💥Are We in a Bubble?, Why Everyone’s Switching to AI Credits, Customer Support Benchmarks, Why SAFEs Blow Up Cap Tables & More
Digesting Insights From the Data
👋 Hi, I’m Andre and welcome to my newsletter Data Driven VC which is all about becoming a better investor with Data & AI. Join 35k+ thought leaders from VCs like a16z, Accel, Index, Sequoia, and more to understand how startup investing becomes more data-driven, why it matters, and what it means for you.
Brought to you by FORESIGHT - The End-to-End Data Platform for Investors
You had the best intentions when you bought all those data sources like Crunchbase, Dealroom, Pitchbook, and Harmonic. You thought everything gets logged with that powerful CRM called Affinity. You wanted to connect it all to ChatGPT and start asking questions.
But then you realized that you’re missing the middle layer between your data and your LLM that connects the data siloes, eliminates the duplicates, and normalizes the data swamp.
This is why you need an “entity resolution” data infrastructure, aka the glue. But do you build it or buy it?
Why Everyone Is Switching to AI Credits
Kyle Poyar reviews the rapid rise of credit-based pricing models in AI. He highlights how incumbents like Microsoft, Salesforce, and OpenAI are normalizing credits as the default way to buy AI, why cost pressures make this shift necessary, and what practices can make credits work for vendors and customers.
Adoption by Big Players: Microsoft added credits for Copilot in January, Salesforce launched flexible credits in May, Cursor moved in June, and OpenAI replaced seat licenses with pooled credits in its Enterprise plan. Dozens of others, including Adobe, HubSpot, Google, and Replit, have followed.
Usage Patterns and Margins: Token usage is exploding, with 70–80% of consumption coming from just 10% of users. That concentration can make heavy users unprofitable, especially since AI gross margins remain tight. Credit pools help vendors contain costs while giving customers clarity.
Credits in Practice: Definitions vary widely. Prices range from fractions of a cent at Salesforce to $0.25 at Lovable. Some companies link credits to costs (e.g., Cursor), others to outcomes like resolved cases or completed workflows (e.g., Salesforce, HubSpot). The latter makes ROI clearer for buyers.
✈️ KEY TAKEAWAYS
Credit-based pricing is becoming the standard for AI, balancing vendor margins with customer predictability. While inconsistent definitions remain a challenge, credits serve as a bridge from flat-rate pricing toward value- and outcome-based models.
Benchmarks for Customer Support Rep Team Size Ratios
Recent analysis by Matt Schulmann highlights how customer support staffing ratios scale with company size, and how AI adoption may shift these benchmarks further. A Pave dataset covering 2,237 companies with at least one support rep provides concrete numbers on full-time employees (FTEs) per support representative.
Support Ratios by Company Size: Smaller companies (1–99 employees) average 23 FTEs per support rep. Mid-sized firms scale up to 50 FTEs per rep at 500–999 employees, while the largest (3,000+) reach 65 FTEs per rep.
Scaling Patterns: Efficiency generally improves with company size, though firms in the 1,000–2,999 range show a dip at 43 FTEs per rep. This suggests structural differences in support operations at that stage of growth.
AI Impact Predictions: With Conversational AI agents like Decagon and Salesforce’s recent 4,000-person reduction, leverage per rep is expected to rise. Benchmarks will likely increase as AI tools automate more of the support workload.
✈️ KEY TAKEAWAYS
Customer support ratios improve with scale, and early signs show AI adoption will accelerate this trend. Companies should anticipate higher leverage per rep as automation reshapes staffing benchmarks.
Oh, the Irony: SAFEs Are Blowing Up Cap Tables
CJ Gustafson explains how repeated use of SAFEs is creating hidden dilution that founders often underestimate. While SAFEs were designed as a quick way to raise early capital, stacking them across multiple rounds delays dilution on paper but compounds it heavily once priced rounds arrive.
Dilution by Stage: Carta data shows that a typical pre-seed SAFE round equals about 10% dilution. Seed and Series A rounds usually add another 20% each, leaving founders with only 38–39% ownership post-Series A once option pools are factored in.
The SAFE Problem: SAFEs delay conversion into equity, creating opacity in cap tables. Some founders layer three or more SAFEs at different valuation caps, which can add 20% or more hidden dilution by the time they price a round. In extreme cases, Carta has tracked companies with up to 10 SAFEs before conversion.
Post-Series A Reality: After dilution and option pools, founders may control less than half of their company before raising a Series B. Two equal co-founders can each fall below 20% ownership, a stark contrast to the intended founder-friendly reputation of SAFEs.
✈️ KEY TAKEAWAYS
SAFEs can be useful at pre-seed, but stacking them across rounds risks severe hidden dilution. Founders should treat priced rounds as a necessary step to maintain transparency and control over their cap tables.
How Many Rounds Does It Take to Build a Unicorn?
Stanford professor Ilya Strebulaev analyzed 1,955 US unicorns and found that fundraising journeys vary widely. While some reach billion-dollar status quickly, most go through multiple rounds before hitting the milestone.
Distribution of Rounds: Nearly 200 unicorns needed fewer than 2 rounds. Around 600 reached unicorn status after 3–4 rounds, the most common path. More than half required 5+ rounds, with 331 raising 8 or more.
Outliers and Extremes: The maximum recorded was 18 rounds before unicorn status. Companies achieving unicorn valuations through private rounds typically needed fewer rounds than those doing so via IPOs or acquisitions.
Dataset Scope: The sample covers VC-backed US companies that became unicorns between January 1997 and July 2024. Some fundraising data may be incomplete, but the trends highlight persistence as a key factor.
✈️ KEY TAKEAWAYS
Unicorn paths differ, but the median experience shows 3–5 rounds are most common. Founders should expect multiple raises and recognize that persistence and timing matter as much as capital efficiency.
Are Startups Healthy or in a Bubble? Both.
Peter Walker’s latest analysis of Series A to Series B “graduation rates” shows how startup health has shifted over the past seven years. The metric reflects what percentage of startups raising Series A go on to close a Series B within two years.
Graduation Rate Trends: 2018 cohorts saw 27% reach Series B within 8 quarters, while 2019 hit 33%. The bubble years spiked higher, with 2020 at 43% and 2021 at 35%. Post-bubble cohorts dropped sharply, with only 13–15% of 2022–2023 companies advancing.
Recent Shifts: Q1 2024 companies are at 13% through five quarters, pacing slightly better but still far below bubble-era levels. This suggests the market is stabilizing after a steep correction.
AI’s Unique Bubble: Unlike the 2021 boom where capital lifted most startups, today’s AI funding wave is highly concentrated. Massive sums flow into a handful of companies, leaving graduation rates far lower than past bubbles. The risk is that when such a narrow bubble pops, the fallout could be sharper.
✈️ KEY TAKEAWAYS
Startup health metrics point to a market that has corrected from excess but remains uneven. AI is fueling a bubble, yet concentrated in fewer players, creating a different risk profile than broad-based exuberance in 2021.
H1 2025 in Rearview
AngelList’s latest data shows stabilization in the venture market, but the environment for raising and scaling is shifting in ways founders should prepare for. AI and robotics still dominate investor attention, valuations are steady at the earliest stages, and funds are cautious with capital deployment after three years of flat performance.
Where Capital Is Flowing: AI and robotics now represent nearly 60% of all deployed capital. Founders outside these sectors should expect tighter competition for dollars and need sharper differentiation, while those in AI/robotics face a favorable but crowded landscape.
Valuations and Round Dynamics: Pre-seed and seed valuations held at $10M and $20M, so early-stage founders can plan around flat pricing when raising. Later-stage rounds, especially Series B, are getting more expensive, which means hitting growth milestones early is critical to justify steeper valuations.
Fund Behavior and LP Pressure: Most funds are marking time with little appreciation since 2021. This means investors are under pressure to show discipline. Founders should expect tougher diligence, more structured rounds, and less tolerance for inflated pricing unless strong traction is proven.
✈️ KEY TAKEAWAYS
Founders should expect flat early-stage valuations, a competitive AI/robotics funding environment, and heightened investor scrutiny due to stagnant fund returns. Planning for disciplined milestones, transparent metrics, and sector positioning is key to navigating H2 2025.
Thanks to Jérôme Jaggi for his help with this post.
Stay driven,
Andre
PS: Check out Foresight - the solution I would’ve loved to see in the world before we spent 7-figures to build our own infra. Now “the glue” for your messy data stack is available off-the-shelf ;)
Thank you for reading this episode. If you enjoyed it, leave a like or comment, and share it with your friends. If you don’t like it, you can always update your preferences to reduce the frequency and focus on your preferred DDVC formats. Join our community below and follow me on LinkedIn or Twitter to never miss data-driven VC updates again.