Vanta absorbs token costs rather than expose markup to customers
Users spot token markups instantly, so Vanta eats costs, scales agent adoption, and defers pricing.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
SaaS valuations halved despite stronger margins; investors misprice growth profiles
Vanta combines historically high margins and rapidly improving growth, yet valuations fell to one-third of peaks. The gap: low-margin AI companies command ten times higher multiples. Christina calls this cognitive dissonance. Investors only fund extremes—lab experiments and early-stage ventures—and avoid mid-stage stable-growth SaaS. Meanwhile, top talent gets poached by higher-valued competitors.
— Christina CacioppoToken charges fail when customers notice; Vanta absorbs costs and scales adoption
Vanta's agents act autonomously—proposing actions like drafting access-revocation scripts and asking for approval. This consumes enormous token volumes. When Vanta tested token-based billing, customers instantly spotted markups and began comparing raw model performance, forcing Vanta into direct competition with Claude. Rather than mark up tokens, Vanta absorbs the cost, accepts lower margins, and focuses on scaling agent-adoption from 25 percent to 75 percent through better user retention and experience. Pricing strategy comes later.
— Christina CacioppoOutcome pricing: everyone wants it, nobody can define what counts
Christina, an economist, notes everyone wants outcome-based pricing—where incentives align. The problem: what's an outcome? Closing a support ticket is output; real outcome is solving the customer's problem. Even retreating to ticket closure, forty-plus measurable units exist: draft documents, close alerts, revoke accounts, etc. Users can't calculate what each agent action should cost. Result: every company wants outcome pricing but nobody can implement it, so they use point-to-token ratios that smart users see through.
— Christina CacioppoAgents need modular codebases; eight-year-old systems need architectural rewrites
Agents work better on modular codebases—it's a token-context problem. Vanta's eight-year-old codebase required a year of modularization so agents could understand whole modules instead of drowning in legacy code. Otherwise an expert needs to monitor the agent constantly, defeating the purpose. Christina frames this as an old problem amplified: new engineers used to need ramp time; now Vanta wants 9,000 instant agent engineers. She grades herself low on this work.
— Christina CacioppoTeams ship proven patterns over risky LLM, unless explicitly forbidden
When Vanta redesigned customer onboarding, the team kept proposing a four-step progress wizard despite feedback to use LLM. The blocker: they felt "not allowed to fail" on this project. The wizard was proven; the LLM approach risked failure. Once leadership explicitly said "this must succeed," the team shipped an LLM-based solution the next day—and it worked well. Christina didn't anticipate this psychological blocker.
— Christina CacioppoTop talent compensation jumped tenfold in five years; still undervalued
On talent wars, Christina notes that compensation bands are a joke—the real band is whatever competitors offer. For the same role, total compensation has grown roughly tenfold compared to five years ago, and she thinks it should be higher. Her philosophy: you never pay top talent enough. She describes their qualities—high agency, execution, deep business understanding, sound judgment, iteration—as worth millions, yet they sound like clichés. The only force keeping salary growth in check is investor worry about equity dilution, but that concern is nearly gone now.
— Christina CacioppoFrench companies chose Chinese models to hedge against supply cuts
During recent Paris customer visits, Christina found that many French software companies now primarily use Chinese AI models. She initially guessed nationalism, but the real reason: after Trump blocked certain models for non-US users, companies realized their business continuity risk. No guarantee US models stay available. So they switched to Chinese models wholesale. Not values—risk management.
— Christina CacioppoLosing the billboard to a cash-rich startup signals market shift
Vanta's iconic billboard at the 101/280 interchange, famous for "compliance that doesn't suck too much," got stolen during renewal by a competitor—Vanta only had a placement order, no long-term contract. Christina offered double, then triple the price, but the competitor said "we have positive cash flow" and wouldn't sell. Vanta lost the location. She joked it felt like watching market capitalization decline in real time.
— Christina CacioppoIn their own words · checked verbatim
You're really only as good as the last thing you shipped.
Christina Cacioppo3:08
Are you Forex as good as Claude? I don't know. Let me go try it out.
Christina Cacioppo9:18
We're eating the token cost right now. We're just taking lower margin.
Christina Cacioppo10:20
I think everyone wants to get to outcome pricing. Everyone wants to get to that.
Christina Cacioppo12:22
We basically don't think we're allowed to fail on this project.
Christina Cacioppo19:40
My comp philosophy is you never pay the top people enough.
Christina Cacioppo23:46
We basically all switched to Chinese models when Trump closed Fable off to not Americans.
Christina Cacioppo33:01
And he's like, we're actually cash flow positive. So like literally, I'm an inelastic seller. And so we lost the spot.
Christina Cacioppo52:40
Figures
| Vanta customer count | ~16,000 | 3:08 |
| Tool integrations supported | 400 | 4:10 |
| Agent-style workflow adoption (current to target) | 25%→75% | 10:20 |
| Top talent compensation growth over five years | ~10x | 23:46 |
| Vanta valuation target (in jest) | $250 billion | 54:42 |
Glossary
- FDE
- Frontline deployment engineer—repackaged as prestigious role, essentially professional services.
- harness
- Model execution framework; the engineering layer connecting agents to real tools and systems.
- outcome pricing
- Billing by business results achieved, not by operations performed or tokens used.
- trust graph
- Vanta's security data layer built from integrating hundreds of software tools.
How to listen
B2B SaaS founders working on AI feature pricing; investors interested in talent compensation and valuation during AI transitions.
Skip her discussion of speaking technique and stage presence (41-42 minutes); low information density.