Juan Herrera

data @tryramp

[Juan Herrera @JuanHerr19]

One of the hardest parts of causal inference, especially if you have an ML background, is letting go of treating everything as a prediction problem.

Take Propensity Score Estimation as an example. We want to uncover the impact of taking a course (T) on Salary (Y)

Took Course (T) -> Salary (Y). 1/n

TL;DR: Include only confounding variables to estimate propensity scores, as including other variables may result in less precise treatment effects.

[Teddy Riker @teddy_riker]

We just published the top use cases of our most AI-native customers who use our CLI and MCP to our website.

Every playbook contains a custom prompt tailored to whatever agent runtime you use: Claude, ChatGPT, Codex, and more.

Get started in one prompt👇

[Ian Macomber @iandmacomber]

A few data points I’m seeing based on the spending patterns of 70k+ @tryramp customers, coding patterns of 1500+ Ramp employees, and my own N=1 anecdote over the last few months.

Half of Ramp customers haven’t spent on @OpenAI or @AnthropicAI yet; the majority are not buying tokens via API, and probably won’t next quarter. But the AI bill is coming due for everyone: over the last 12 months, we’ve seen the percentage of “SaaS vendors with AI-usage billing SKUs” double. Companies using the software they already know (Github, Notion, Zendesk, Hubspot, Salesforce, Adobe) are getting familiar looking invoices with unfamiliar new cost models: AI credits, agent runs, model tiers, usage caps, overages, and of course, tokens.

This is scary! My N=1 anecdote: I got an email from one of our data vendors, introducing an AI credit model on top of our PEPM (per-user-per-month) pricing. This wasn’t in our budget, I didn’t know what a credit represented, I didn’t know if this would increase our bill by 0% or 1% or 100%. It took me about 30 minutes to stitch together everything across the website + admin panel.

I literally have better data than anyone in the world on what SaaS vendors cost, and this was still painful for me. This transition + feeling of uncertainty will happen to millions of SaaS budget owners, across thousands of SaaS vendors.

But it’s just as challenging for the SaaS vendors! The end of token subsidies from OpenAI + Anthropic is over. The SaaS AI features + workloads that customers love are starting to get really expensive, in a way that seat pricing can’t recover. We’re seeing vendors look for opportunities to shift workloads to non-frontier models, to other providers (@deepseek_ai was a trending vendor on Ramp this month for a reason!).

Ramp has always been about turning spend into intelligence. Allocating compute and frontier intelligence IS the story of the next five years of business. What do companies like us spend? Which vendors are normal for our stage and industry? Which teams are outliers? Which AI charges are new, which are growing, and which are tied to outcomes worth funding?

Today, Ramp raised $750M at a $44B valuation. The job has never been less finished. Our data will be there every step of the way to help companies save time, save money, allocate intelligence, and navigate the post-AI era.

[Juan Herrera @JuanHerr19]

@zeeg Hey David, totally unrelated but curious about what kind of analytics/metrics you run to track tool performance for the MCP

[Ramp @tryramp]

"We can let Ramp be the bad cop. That changes the relationship between finance and the rest of the company." - @ProsperLoans' director of enterprise risk management

Bad cop reporting for duty o7

[Juan Herrera @JuanHerr19]

Get started here

https://t.co/QB2wwzWkCT

[Juan Herrera @JuanHerr19]

@tryramp daily MCP/CLI adoption in the last few months.

A few things people are building with it:
- CFO dashboards, automatically refreshed
- accounting codes, missing receipts, memos on a schedule
- procurement workflow automation

[Juan Herrera @JuanHerr19]

@kushbhuwalka @agentmail @paysponge How much does it cost?

[Juan Herrera @JuanHerr19]

Get access here: https://t.co/QB2wwzWkCT

[Teddy Riker @teddy_riker]

50+% of Ramp MCP and CLI users are building recurring reports to summarize things that need their attention on a given day

We now have a single tool that will aggregate:

- transactions, reimbursements, bills pending approval
- expenses missing receipts, memos, and accounting codes
- procurement tasks
- upcoming contract renewals
- payment runs
- vendor updates

and other Ramp homepage action items into one prioritized feed.

Just ask "show me Ramp tasks requiring my attention"

(note: if you are using Claude or ChatGPT, you will need to reauthorize the Ramp connector to get access)

[Juan Herrera @JuanHerr19]

@akothari @NotionDevs Is this daily tool calls?

[Juan Herrera @JuanHerr19]

@yul_magdaleno Que vivan los boleros románticos!

[Zack @zack_field]

Ramp Agent Cards is now self-serve!

> ramp funds enroll

ask your agent the next time you need to make a purchase!

[Juan Herrera @JuanHerr19]

@zeeg Makes sense. What’s the trade off with the tools added to the MCP? Are there any stuff you rather have the user use the app for?

[Ramp @tryramp]

Ramp Data is now LIVE in Claude, ChatGPT, Bloomberg, Perplexity, and Grok. Ask what 50,000+ businesses are paying for software or where adoption is shifting and you’ll get answers based on real spend data.

It's free, public, aggregated, and anonymized. Enjoy, data lovers.

[Juan Herrera @JuanHerr19]

@_renatov @TrevMcKendrick Have you tried it? Whats your experience?

[Juan Herrera @JuanHerr19]

Go give this a read

[Teddy Riker @teddy_riker]

https://t.co/dyiVwtFQNJ

[Juan Herrera @JuanHerr19]

@sama @_dmca What’s the breakdown between new and recurring users?

[Juan Herrera @JuanHerr19]

@SolveBuild @RhysSullivan @mattzcarey Are the prompts mostly workflows you want the client to pick up more easily?

[Juan Herrera @JuanHerr19]

@mehtadata_ @Benioff Did you add the workflows as resources?

[Juan Herrera @JuanHerr19]

@SolveBuild @RhysSullivan @mattzcarey How does this work so far?