Ignite Reading's tutor capacity model went from a week of manual work to a handful of prompts — and surfaced a hiring ratio no one had measured before.

The Challenge

Joy Stanton manages people operations for Ignite Reading, an organization that deploys approximately 1,200 part-time tutors across programs serving students in under-resourced schools. Tutors don't arrive in a steady stream — they're hired in cohorts, typically 200 to 500 at a time, tied to specific program launches and seasonal cycles. That means workforce planning isn't a static exercise. It requires understanding how each cohort behaves over time: when tutors leave, at what rate, and how that compares to previous years.

Joy’s team had been tasked with building a tutor capacity model, a framework the organization could use to forecast hiring needs, set realistic targets, and plan for program growth. The outputs were clear: resignation rates, attrition patterns by cohort, year-over-year trends, and the relationship between offer letters sent and tutors who actually started. What wasn't clear was how to surface that data in any practical way.

The data existed within Rippling, but accessing it required knowing the right queries, filters, and time windows — and then stitching the outputs together across multiple pulls. For someone without a data analysis background, the process would have taken a week of work. Even then, the model would have been built on assumptions that hadn't been validated. As Joy later discovered, some of those assumptions were wrong.

The Solution

Joy used Rippling AI to iteratively build the model, working through a chain of prompts over several days. Each prompt answered a specific question, and each answer opened the next one. The first prompt established a baseline:

"What's our part-time employee resignation rate over the last 60 days? Break it down by month, week, and day."

From there, she drilled into cohort-level attrition for the current winter program, then pulled 90-day attrition data with a year-over-year comparison. A subsequent prompt requested full 2025 calendar-year attrition, alongside offer letters sent versus actual starts — a two-year view of the gap between people who accepted an offer and those who showed up. The final prompt synthesized everything: overall attrition since January 1, 2024, with the full picture assembled in one place.

What Rippling AI returned was a complete lifecycle view of how Ignite Reading's tutor workforce actually behaves — the kind of model that would have required a week of manual data work to approximate, and that had never been built with real numbers before. One finding stood out immediately: on average, only 63% of tutors who received an offer letter actually started. For an organization planning cohort sizes in the hundreds, that ratio had significant implications for every hiring target they had ever set.

Rippling AI also did something Joy hadn't anticipated: it flagged its own assumptions and surfaced a data anomaly without being asked. When pulling cohort-level data, the system noted that it had defined a cohort as 50 or more part-time hourly employees and flagged one cohort showing zero actual starts. Rather than silently skipping the anomaly, Rippling AI explained it: those start dates had shifted, and the employees had been moved into a different cohort. It was exactly the kind of detail that corrupts a model built on raw exports, and it surfaced automatically.

The whole process, from the first prompt to the complete model, took roughly an hour. Joy's estimate for doing it manually: a full week, with no guarantee that the underlying assumptions would be correct.

The Impact

The capacity model Joy built is already in use. Ignite Reading has adjusted its approach to summer school pilot hiring based on the offer-to-start ratio the model surfaced — a direct change to operational planning that flows from data that previously didn't exist in a usable form. The same model will inform how the organization plans for fall program launches, giving leadership a foundation for forecasting that replaces guesswork with a two-year view of actual tutor behavior.

Beyond the numbers themselves, the model changed what Joy could bring to cross-functional conversations. When she presented the findings at a meeting that included team members outside of people operations, the response was immediate.

A week of work, done in an hour. Building a tutor lifecycle model from scratch would have taken a full week manually. Rippling AI completed the same analysis iteratively in roughly an hour.

Wrong assumptions, corrected. The 63% offer-to-start rate — the ratio of accepted offers to tutors who actually showed up — was a finding that hadn't been quantified before. It corrected assumptions baked into hiring targets and changed how Ignite Reading approached summer pilot planning.

Anomalies surfaced automatically. Rippling AI flagged a cohort showing zero actual starts and explained why, without being asked. 

Cross-functional visibility, on demand. People operations data that had previously been inaccessible to other teams was now available in real time and legible to anyone in the room. The capacity model became a shared planning tool the moment it was presented.

For a lean people operations team managing a workforce that cycles through seasonal cohorts, the ability to answer complex workforce questions in real time is a meaningful shift in what's operationally possible. For Joy, it's also a shift in what she can do with her own time: less pulling data, more thinking about what the data means.