Jev case study: Jack & Jill cut candidate screening costs 88%

On 12 October 2026, TypeSafe AI posted a customer case study on its blog titled "How Jack & Jill Cut Candidate Screening Costs By 88% With Jev". TypeSafe files it under Case Study.

Jack & Jill is a hiring marketplace run by AI agents. According to the post, Jack is a career agent for professionals and Jill is an agent for hiring teams; when they find a mutual match, they introduce the candidate and the company directly, with no application. TypeSafe says the company works with over 5,000 companies and more than 400,000 professionals.

At one stage of its pipeline, Jack & Jill scores candidates against a role. Before Jev, that stage called Gemini 3.1 Flash Lite. TypeSafe says Jack & Jill went from its first test to using Jev for 100% of calls at that stage within 10 days.

The stage is a good example of the kind of work Jev is built for: a defined question (how well does this candidate fit this role?) answered with a score, many thousands of times a day. TypeSafe says the volume is about 6,000 candidates per open role per day, with a shortlist of the top 2,000.

All of the figures below are TypeSafe's own, taken from the case study:

• Cost: $0.755 with Gemini against $0.092 with Jev per 1,000 candidates scored, which is 88% lower.
• Savings: about $265,000 a year at current volume, and a projected $500,000 over the next twelve months if the business grows as expected. The second number is a projection, not a result.
• Quality: the right candidates stayed in the top-2,000 shortlist 94.6% of the time with Jev against 93.9% with Gemini. TypeSafe itself says the 0.7-point gap is not statistically significant. The AUC (a standard measure of how well a score separates good matches from poor ones) was 0.933 against 0.924.
• Speed: a median of 10.3 seconds with Jev against 20.3 seconds with Gemini, roughly twice as fast.
• Test size: 150 live roles.
• Elsewhere in the product: TypeSafe says Jack & Jill now uses Jev in more than 15 workflows, and that one profile check dropped from about 3.6 seconds to 0.2 seconds.

This is a vendor's own case study, so a few points deserve caution:

• It is TypeSafe's account. The post does not say that anyone outside TypeSafe checked the figures, and Jack & Jill's own version of the story was not part of what we read.
• The quality results come from an offline test, as the post's own method note says, not from watching hiring outcomes.
• The speed figure is described inconsistently. One table labels it per candidate and another per role, so treat the 2x as approximate.
• The price comparison uses list prices at the time of measurement: Jev at $0.042 per million input tokens with free output, and Gemini at $0.25 for input and $1.50 for output. Prices change, so redo the sum with current numbers before relying on the 88%.

The result fits what Jev is designed for: high-volume scoring and classification, where an LLM's slower, pricier answers are not needed. If you have a similar step in your own product, the quickest check is to run the same questions through both models and compare the cost and the quality on your own data, as Jack & Jill did on 150 roles.

Our cost calculator lets you put your own volume into the Jev price, and the use cases pages show the kinds of questions this class of model handles well.

• Cost calculator
• Use cases