The $30,000-a-Year Calendar Assistant Nobody Asked For
Here's a number that should make every CFO wince: thirty thousand dollars per year. That's what one Rippling employee was burning through on Anthropic's Claude API just to have an AI sort their calendar and draft email replies. No malice involved. The worker genuinely found it helpful — "Claude is so helpful for me," they said, according to CEO Parker Conrad. But the return on that spend? Barely there.
This is the new problem AI has created for enterprise software. For decades, companies paid per seat. You bought fifty licenses for your CRM, fifty for your project management tool, and you knew exactly what your bill would look like at month's end. Predictable. Boring, maybe, but safe.
Generative AI broke that model. Now you're paying by the token — by the actual computation consumed every time someone asks a question, generates a summary, or drafts an email. And unlike a Slack subscription that costs $7.50 per user regardless of how much you type, an LLM API bill scales with enthusiasm. The more helpful the tool feels, the faster it drains your budget.
Rippling found this out the hard way when they turned their own Data Cloud on their internal workforce. What emerged wasn't a story about bad actors or policy violations. It was something more interesting and, honestly, more common: well-meaning employees discovering that AI could do their administrative drudgery and then, quite naturally, letting it do all of their administrative drudgery — repeatedly, expensively, and without anyone noticing until the invoices arrived.
Merging Your Org Chart with Your API Logs
The insight that made Rippling's Data Cloud worth building was deceptively simple. Most companies have two completely separate data worlds: their HRIS system, which knows who reports to whom and how they're performing, and their application logs, which know how much API compute each person is consuming. Nobody connects them.
So you've got your engineering manager looking at GitHub pull requests, trying to guess which developers are getting value from AI tools and which are just generating noise. Meanwhile, Anthropic's usage dashboard shows raw token consumption with zero context about whether that consumption produced anything worth shipping.
Rippling's Data Cloud collapses this gap. It pulls application audit logs — Salesforce ticket volumes, GitHub PR data, Claude and GPT API consumption records — and stitches them directly into the HR workforce platform. The result is a system where you can ask questions like "Which teams are drowning in unresolved tickets relative to their headcount?" and get an answer that accounts for both workload and staffing levels.
Conrad demonstrated this live during the June 25 launch. He pulled up a dashboard cross-referencing Salesforce support ticket volume with employee scheduling data and immediately spotted that Rippling's enrollments team was severely understaffed while the travel team carried more than double the unresolved tickets of the platform team. That's not just a usage metric — that's an org chart insight powered by application data.
The platform handles permissions automatically. When someone changes roles, updates their status, or reports to a new manager, the Data Cloud adjusts data access accordingly. You're not building a custom ETL pipeline for every new SaaS integration. Fivetran, Snowflake, and Tableau do pieces of this puzzle individually — Rippling is arguing they should all live inside the HR system instead.
Code Slop: When AI Help Becomes a Productivity Tax
The most compelling use case Conrad showcased wasn't about email or scheduling. It was about engineering productivity — and specifically, the phenomenon Rippling internally calls "slop."
Here's how the analysis works. Rippling cross-references three data sources: Anthropic's LLM consumption logs, GitHub pull request history, and the company's own performance review ratings. The pattern that emerged was intuitive once you saw it. High performers spend the most on AI — which makes sense, since they're writing more code and asking more questions. But the dashboard also surfaced a less flattering group: engineers with high AI token spend and high peer rejection rates on code reviews.
"If your peers are telling you to go back and do this over all the time, maybe you're just generating a lot of slop," Conrad said.
This is the distinction that matters. There's productive AI utilization — an experienced developer using Claude to accelerate a complex refactor, producing clean code that ships on the first review. And then there's unproductive utilization — a developer generating high volumes of AI-assisted code that colleagues keep asking them to redo. The token spend is identical. The business value isn't even in the same neighborhood.
Rippling has already acted on these findings, cutting spending limits for certain employees based on the data. But the product goes further than just throttling access. You can configure automated alerts that notify managers when someone exceeds a threshold, or set up automatic shutdowns that cut off API and UI access entirely once spending crosses a line you've drawn.
The implications for engineering leadership are significant. For the first time, you can distinguish between an engineer who's leveraging AI as a force multiplier and one who's using it as a crutch — not through subjective judgment, but through the hard data of peer review outcomes and consumption patterns.
The Banking Play: HR Platforms vs. Fintech Spend Tools
While the Data Cloud grabs headlines for AI governance, Rippling made an equally strategic move this week with Business Banking — a high-yield checking account and same-day payroll processing feature that directly challenges fintech spend management players.
The pitch is familiar: eliminate the mental overhead of managing two timelines. Traditional payroll systems require you to process payments two to four days in advance. Rippling's banking product lets companies run payroll on the same day employees get paid, with changes accepted as late as 1 p.m. on payday.
But the real competition is with Ramp, which just raised $750 million at a $44 billion valuation — roughly three times what Rippling's investors valued the company at last year ($16.8 billion). Ramp has been positioning itself as the financial operating system for companies navigating AI costs, and Conrad welcomes the comparison even though Rippling's banking business remains far smaller.
"There are some advantages to centralizing all of this," Conrad noted. The argument is that if you're already tracking AI spend, employee performance, and application usage inside your HR platform, why maintain a separate fintech stack to manage corporate cards and expense reports? The data synergy is real, even if Rippling's banking revenue is still in its early days.
The base Rippling AI SKU runs around $20 per month, with usage-based charges kicking in for heavier consumers. About 560 companies are currently using the Data Cloud, generating roughly $5 million to $7 million in new monthly revenue. Not bad for a product that's barely a week old.
Practical Controls for Teams Managing AI Spend
If you're reading this and your organization is currently flying blind on generative AI costs, here's what the Rippling example teaches you — and what tools like Zylo and Torii can help you implement today.
Set hard spending limits with automatic enforcement. The goal isn't to prevent employees from using AI tools. It's to prevent runaway consumption without meaningful output. Configure per-user or per-team spend caps, and set up automated alerts — or access cutoffs — when thresholds are breached. Rippling's own experience shows that without these guardrails, helpful AI tools will quietly consume budgets faster than anyone notices.
Cross-reference consumption with output metrics. Raw token spend tells you nothing about value. You need to connect API logs to actual work product: code review acceptance rates, support ticket resolution times, content publication velocity. High spend combined with low output quality is the slop pattern — and it's identifiable once you have both data streams in one place.
Evaluate model economics regularly. Rippling recently migrated much of its backend AI stack from Anthropic to OpenAI's GPT-5.5 model, citing improved performance and better cost-efficiency. The model landscape shifts constantly. What was the right choice six months ago may not be today. Build regular model-evaluation cycles into your AI governance process.
Catalog every AI tool in your stack. Shadow AI — employees using unauthorized generative tools — is a growing risk. Platforms like Zylo specialize in discovering and tracking usage-based software licenses across enterprise portfolios, while Torii offers a dedicated AI governance dashboard that helps IT and finance teams inspect data compliance risks and track consumption costs across OpenAI, Anthropic, and other provider APIs. You can't manage what you haven't found.
Centralize where possible. The broader lesson from Rippling's Data Cloud launch is that the most powerful insights come from connecting data domains that usually live in isolation. Your HRIS knows who's performing well. Your application logs know what tools they're using. Your finance system knows what it costs. The companies that win on AI ROI will be the ones who stop treating these as separate problems and start treating them as one.
The era of per-seat software is over. The era of consumption-based AI spend is here — and it demands a level of visibility that most organizations simply haven't built yet. The tools exist now. The question is whether you'll use them before the bills become impossible to ignore.
