Keith Peiris built a presentation tool called Tome to 25 million users, then shut it down because none of the founders liked what they had made. Lightfield, the CRM he built afterward, has since raised a $47 million Series A led by a16z.
Enterprise software has spent two decades organizing customer data into fixed fields and tables. Peiris's bet is the opposite: store almost nothing in a fixed schema, let a language model reconstruct the record from raw emails, calls and meetings, and fill in the fields later.
"I think in many ways your CRM is maybe harder to move off of than your bank."
Peiris ran Tome's presentation product to 2 million users a month before killing it, then rebuilt three times before landing on what Lightfield calls a "business world model," and he still runs the company's own sales meetings from a spreadsheet.
The full interview is covered here so you can skip it. 53 minutes of audio, 11 minutes of reading.
Here are the 8 lessons that matter.
👤 Guest: Keith Peiris, Co-founder & CEO of Lightfield, who built Tome to 25 million users before shutting it down to start over
🎙️ Hosts: Alex Rampell and Joe Schmidt, general and AI-apps partners at a16z who cohost its interview show
📰 Published: 16 September 2026 on the a16z podcast
🔴 YouTube | 🟣 Apple Podcasts | ⏱️ 53 min | ✅ Time saved: 42 min
Key Takeaways
Peiris shut down Tome, a presentation tool with 25 million users, because the founders didn't like the product "None of us like the product which is kind of a funny thing to say"
Lightfield has raised a $47 million Series A led by a16z
Lightfield stores customer data as a chronological "activity log" instead of fixed fields, then infers the CRM record from it The team tried a fully unstructured version first and found the queries took too long
Early customers were recruited with negative pricing — free desk space in Lightfield's own office in exchange for using the product
Pure seat pricing and pure consumption pricing both failed before Lightfield landed on a hybrid: platform fee for core CRM work, consumption for pipeline generation and automations "was using 10,000x more than the tail" once seat pricing let one account run wild
Lightfield gives the product away free to everyone inside a customer's company, not just the sales team, to build internal network effects against being ripped out
The company runs with no fixed roles — a 40-person team meets one daily standup and whoever is free picks up the next problem
Peiris's biggest fear is speed: a16z's own portfolio company Eleven Labs left a rival CRM for Salesforce after waiting four months for a dashboard
1. The Product Nobody Loved
Peiris and his co-founders built Tome, a GPT-3-era presentation generator, to explosive scale — 2 million users a month, with people lining up when the company didn't have enough inference capacity to serve them.
The growth was real, but the product was wrong. "None of us like the product which is kind of a funny thing to say." He said a founder has to love what they're building, and the team could never make Tome produce a presentation good enough for professional use
The model's limits, not the market's, killed it. He said the technology couldn't hold enough context about the presenter, the audience and the relationship between them, and "no amount of general reasoning gets you past that"
Waiting for better models was considered and rejected. The team discussed shrinking down and waiting through the jump from GPT-3.5 to GPT-4, but concluded the underlying context problem wouldn't be solved by a bigger model alone
The team also weighed burn against patience: waiting out a model generation is a luxury a startup spending cash does not have
2. Finding the CRM Problem
Talking to Tome's roughly 25 million users, the team found its B2B use case was sales and marketing teams using the tool for pitch decks — and ran twelve free pilots to find out what those teams actually needed.
The pilots expanded past decks almost immediately. Customers asked Tome to research prospects, qualify leads and help understand companies for expansion — work that had nothing to do with presentations
Access to a company's "context box" turned out to mean its CRM, call recorder and data warehouse. Peiris said the team assumed integrating that data would be straightforward
The data itself was the hard part, not the integration. "The hardest part of doing this work was actually making sense of all of the data across all of these disparate systems," which he described as incomplete and conflicting — the call recorder and the CRM often disagreed about the same account
That reorganization problem became the actual company. Peiris said reorganizing a company's reality into something both machines and humans could understand felt like a more interesting and more enduring business than the one Tome was already running
3. The Business World Model
Lightfield's core design bet is that a chronological record of every interaction — emails, calls, meetings, product usage — is a better foundation for a CRM than a set of fixed fields.
The team built an activity log first, modeled on a Facebook-style timeline, because three of Lightfield's five founders came from Facebook. The log records first contact, every meeting and document, and eventually product usage and payment behavior
The activity log drives the traditional CRM fields, not the other way around. The system uses the log to trigger record updates, but the chronological log stays as the underlying record
A fully unstructured version was tried and abandoned. Peiris said the team "actually tried going fully unstructured" and found the queries took too long, landing instead on a semi-structured approach that stores unstructured data inside the activity log and lets the system infer causality from it
The company deliberately avoided a data model up front. Peiris said CRM consultants told him the most consequential decision they help customers make is the data model, and getting it wrong is unrecoverable: "It's over, right? If you get the wrong stages, the wrong fields, you can't get the reps to go back in time and fill it out." Lightfield's answer was to connect a customer's email, calls and data warehouse and assemble the schema afterward, so fields can be refilled later if they change
One customer, Power, models both sides of a two-sided marketplace inside Lightfield — matching pharmaceutical companies looking for clinical trial participants with patients seeking frontier treatment, using automations that scrape the FDA and clinicaltrials.gov. Peiris said Lightfield helped one such patient with Alzheimer's find a treatment within days
4. Negative Pricing
Lightfield's first ten customers were paid to show up, not the other way around.
The company posted an offer on X and LinkedIn: free desk space in Lightfield's own office for anyone who would use its four-month-old CRM. Ten startups took the offer
Those customers used the barely-finished product constantly and complained about it hourly. Peiris said they were "in it every day" and gave feedback "every two hours," despite the product missing features and running slowly — a signal he read as real engagement rather than politeness
The naive starting assumption became the strategy. Coming from outside the CRM industry, the team assumed the most important thing was modeling the business relationship rather than powering low-level automation or forecasting, and built around that instead
5. Greenfield Versus Brownfield
Rampell framed the company's market entry in terms he said the firm uses often: selling into a market an incumbent already owns is harder because "it's been trampled."
His framing: Rampell says the best companies "have hostages not customers." He named SAP as an example of a brownfield incumbent that customers stay with reluctantly
He drew the cloud transition as the model for breaking into a brownfield market — not competing head-on with an incumbent's product, but redefining the problem, as cloud vendors did to on-premise CRM systems running on aging mainframes
The alternative strategy is greenfield: ignore the incumbents and win only new companies that have no existing system to replace. Lightfield's early approach followed this path, winning newly founded startups with no CRM history rather than trying to displace Salesforce accounts directly
Rampell said he personally once resisted paying $85 a month for Salesforce and used the free SugarCRM instead, until a new VP of sales insisted on switching — illustrating how a brownfield incumbent gets reinforced by the people companies hire, not just the product itself
6. Give It Away, Win the Firm
Lightfield's answer to a new VP of sales who already knows Salesforce is to make the CRM free for everyone else in the company first.
Every seat outside the sales-led plan is free. Peiris said this both helps Lightfield understand what engineering, support and finance are doing, and builds internal network effects that make the product harder to rip out
The tactic has worked against experienced sales hires who resist switching. He described cases where a new VP of sales says they only know Salesforce, and the rest of the company pushes back because engineering, finance and customer success already depend on Lightfield
Lightfield leans toward pragmatism over ideology in the interface. The company kept spreadsheet views and dashboards for people who want them, while also supporting natural-language and CLI use for those who don't — "I still run all of our meetings from the spreadsheet view," Peiris said of his own use as the company's de facto sales manager
7. From Seats to Outcomes
Lightfield tried both ends of the pricing spectrum before landing on a hybrid.
Pure seat pricing was tried first because it matched the market Salesforce and HubSpot had trained customers on. It was well received, but usage across accounts was wildly uneven — the busiest account, Peiris said, "was using 10,000x more than the tail."
Pure consumption pricing, tried next, killed usage entirely. Peiris called it "the worst three weeks of the company's life" once customers stopped touching the product rather than run up a credit-based bill
The landing point splits Lightfield's work into four buckets. Everyday CRM work (capturing meetings, updating records) sits inside a platform fee; pipeline generation and workflow automations are billed on consumption; forecasting and intelligence work is priced as a premium on top
Outcome-based pricing remains out of reach for now, because outcomes depend on the customer's own product-market fit rather than Lightfield's effort. Prospecting for a company with strong demand would produce great results cheaply; prospecting for an early-stage company with a weak product would not. Peiris said Lightfield has "to charge for the work," and "we can't quite charge for the outcome"
8. No Swim Lanes at Lightfield
Lightfield's 40-person team runs without fixed functional roles, a structure Peiris says fixes a mistake from Tome.
At Tome, functional leaders defended their own areas and resisted feedback outside them, which Peiris says slowed decisions and made the company hard to pivot
Lightfield's alternative: no one owns a swim lane. The whole team meets at one daily standup, ranks the most important problems, and whoever is available picks up the next one
Planning is continuous rather than fixed in advance. The list of priorities can change daily and is reassessed weekly, rather than being locked into a long-range roadmap
AI tooling lets non-specialists contribute across functions. Peiris said anyone can ramp up on a customer through Lightfield's own record, on the design system through Figma, or create tasks in Linear — engineers, designers and customer-success staff all run projects rather than staying in one lane
The company still enforces a quality bar before anything ships, running company-wide "bug bashes" — a practice Peiris said he learned at Instagram — before a feature reaches customers
Bonus Insights
Peiris says Silicon Valley is "almost overhyped" on AI while the rest of the country remains underhyped, because many people tried an early version of a chatbot years ago, saw it fail, and wrote the technology off
Winning Silicon Valley customers is a marketing strategy, not a revenue strategy, in his account. He said most of Lightfield's actual revenue scale comes from outside the Bay Area, and the value of an early Silicon Valley logo is being able to point to a reference customer in healthcare or fintech when selling into those industries later
His biggest fear is speed, and he named a16z's own portfolio company Eleven Labs as the cautionary tale. He said he had read about Eleven Labs waiting roughly four months for a rival startup CRM to build dashboards it needed, and the company eventually moved to Salesforce out of frustration
DIY attempts to build an in-house CRM or "company brain" instead of buying one are common but usually abandoned, Peiris said, once a company discovers that modeling its own customers is harder than expected
His advice to a founder mid-pivot: find the pain, get inspired to solve it, and ignore the noise. He said complaints at the time about office space or stock option repricing "honestly...none of that matters" next to focusing on the customer
Peiris's bottom line is that a CRM built around a chronological record and inferred structure beats one built around a fixed schema, and that the model has to keep proving itself against speed — the wedge that let his company in also lets a customer leave if a rival gets there faster.
Products, Companies & Tools Mentioned
Lightfield (Peiris's CRM, now on a $47 million Series A led by a16z, built around a "business world model" instead of fixed fields)
Tome (Peiris's earlier presentation company, built to 25 million users on GPT-3-era demand and shut down because the founders didn't like the product)
Salesforce and HubSpot (The incumbent, seat-priced CRMs that Lightfield's early seat-pricing model was designed to match, and that Rampell says have "hostages not customers")
SugarCRM (The free CRM Rampell used at an earlier company to avoid paying for Salesforce, before a new VP of sales made him switch)
Power (A Lightfield customer that matches pharmaceutical companies seeking clinical trial participants with patients seeking frontier treatment, modeled entirely inside Lightfield)
ElevenLabs (Cited as a company that left a rival startup CRM for Salesforce after waiting months for dashboards it needed)
Figma and Linear (The tools Peiris says let non-specialists at Lightfield contribute to design and engineering work directly)
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: