Tracelight CEO Pete Fuller says the startup will compete for the top 1 percent of Excel users rather than fight Microsoft and Anthropic for the billion general users.
Fuller says Anthropic's Claude for Excel launch was a huge moment that brought Tracelight much more attention, even though he did not expect the labs to ship Excel add-ins.
Fuller argues AI products have enormous overhang, so Tracelight is building memory to improve with use and staying model-independent as customers fear dependence on one lab.
Tracelight bets the top one percent of Excel users want their own AI
Tracelight is betting that the most demanding Excel users are a market the big labs and Microsoft will not serve well. It builds at the application layer, stays independent of any one model and adds memory so the product improves with use. Fuller says frontier models have not yet saturated the tasks these users care about, so there is still room for a focused product.
Takeaways
Tracelight targets Excel's top one percent of users. Fuller argues the most demanding Excel users need different tools from the other billion. Tracelight competes for them in management consulting and high finance, especially private market work like valuations and due diligence, rather than fighting Microsoft and Anthropic for everyone.
Big lab Excel add-ins helped rather than hurt. Fuller did not expect Anthropic and OpenAI to launch Excel add-ins, though he expected Microsoft to. Anthropic's Claude for Excel was a huge moment that brought Tracelight much more attention, and he is comfortable competing with Microsoft.
Model independence and orchestration are the strategy. Tracelight starts at the application layer. Customers are nervous about depending on one lab, so it uses Gemini to read PDFs and Claude to model in one workflow. Fuller says models may commoditize, with money made from complements.
Product overhang leaves room for application builders. Fuller says the harnesses built around models still feel inadequate, so there is enormous product overhang. Tracelight plans to build memory so the product improves with each use, and frontier models have not yet saturated its users' tasks.
Enterprise sales drive growth, and love drives retention. Enterprise revenue is the main growth engine. Deals are slow but now expanding quickly, mostly from inbound demand. Fuller says user love is the real stickiness, and buyers' biggest fear is choosing the wrong vendor.
Excel lacks the tooling coding already has. Fuller says coding was set up for AI before ChatGPT, with diffs, git and version control, while Excel has none of these. He says Microsoft needs significant changes. Barnhurst adds that Excel is 40 years old.
Modelers will build systems that become code. Fuller predicts financial modelling work will shift from building models to building systems that become code. He expects users to add more value this way and says he is very optimistic about their future.
Tracelight's path so far
Fuller worked in early stage venture at Entrepreneurs First, then five years at McKinsey with its AI practice QuantumBlack, and briefly at Amazon.
Tap a name to go there · tap a line to see why they are linked
On this episode 4
Show 1
He co-founded Tracelight in mid 2024 with Alex, a former Jane Street engineer, and Janek.
The team began with AI for Excel, inspired by a Microsoft research paper on representing spreadsheets to language models.
It later discarded its database and expanded into an all-in-one AI workspace with add-ins across Microsoft Office.
Next, growth comes more from the web app than new add-ins, with audit and tax as adjacent markets.