20 Million Users, Yet None of Tome's Five Founders Could Make a Decent Slide Deck

Source: a16z | Published: 2026-09-16T14:30:12Z

Tome had 20 million users, but its five founders privately admitted they couldn't make a decent presentation themselves — and that's the real reason Keith ultimately walked away.

That was the real reason Keith Peiris ultimately decided to walk away from Tome. Not slowing growth, not fundraising pressure, but the team's honest assessment of their own product: they couldn't imagine a discerning professional presenter becoming dependent on this tool. The leap from GPT-3.5 to GPT-4 was right in front of them, and some suggested cutting the team and waiting for the technology to mature. Keith's conclusion: no matter how good the model got, it still didn't understand the fundamental problem — the relationship between presenter and audience. Waiting would only produce a better throwaway tool, not the company they wanted to build.

From Twenty Million Users to "Use Our CRM and We'll Give You Free Office Space"

Among Tome's users, B2B adoption was concentrated in sales and marketing. Keith secured free pilots with 12 large companies, originally planning to help them build new business decks and proposals. But those sales teams quickly started asking: can you help us qualify leads? Can you analyze accounts for expansion decisions?

So Keith's team started integrating with customers' CRMs, call recorders, and data warehouses. They quickly discovered that the real challenge wasn't completing these tasks — it was the data itself. Systems contradicted each other. CRM records didn't match what was said on recorded calls.

Cleaning up that data mattered more than generating any output.

That realization led them toward CRM. But their first version — a go-to-market assistant — had users, not paying customers. The reason was simple: they didn't own the data, there were ten similar products on the market, and they had no pricing power.

So they cut the team and rebuilt from scratch, working in the dark for four months. Then came a new problem: no one wants to adopt a four-month-old CRM.

Keith posted on X and LinkedIn: come to our office, use our CRM, free. They found 10 startups. These companies used the barely functional product every day, firing off feedback in Slack every two hours — complaining it was too slow, features were missing — but they wouldn't leave.

This was completely different from Tome. Tome had twenty million users, but it didn't feel like this. Here was a barely working product that people actually cared about.

Activity Log: A CRM Primitive Borrowed from Facebook's Timeline

Three of Lightfield's five founders came from Facebook. They brought a simple intuition: the most important thing in a CRM is relationship modeling, and relationships are fundamentally about timelines.

They called this the activity log — a record of when you first contacted a company, what was said, which meetings were held, which documents were exchanged, followed by product usage behavior and payment history. The system uses this log as a foundation, then reverse-engineers updates to traditional CRM fields.

On the data structure side, they first tried fully unstructured storage, but queries were too slow — a needle-in-a-haystack problem. They landed on "semi-structured": unstructured data lives in the activity log, and the system uses the log to infer causal relationships.

The more critical design decision: completely schema-free. In traditional CRMs, if you build the wrong data model, there's no recovery — wrong stages, wrong fields, and you can't make the sales team go back and fix the data. Lightfield's approach: connect your email, call recordings, and data warehouse first, assemble relationships in real time, and if you want to change fields later, just backfill them from the activity log. You log in, hit sync, wait five minutes, and it's there.

An Alzheimer's Patient Found a Cutting-Edge Treatment in Days

Lightfield has a customer called Power, which helps pharmaceutical companies find clinical trial participants while also running a patient marketplace — where patients with various conditions search for cutting-edge treatments. The business has both B2B and B2C sides and involves complex two-sided matching.

They loaded their entire business model into Lightfield, built automated pipelines pulling data from the FDA and ClinicalTrials.gov, constructed a world model of global clinical trials, and ran matching on both the patient and pharma sides.

An Alzheimer's patient used this system to find a cutting-edge treatment in a matter of days.

This case illustrates why Lightfield's architecture matters: it allows fully customized object and relationship models, because schema doesn't constrain you here. For a company like Power, with a genuinely unique business model, this is a capability that simply didn't exist before.

Pure Consumption Pricing: The Company's Worst Three Weeks

Lightfield's pricing went through two extremes.

The first version was pure seat pricing, modeled on Salesforce and HubSpot. Customer acceptance was decent, but a serious problem emerged: usage distribution was wildly uneven, with top users consuming ten thousand times more than bottom users. Per-seat pricing wasn't sustainable.

So they switched to pure consumption pricing, with all operations converted to credits. The result: nobody touched anything. Lots of signups, zero behavior.

Those were the worst three weeks in the company's history.

They eventually interviewed customers and bucketed Lightfield's work into four categories: day-to-day CRM work (updating records, filling fields) that customers want at a fixed cost — no budget uncertainty; pipeline generation, where customers are willing to pay usage-based fees because the ROI is visible; workflow automation, also usage-based; and intelligent forecasting and scenario planning, for which customers are willing to pay a separate alpha premium.

The final model: a platform fee plus per-seat coverage for core CRM, with everything else on consumption.

Why They're Not Doing Outcome-Based Pricing

On outcome-based pricing, Keith's view is direct: if they run outbound for OpenAI, conversion will be high. If they run outbound for a seed-stage company with no website, results will be terrible. Sales outcomes depend heavily on the customer's own product-market fit.

So for now, they can only charge for the work — not the results.

Greenfield Customers Are a Marketing Asset, Not Just Revenue

Lightfield's early customers are primarily Silicon Valley startups. Keith is candid about this: Silicon Valley isn't the most efficient source of revenue — real scale revenue comes from broader markets.

But he sees serving Silicon Valley startups as an efficient way to build marketing assets. These companies might have three go-to-market employees today, but they've raised $200 million and will be large someday. Nail it, and you have reference logos. When you walk into health tech or manufacturing, you can use those logos to open doors.

"When I walk in and say 'we've done great work in health tech, let's talk' — that's more effective than any advertisement."

Lightfield performs well in health tech partly because they invested heavily in security early on — signing BAA agreements, running penetration tests. Those investments are now credentials that unlock access to heavily regulated industries.

No Swim Lanes: Priorities Can Be Rewritten Every Day

One of the internal reasons Tome failed was what Keith calls "playing house" — product, marketing, and CS leads each guarding their own turf, with cross-functional feedback triggering defensiveness. That made the company unable to pivot quickly when it needed to.

Lightfield has 40 people and one daily standup for the whole company. Everyone stack-ranks the most important problems together, and whoever has capacity solves them — whether it's an engineering, CS, or product issue. Priorities can shift daily and are reassessed weekly.

This works because the tools amplify generalist capability: anyone can use Lightfield to quickly get up to speed on a customer; engineers can give LLMs access to Figma; CS can use Lightfield to automatically create tasks in Linear. Expertise still matters, but any project can be led by an engineer, designer, or CSM.

What Keeps Keith Up at Night: An a16z Portfolio Company Switched Back to Salesforce

Keith mentions that before founding Lightfield, he read a report documenting companies that had migrated back to Salesforce from newer CRMs. One was an a16z portfolio company: they'd been using a startup CRM but waited four months for dashboards that never came, and eventually switched back.

That story keeps him on edge. Lightfield has strong appeal for new companies, but the team must keep building — so users never find themselves thinking "maybe I should go back to the old world." Losing an early customer isn't just one contract; it's a company that might grow into a hundred-person sales org.

For Anyone Mid-Pivot

Asked what advice he'd give his past self, Keith's answer was simple: almost all the noise around you doesn't matter. At the time, people were debating how to reprice options; someone thought the office food wasn't good enough — none of it matters.

Find a real pain point, have genuine passion for that pain point, and keep your eyes on your customers. Everything else is noise.