a16z Podcast

Keith Peiris on why intelligence beats schema in the CRM

Keith Peiris· Co-founder & CEO at Lightfield
·~52 min·English·a16z
AgentsAI CompanyBusiness StrategyLLM
TL;DR

The CEO of Lightfield explains how an AI-native CRM wins by treating intelligence as worth more than structure: schema-less data, an activity-log architecture, experimental pricing, and an org with no swim lanes.

01Core Mental Model

Intelligence Is Greater Than Schema

Lightfield ingests your email, calls, and warehouse, assembles the relationships, and lets you fill in the fields later, so the data model is no longer a decision you can never undo.

If you get the wrong stages, the wrong fields, you can't get the reps to go back in time and fill it out. It's over.

— Keith Peiris, a16z Podcast
Key Insight
The varchar(25) history the hosts recall shows structure was always a workaround for scarce compute, where you predefined a column's size to save bytes. When intelligence is abundant, the rigid schema becomes optional rather than foundational.
02The Pivot

None of Us Liked the Product

Tome hit explosive growth as an AI presentation tool, yet Keith hard-pivoted on instinct because a founder has to love what customers use, and the model could not carry enough context to serve professionals.

deep down, I think at an instinctual level, um none of us liked the product.

— Keith Peiris, a16z Podcast
Key Insight
The signal to pivot was not a metric but the founders' inability to picture a discerning professional depending on the tool. He treats a missing-context problem as a ceiling that more raw reasoning will never lift.
03Finding the Problem

Follow the Heat

Following his B2B users into sales and marketing and running twelve free pilots, Keith found the real work was not making decks but reconciling systems that disagreed about the same customer.

if you can reorganize reality for a company in a way that machines can understand and also so for humans to understand

— Keith Peiris, a16z Podcast
Key Insight
The winning problem was not the one he set out to solve. It surfaced from doing the unglamorous reconciliation work, where the call recorder and the CRM held conflicting views of reality, that customers valued most.
04The Architecture

The Activity Log

Borrowing from the Facebook timeline, Lightfield records the chronological relationship first as an activity log, then derives the familiar CRM fields and stages on top of it.

the system can use the activity log to sort of infer causality and work from there. And that's why it like works so much better than a data lake.

— Keith Peiris, a16z Podcast
Key Insight
By making the relationship timeline canonical and the schema derived, changing your mind about a field becomes a re-read of the log rather than a painful migration. A fully unstructured data lake could not do this because the queries were too slow.
05The Cold Start

Negative Pricing

No one wants a four-month-old CRM, so Lightfield gave away free office space to ten startups, and their hourly, angry feedback was the product-market pull Tome's millions of users never gave.

barely working, barely finished product that people are in all the time, where they care about it so much, they're going to give us feedback about it on an hourly basis.

— Keith Peiris, a16z Podcast
Key Insight
The metric that predicted the second company was not scale but engagement under friction. Users who complain every two hours are users who have already made the product part of their day.
06Go To Market

Give It to the Whole Company

Rather than fight to convert a Salesforce-trained VP of sales, Lightfield gives itself free to everyone in the company, so engineering, finance, and customer success all come to depend on it.

This is like how the engineers understand customers. This is how finance does revenue recognition. This is how customer success does account scoring.

— Keith Peiris, a16z Podcast
Key Insight
Giving it away is not a growth hack but a moat. It converts a single-buyer choice into a whole-company dependency, which is what answers a new executive who only knows the incumbent tool.
07Monetization

The Efficient Frontier of Pricing

Lightfield ran both pricing extremes to find the frontier fast, learning that pure seats and pure consumption each failed, then landed on a platform fee plus seat for core CRM and consumption for everything else.

we sort of landed in this platform fee plus seat for um core CRM and we do consumption for everything else and it's landed pretty well.

— Keith Peiris, a16z Podcast
Key Insight
He did not reason his way to a price. He ran both extremes as fast, cheap experiments, because the frontier is found, not designed. Outcome-based pricing stays off the table because the outcome depends on the customer's own product-market fit, not the tool.
08How It Ships

Nobody Has a Swim Lane

The swim lanes that let Tome stop talking to itself are gone: at Lightfield forty people share one standup, stack-rank the problems, and whoever is free picks the next one up.

the bar to start a project is very low at Light Field, but the bar to ship the project is pretty high.

— Keith Peiris, a16z Podcast
Key Insight
The generalist model is downstream of the same thesis. When the LLM can connect anyone to Figma, Linear, and the CRM, specializing by role becomes the bottleneck, so rigid structure gives way to intelligence once again.