Lightfield Raised $47M to Kill the CRM Field
A16z led a Series A for a CRM built for companies run by agents — 5,000 customers, 400% net dollar retention, and a founding team that walked away from a presentation tool with 25 million users to build it.
Lightfield raised a $47 million Series A led by Andreessen Horowitz on September 9, with Maverick Capital, Coatue, Greylock, Lightspeed, Audacious Ventures, and Alumni Ventures participating. The company reports 5,000 business customers and 400% net dollar retention.
The founders, Keith Peiris and Henri Liriani, previously built Tome — an AI presentation tool that reached 25 million users and raised $80 million — and pivoted away from it last year.
The 400% number
Net dollar retention of 400% means the average customer cohort quadrupled its spend over the measurement period. Best-in-class SaaS lives at 120–140%. Hypergrowth infrastructure companies occasionally touch 200%.
Four hundred is a different kind of number, and it usually means one of three things: a small base where a handful of expansions dominate the ratio, usage-based pricing on a workload that is itself exploding, or land-and-expand from a deliberately tiny initial contract.
None of those are disqualifying — usage-based pricing on agent volume is a legitimately excellent business model when the volume compounds — but 400% NDR at Series A is a measurement of the pricing model at least as much as the product. The durable version of that number is the one that survives a base of five-figure starting contracts. Lightfield is not there yet.
The actual thesis
Lightfield's pitch: a system of record that structures customer interactions into a world model, so autonomous agents can reliably run relationships, prospecting, and performance analysis.
That phrasing is doing precise work. The bet is that the CRM's fundamental problem was never the interface — it was that the data was entered by humans who did not want to enter it.
Every CRM ever built rests on the same broken foundation: a salesperson finishes a call and is asked to translate a rich, ambiguous conversation into dropdown values and a text box. They do it late, partially, or not at all. Every downstream artifact — the pipeline forecast, the territory model, the QBR deck — inherits that decay. The entire category has spent twenty years building better dashboards on top of data nobody wanted to type.
Agents change the input side, not the output side. If the system observes calls, email, and messages directly and derives structure from them, the field stops being a chore and becomes an inference. And critically, an agent reading a derived world model can act on far more context than a human reading a record with four filled fields out of thirty.
This is a genuinely better idea than "CRM with a chatbot bolted on," which is what most incumbents shipped in 2025 and 2026.
What has to be true
Three things, and only the first is technical.
Inference from raw interaction has to beat human entry on accuracy, not just completeness. A CRM full of confidently wrong inferred fields is worse than a sparse one, because sparse records signal their own uncertainty and inferred ones do not. This is the failure mode to watch, and it will not show up in a demo.
Someone has to own the resulting system of record. CRM's moat was never features; it was that the data lived there and integrations pointed at it. If the world model is derived from interaction streams, it can in principle be re-derived by anyone with access to the same streams — which is a weaker lock-in than Salesforce spent two decades building. Lightfield's defensibility depends on being where the streams terminate.
The incumbents have to stay slow. Salesforce, HubSpot, and Microsoft all own the interaction data already, all ship agent products, and all have every incentive to derive structure from conversations rather than ask for it. The startup's advantage is architectural — no legacy schema to preserve, no installed base to avoid breaking — and architectural advantages have a shelf life.
The valuation footnote
Secondary-market analysis puts Lightfield's post-round valuation at roughly $121 million, with more than $100 million raised in total across all rounds including the Tome era.
A $47 million Series A into a ~$121 million valuation is a large round at a modest price — which is what a strong pivot from a known team usually looks like. The founders carry credibility from 25 million Tome users and the discipline of having shut it down, and investors are pricing the second act with the first act's cap table still attached.
The read
The AI-native CRM thesis is one of the cleanest applications of agents to enterprise software, because the target is a specific, universally acknowledged failure — humans do not fill in fields — rather than a vague promise of productivity.
The reason to watch this specific company rather than the category: a16z, Coatue, Greylock, and Lightspeed all wrote into a Series A at a valuation that leaves substantial room, which is what conviction looks like when it is not competing with a bidding war.
The reason to stay skeptical: 400% NDR at 5,000 customers is a pricing artifact until it is a retention fact, and the incumbents own the data streams Lightfield needs to terminate at its own system.