wide access
Aaron explains the inputs and workflow for creating twins: selecting key audiences, conducting hour-long grounding interviews, and building a representative “persona core.” The value is democratizing customer understanding across functions without repeated study cost.
- •Audience selection: current customers vs. target prospects
- •Creation method: extensive AI-run grounding interviews with real people
- •Output: a digital twin anyone in the company can query anytime
- •Use cases: messaging tests, pricing scenarios, rapid iteration without new budget
- 19:39 – 21:18
Aaron’s founder journey: learning under founders, then starting before the “perfect idea”
The discussion closes with Aaron’s path to entrepreneurship, including working closely with founders at startups like Triplebyte. His advice is to start now—don’t wait for a fully formed idea—and learn by doing with the right co-founder.
- •Early career: product roles at startups; first product hire at Triplebyte
- •Value of proximity: pattern-matching startup phases by working with founders
- •Founder advice: life is short—commit before everything is figured out
- •Partnering with co-founder and exploring iteratively as a starting point
