Applying AI in B2B

Episode Summary

Executive Summary: The episode argues that enterprise AI is moving from hype to workflow transformation, especially in sales and marketing. Oleg Reginsky and Peter Lawton frame AI as either autopilot (automation of repetitive work) or co-pilot (decision support), emphasizing that data capture, network effects, and early model training will reshape competitive advantage, product design, and how knowledge workers learn on the job.

Main Topics: Why B2B AI differs from B2C (Priority: 5/5): Enterprise AI is harder to deploy than consumer AI because there are fewer customers, higher stakes, stronger security/privacy requirements, and more need to prove ROI before rollout. Data as the foundation of enterprise AI (Priority: 5/5): AI value depends on collecting high-volume, reliable activity data from work processes; companies must weigh the cost of gathering/cleaning data against the value of the resulting models. Network effects and AI moats (Priority: 5/5): The more customers, sensors, or workers contribute data, the smarter the shared graph becomes, creating a virtuous cycle that improves predictions and attracts more users. Autopilot vs. co-pilot (Priority: 5/5): Autopilot removes repetitive work; co-pilot augments judgment with recommendations. Together they shift workers toward higher-value, more human tasks. How AI changes software UX (Priority: 4/5): Future enterprise software will push personalized actionable insights through feeds and guided workflows rather than forcing users to navigate many tabs and reports. Adoption, trust, and enterprise security (Priority: 4/5): Bottoms-up adoption is constrained when AI needs proprietary company data; successful products must work with IT/security teams and demonstrate 10x value to users. Founder lessons and market selection (Priority: 4/5): Oleg’s prior startups taught him that timing and market size matter; product ideas came from firsthand pain with bad CRM hygiene and poor sales visibility.

Key Arguments: B2B AI is fundamentally different from B2C because enterprise buyers demand proof, security, and clear ROI before sharing valuable proprietary data. The best AI opportunities exist where human activity data is generated but not currently captured, enabling models to recommend the best next action. Shared data networks create compounding returns: more contributors improve the model, which improves outcomes, which draws in more contributors. AI can create an arms race among enterprises; companies that delay collecting relevant data may fall behind permanently. Future software will increasingly use graph-based data models because graphs better represent relationships and support machine learning. User experience will shift from manual searching and reporting to pushed, prioritized action feeds that reduce overwhelm and increase focus. The workforce will not simply be automated away; instead, repetitive work will be removed and humans will concentrate on judgment, empathy, and relationship-building. Successful adoption requires proving that the system delivers dramatically more value than the data or privacy friction it introduces. Sales is especially ripe for AI because much of a salesperson’s time is spent on manual data entry and prospecting that machines can handle better. In-product learning will replace some external retraining by teaching users better workflows as they work.

Data Points: Salespeople time on manual data entry: About one third - Used to illustrate how much of a salesperson’s week can be automated by AI. Salespeople time on prospecting: About one third - Framed as a look-alike modeling problem AI can improve. Salespeople time on face-to-face selling: About one third - The remaining portion reserved for human relationship-building. Time for CRM cleanup anecdote: 1 week - Oleg described being forced to clean Salesforce record by record for a week. Symantria market size: 20–30 companies - Example of a successful but small market for sentiment analysis API. Symantria market share: ~80% - They owned most of the niche market within three years. Symantria revenue scale: Single millions - Despite dominant share, total revenue remained limited because the market was small. Startup timing window: 5–10 years - Oleg predicted most industries will shift to activity-data collection and next-best-action models within this period. Historical CRM database eras: 3 eras mentioned - Hierarchical databases, on-prem SQL, and cloud SQL were cited as successive data-model shifts.

Pivotal Quotes: "There are two modes in which AI operates with people. We call it one autopilot, and another one is co-pilot." — Oleg Reginsky: Core framework for understanding AI’s role in enterprise workflows. "You have to be visually in a very simple, explainable way, promising and delivering 10x the value of being on the system than being of the system." — Oleg Reginsky: Explains how to overcome user privacy concerns and adoption resistance. "If you miss the AI boat, the results are very different." — Oleg Reginsky: Highlights the strategic importance of early data collection and model training.

Implications: Enterprise AI winners will own unique activity data, not just models. Expect software to become more prescriptive, workflows more automated, and knowledge workers more focused on judgment-heavy tasks. Early data moats may determine future market leaders.