MurmurCast
DiscussionInsightful

CRM As A Business World Model: The Future For GTM Teams? | Keith Peiris, CEO @ Lightfield

Topline1h 6m

Keith Peiris, CEO of Lightfield, discusses why his AI-native CRM is disrupting the go-to-market space, having raised $47M Series A and already serving 5,000+ customers. The conversation covers how Lightfield tackles AI reliability through hybrid deterministic/probabilistic approaches, debates whether the SaaS partner ecosystem is dead in the age of AI, and explores whether ecosystem partnerships or integrated platforms will dominate the future of revenue tech.

Summary

Keith Peiris joins Sam Jacobs, Asad Zaman, and AJ Bruno to discuss Lightfield, an agentic CRM that launched in November 2025 and has already achieved 5,000 customers despite being one of the newest players in a crowded space. Peiris explains that Lightfield was born from the observation that as AI models become smarter and more capable, companies need a comprehensive "business world model" rather than traditional CRMs. He argues forecasting has become fundamentally harder due to rapid product cycles, dropping model costs, and constantly shifting market segments—requiring richer data than traditional pipeline analysis.

The discussion delves into the critical challenge of using probabilistic AI technology for deterministic business outcomes like forecasting. Peiris acknowledges that models will always be non-deterministic but frames this as a "harness problem" rather than an unsolvable one. He outlines a hybrid approach: for critical fields, requiring rep approval of AI suggestions to maintain auditability; using real code instead of LLMs for mathematical calculations; and reserving pure AI inference for insight generation. He notes that math should never be delegated to LLMs, but finding patterns and context in customer data is where LLMs excel.

The hosts challenge Peiris on whether human auditing of AI-generated forecasts creates more work than it saves. Peiris responds that through careful implementation—converting prompts to deterministic code for calculations, establishing strong business model foundations upfront, and making fine adjustments iteratively—the system becomes increasingly reliable over time. He positions this as fundamentally different from asking humans to rebuild AI work from scratch.

A major debate emerges about whether the SaaS partner ecosystem is dead. Asad argues that as AI makes building features easier, companies will build more things themselves rather than partnering, reducing ecosystem opportunities. Sam and AJ counter that being 10x better at something (as Granola is at call recording) still creates moat and customer loyalty, and that companies prefer to buy from specialists rather than manage homegrown solutions. Keith takes a middle position: Lightfield is "all in" on the system of record but deliberately avoiding becoming the best call recorder or forecasting tool, instead maintaining strategic partnerships with best-of-breed companies like Granola. He argues that building products to Granola's quality requires serious R&D focus that would distract from the core business.

On the question of why now for Lightfield, Peiris identifies two key factors driving CRM replacement: first, believing that a unified customer/prospect model improves all parts of the revenue funnel (e.g., SDRs pulling real-time insights from CS data, expansion teams seeing renewals early); and second, enabling better scenario planning and strategic decisions about product builds and sales process changes. He emphasizes that companies rip out CRMs for structural reasons, not incremental feature improvements.

Regarding pricing, Peiris describes an evolution from seats-based to pure consumption to a hybrid model. The hybrid approach covers predictable core functionality (AI CRM, call recording, transcription, field updates) with platform and seat fees, while layering consumption pricing on high-ROI activities (pipe generation, workflow automations) and uncapped usage for business intelligence queries where CEOs and revenue leaders have uncapped budgets. He found that when everything was priced on consumption, customers were afraid to click buttons; when it was pure platform fee, power users subsidized light users.

The conversation explores deal economics and why Lightfield's $25-30K ACVs for 200-person companies rival Salesforce's enterprise pricing. Peiris attributes this to agentic work and consolidation of value that traditional CRMs couldn't provide. He emphasizes that Andreessen Horowitz's willingness to fund a $50M Series A was contingent on believing Lightfield could reach a $100B outcome without needing to rip Salesforce off enterprise accounts like Costco—implying the thesis depends on disproportionately large deals in SMB and mid-market.

The hosts discuss that while CRM is historically a crowded space, the reimagining required for AI-native systems—automatic data ingestion via APIs, capturing everything rather than just form fields, building a business model rather than a pipeline tracker—represents a genuinely different approach. This contrasts with how quickly features become commoditized when multiple companies all aim for the same feature set (sequencing, call recording, etc.), creating a "Frankenstein app" problem.

Finally, Peiris addresses customer acquisition strategy, noting that Lightfield started with a PLG motion for seed/Series A companies, refined the product based on feedback, then expanded to the mid-market. The 5,000-customer count reflects that many early customers scaled with the product (some grew from 0 to 100 AEs), providing reference points for upmarket expansion. Current ICP is 20-100 reps.

About this episode

<div> Keith Peiris, Co-Founder and CEO of Lightfield, joins Sam Jacobs, AJ Bruno, and Asad Zaman to talk about building an agentic CRM that launched in November 2025, has been adopted by more than 5,000 customers, and just raised a $47 million Series A led by Andreessen Horowitz. Topics include why every system of record was really a forecasting tool, how to get reliable output from non-deterministic models, and the two reasons companies actually rip out Salesforce or HubSpot. Plus, a spirited debate on whether the SaaS partner ecosystem is dead, why AI-native deal sizes at 200-person companies can rival enterprise contracts, and how Lightfield landed on a mix of seat-based and consumption pricing.</div> <div>  </div> <div>Key Takeaways:</div> <div>  </div> <div>- Plan for models that will always make some mistakes. Keith builds Lightfield on the assumption that models stay non-deterministic, and calls reliability "a harness problem" that they're making fantastic progress against.</div> <div>- Use LLMs for insight and plain code for arithmetic. When Asad asked whether auditing AI forecasts saves anyone any time, Keith described a one-time implementation where Lightfield's head of finance checked every formula: "We converted some things from prompts to real code, right? I don't want LLMs doing math. I want real code doing math." That got the dashboards 95% of the way there, leaving the models to flag which deals were on the cusp because of missing features or competition.</div> <div>- Slightly better AI output won't justify replacing a CRM. After several rip-and-replace projects off Salesforce and HubSpot, Keith said marginally better agentic emails are "not a good enough reason to change a system of record." The two reasons that do move customers are believing one model of the business improves every stage of the revenue funnel, with SDRs pulling real-time case studies from customer success, and higher-level scenario planning, where "the modeling really matters. The data completion really matters."</div> <div>- Put predictable work in the seat price and meter the work with visible ROI. Lightfield tried both extremes, and pure consumption backfired: "It's like going to a store where everything's really expensive, that our customers didn't touch anything." The seat and platform fee now covers the core CRM, call recording, and the data model, while pipeline generation, AI workflow automations, and scenario planning run on consumption, because "most revenue leaders, CEOs… have an uncapped budget for business intelligence."</div> <div>  </div> <div> Connect with the Hosts & Guests:</div> <div>  </div> <div> Host: Sam Jacobs, CEO at Pavilion - https://www.linkedin.com/in/samfjacobs/</div> <div> Host: AJ Bruno, CEO at QuotaPath - https://www.linkedin.com/in/ajbruno3/</div> <div> Host: Asad Zaman, CEO at STA - https://www.linkedin.com/in/azaman1/</div> <div> Guest: Keith Peiris, Co-Founder & CEO at Lightfield - https://www.linkedin.com/in/keithpeiris/</div> <div>  </div> <div> Topline is more than a Podcast:</div> <div>  </div> <div> Subscribe to Topline Newsletter: https://toplinemedia.substack.com/</div> <div> Check us out on YouTube for the #1 video podcast for founders, operators, and investors in B2B tech: https://www.joinpavilion.com/topline-podcast</div> <div> Join the free Topline Slack channel to connect with 600+ revenue leaders to keep the conversation going beyond the podcast: https://www.joinpavilion.com/topline-slack</div>

Key Insights

  • Peiris argues that forecasting has fundamentally changed because models are rapidly improving, product cycles are shortening, and market segments are constantly rewritten, making traditional pipeline analysis insufficient.
  • Lightfield's core thesis is that AI models will become progressively smarter and will need a comprehensive business world model (not just a CRM) to effectively help operators with forecasting, analysis, and decision-making.
  • Peiris frames the non-deterministic nature of AI as a harness problem rather than an insurmountable obstacle, proposing that auditability, use of deterministic code for math, and human approval workflows can make AI-assisted forecasting reliable.
  • The hosts debate whether easier software development will kill the SaaS partner ecosystem; Asad contends it will consolidate everything into platforms, while Sam and AJ argue that 10x-better specialist products (like Granola) still command customer loyalty.
  • Peiris deliberately positions Lightfield as best-in-class on the system of record but maintains partnerships with specialists like Granola, arguing that competing with Granola's quality would require R&D focus that would distract from core CRM goals.
  • Only two structural factors drive customers to replace their CRM: belief that a unified customer/prospect model improves all parts of the revenue funnel, and need for better scenario planning and strategic business decisions.
  • Lightfield achieved $25-30K average contract values with 200-person companies (comparable to Salesforce's enterprise pricing) through agentic work and consolidation of value, not through undercutting.
  • Peiris describes CRM history as fundamentally about forecasting reliability—Salesforce succeeded because sales reps actually used it, enabling trustworthy pipeline forecasts, unlike earlier systems.
  • Lightfield's pricing evolved from seats-based to pure consumption to hybrid, because pure consumption deterred usage and pure platform fees created subsidy dynamics; the final model uses platform fees for predictable functions and consumption for high-ROI activities.
  • Revenue leaders have uncapped budgets for business intelligence (running GPT-6 on the business model to scenario plan), meaning token consumption can be monetized differently than transactional features.
  • The 5,000-customer figure reflects acquisition from early PLG with seed/Series A companies, who then scaled with Lightfield and became reference customers for upmarket expansion.
  • Peiris argues that traditional CRM competitors created feature parity by all adding similar capabilities (sequencing, call recording, forecasting), resulting in undifferentiated products that don't integrate naturally and feel like left-menu lists of disconnected tools.

Topics

AI-native CRM design and business world modelsDeterministic vs. probabilistic AI in forecasting and accuracySaaS partner ecosystem viability in the age of AIDeal economics and pricing models for AI-driven softwareFeature commoditization and integrated vs. best-of-breed strategiesCRM replacement motivations and driversAgentic work and its impact on ACVHybrid pricing strategies (seats + consumption)Data capture and auditability in AI systemsGo-to-market competitive dynamicsCustomer acquisition from PLG to sales-led motionThe role of product excellence and customer love in competitive moats

Transcript

Forecasting now, prediction now is really hard. People are putting out new products every couple of weeks. The model costs are dropping. Segments are being written and rewritten. That's Keith Paris, co-founder and CEO of Lightfield, who just raised a whopping $47 million Series A led by Andresen Horowitz. Why? He's building an agentic CRM already being used by more than 5,000 companies. In today's episode, he shares what makes an AI-native CRM powerful, but also how they're tackling the issue of accuracy and reliability for sellers. The models are always going to be non-deterministic. I see that deterministic problem as a harness problem, and let me explain. We also debate whether the SaaS partner ecosystem is dead. Keith shares…

Full transcript available for MurmurCast members

Sign Up to Access
View original source →

More from Topline

Get AI summaries like this delivered to your inbox daily

Get AI summaries delivered to your inbox

MurmurCast summarizes your YouTube channels, podcasts, and newsletters into one daily email digest.