Summary of "The AI Opportunity that goes beyond Models"

High-level thesis

AI is the next major product cycle, following PC, Internet, Cloud, and Mobile. Adoption is accelerating because models + cloud + smartphones enable massive scale and real economic value.
Most net-new software revenue today is AI-driven across both infrastructure and application layers.
The most investable and defensible startups turn AI into sustained business moats — not merely model wrappers.

“Best companies have hostages, not customers.” Focus on systems of record and embedded workflows that make switching costly.

Three repeatable investment / company archetypes

Traditional software going AI-native
- Description: Incumbent categories (ERP, payroll, helpdesk, accounting, marketing, etc.) are being upgraded with AI primitives.
- Playbook: Ship AI features that integrate into existing systems of record; convert customers at inflection points (new company creation or when a customer hits complexity thresholds).
- Risk: Brownfield displacement is hard; best to target greenfield or customers at a clear inflection.
“Software eats labor” (new category creation)
- Description: Replace or augment human labor in roles like collections, intake, transcription, legal intake, and call centers.
- Why it matters: These markets often exceed traditional software TAM because they map directly to human jobs.
- Playbook: Build products that perform 70–90% of tasks humans did, and capture value via percentage uplift (e.g., higher collection rates) or outcome-based pricing.
Walled gardens / proprietary data + finished product
- Description: Acquire or create exclusive, hard-to-recreate datasets and deliver finished, high-value products (not raw data).
- Playbook: Digitize archives, license content exclusively, build specialized models and workflows, and sell the finished workflow/product (higher ASP than raw data).
- Example verticals: medical literature, legal records, flight ADS-B data, WHOIS history, contracts/procurement.

Key frameworks, processes and playbooks

Bingo board: systematic checklist of verticals/systems of record (ERP, payroll, collections, support, procurement, etc.) used to evaluate AI opportunities.
Greenfield vs Brownfield: choose brand-new customers (greenfield) or target inflection points where customers must migrate away from legacy systems (brownfield).
Differentiation vs Defensibility:
- Differentiation: AI features can distinguish a product.
- Defensibility: typically requires unique data, control of end-to-end workflows, and high switching costs.
Process / investment playbook:
- Process + Interrupt model: maintain a repeatable discovery/writing/benchmarking process but prioritize exceptional inbound opportunities.
- Adverse vs Positive selection: prefer high-quality, sought-after deals (positive selection).
- Two-key decisioning: partner conviction plus investment committee signoff; defer to on-the-ground experts when needed.
Aggregation thesis for consumer AI: aggregators that orchestrate multiple models (a single pane of glass) win over single-model providers.
Pricing playbooks: move from per-seat to outcome/value pricing where appropriate (e.g., charge per recovered dollar in collections rather than per seat).

Concrete case studies & examples (with business takeaways)

Ramp (corporate card / expense management)
- January 2025 spike in AI usage across enterprise customers — signal of enterprise pull for productivity gains.
Salient (auto-loan servicing / collections)
- AI agent collects evidence, supports many languages, and achieves ~50% higher collection rates versus human staff.
- Pricing should be tied to increased recovery rather than pure cost savings.
Eve (plaintiff legal platform)
- Voice agent + end-to-end intake dramatically increases case evaluation throughput.
- Owning intake + outcomes creates non-public outcome data that both improves intake and expands addressable market (can profitably pursue smaller cases).
Open Evidence
- Exclusive licensing of medical journals + AI interface yields higher-value product for clinicians than generic LLMs.
Vlex
- Digitized legal records in Spain; adding AI quintupled revenue — demonstrates that walled garden data + finished product substantially increase monetization.
FlightAware, PitchBook, DomainTools, AskLeo
- Public or obscure raw data becomes valuable when aggregated, cleaned, indexed, and embedded into workflows.
Slingshot (consumer mental health)
- Uses AI scribe to collect proprietary therapy notes and train a consumer product — pipeline from professional data collection to consumer monetization.
Toast, ServiceTitan, MindBody
- Examples of vertical operating systems that succeeded by owning more of the workflow and financial services; useful playbook for vertical AI startups.

Key metrics, KPIs and observed signals

Adoption: ~15% of adults use ChatGPT weekly (adoption signal).
Usage: rapid growth in minutes per user for conversational AI — described as “astronomical.”
Enterprise signal: Ramp’s enterprise usage jump (Jan 2025) as an example of product-market pull.
Product impacts:
- Salient: +50% collections rate (core growth metric and pricing lever).
- Legal intake: AI improves conversion and qualification (historical baseline ~1 case accepted per 100 leads; AI can shift minimum viable case economics).
- Call centers: example client had $50M/year call center with 40–70% annualized churn per employee — large labor-replacement opportunity.
Growth velocity: several AI startups are scaling from zero to tens or hundreds of millions in revenue much faster than historical software norms (1–2 years possible).

Actionable recommendations for founders / operators

Focus on outcomes: price for outcomes or finished products instead of raw model calls or per-seat metrics where possible.
Own the system of record / end-to-end workflow: deep embedding increases switching costs and retention.
Prioritize proprietary data: acquire, digitize, and exclusively license sources labs don’t have; convert raw data into finished, actionable products.
Target greenfield or clear inflection points: go after customers without legacy products or those hitting complexity thresholds (multi-entity accounting, multicurrency, scaling stages).
Verticalize strategically: vertical operating systems paired with adjacent financial services or platform hooks (payments, lending) can drive massive value.
Use AI to augment sales/engineering: large buyers often need forward-deployed engineering for integration—invest accordingly.
Consider acquisition roll-ins: buy distribution/customers where organic GTM is hard, then transform with AI.
Differentiate for defensibility: multilingual support, regulatory statute ingestion, or state-level legal nuance are practical moats.
For consumer AI, build aggregator experiences that orchestrate multiple models and deliver superior UX/finished outputs.

Go-to-market and sales observations

Enterprise buying: strong inbound demand in many categories; some AI-native companies have required minimal outbound sales (e.g., Eve).
Sales investment: for large corporates expect sales headcount plus forward-deployed engineering and integration resources.
Retention: enterprise retention is healthy when products are embedded and drive measurable value; consumer retention is promising for differentiated products.
Pricing: movement away from per-seat models (deprecated for support) toward outcome or outcomes-plus-platform pricing.

Risks & competitive dynamics

Easy productization: low-friction product creation (“vibe coding”) raises competitive risk; defensibility must come from data, workflow embedding, and distribution.
Incumbents vs startups: incumbents will adopt AI and monetize their hostages; startups must find greenfield, inflection points, or vertical data moats to outpace incumbents.
Model-provider encroachment: OpenAI, Google, and others may move into applications, increasing the premium on proprietary data and end-to-end productization.
Talent & speed: execution speed and team expertise are critical — sourcing and supporting top founders and teams matters more than ever.

Miscellaneous concrete playbook items

Convert raw data into “finished meals” (finished outputs) instead of selling raw signals.
Make products multilingual and regionally aware (statutes, laws, customer language) to expand TAM and defensibility.
Benchmarking & content strategy: publish public benchmarks and content to build thought leadership and inbound deal flow (Apps Fund approach).

Presenters / primary sources mentioned

Alex Rampel (The Apps Fund) — lead presenter
David (Apps Fund partner / senior speaker referenced)
Anish (Apps Fund partner / senior speaker)
Jen (moderator)
J (speaker, referenced)
Ari (CEO of Salient) — case study CEO referenced
Frank Chen (partner, referenced)
Chris Dixon (referenced for prior product cycles post)
Noam Shazeer (co-author of “Attention is All You Need” — referenced)

Primary companies / case examples cited

Ramp, Salient, Eve, Slingshot, Open Evidence, Vlex, FlightAware, PitchBook, DomainTools, AskLeo, 11 Labs, OpenAI, Microsoft, Toast, ServiceTitan, MindBody, Salesforce, Workday, Netsuite, Mercury, QuickBooks, UiPath, Zendesk, Bill.com, SAP, Ancestry.com

Optional follow-ups (offers)

One‑page checklist founders can use to evaluate defensibility for an AI startup (data sources, workflow embedding, pricing levers, GTM motions).
Prioritized “bingo board” verticals list with TAM and specific product ideas.