Recruitment Case Study: AI Cut Fill Time 45 to 19 Days

You Know That Feeling When a Role Drags On for 45 Days?

The hiring manager is asking for an update for the fifth time. You've screened dozens of resumes manually, played email tag with candidates, and coordinators keep double-booking. According to SHRM's 2024 report, the average time-to-fill hovers around 41 days—but for senior tech roles, it can easily exceed 45. We lived this nightmare with a 200-person SaaS company. Then we cut their time-to-fill to 19 days by combining AI tools with a few operational tweaks. Here's exactly how we did it, with templates you can use tomorrow.

The Challenge: A Tech Company Stuck at 45 Days

The company hired 15 engineers a year, but each search took 45 days on average. The recruiting team of three spent 60% of their week on non-strategic tasks: sourcing, screening, and scheduling. Offer acceptance rate was 78%, but candidates often ghosted because the process was too slow. We knew we couldn't throw more recruiters at the problem. We needed to rebuild the workflow around AI.

• Time-to-fill: 45 days
• Sourcing time per role: 13 hours
• Screening time per 100 applicants: 5.5 hours
• Outreach response rate: 11%
• Scheduling coordination: 8 emails per interview

The painful truth: their recruiters were spending 13 hours a week just sourcing. That was 34% of their workweek.

The AI Stack That Changed Everything

We selected a lean stack of AI tools that integrated without disrupting the existing ATS (Greenhouse). No rip-and-replace. We focused on four functions, each with a tool that specialized in one job. The key: we used AI to automate the high-effort, low-judgment work, not candidate evaluation. We combined [AI sourcing platforms](INTERNAL:tools/ai-sourcing) with [automated interview scheduling](INTERNAL:workflows/automated-scheduling) to create a seamless funnel.

1. AI Sourcing (HireEZ): to scan 50+ platforms and build ranked candidate lists in minutes.
2. AI Outreach & Personalization (Gem): to auto-draft contextual emails based on a candidate's GitHub and LinkedIn activity.
3. AI Scheduling (GoodTime): to eliminate back-and-forth by syncing calendars and offering auto-optimized time slots.
4. AI Screening Assistant (Ideal): to filter resumes against a predefined, structured intake form—not to reject, but to prioritize.

We didn't implement AI to reduce headcount. We did it to give recruiters back 20 hours a week so they could spend more time engaging top candidates.

Phase 1: Audit Your Current Funnel (2 Days)

You can't improve what you don't measure. We pulled exactly 5 baseline metrics from their ATS and self-reported recruiter time.

1. Pull last 10 closed requisitions and calculate median time-to-fill in days.
2. Have each recruiter track time spent on sourcing, screening, and scheduling for one week using a simple tally sheet.
3. Measure outreach response rate by dividing positive replies by total outreach emails sent.
4. Count the average number of emails exchanged between recruiter, candidate, and coordinator to schedule one interview.
5. Record offer acceptance rate for the past 6 months.

Phase 2: Build and Integrate Your AI Stack (1 Week)

Once we had the diagnoses, we chose tools that plugged the specific time sinks. We prioritized tools with native Greenhouse integrations to avoid data silos.

• Connect AI sourcing tool (HireEZ) to ATS and set up targeted job searches for each role type.
• Configure Gem sequences with AI-generated personalization tokens based on recent candidate activity.
• Set up GoodTime with recruiter and hiring manager calendars, enable "Instant Interview" for high-priority candidates.
• Create Ideal intake forms with 5 key "must-have" skills per role (co-created with hiring managers).

In my experience, the integration step is where most teams stumble. We spent a full day testing each tool with dummy data before we let it touch a live candidate. That upfront investment avoided embarrassing misfires later.

Phase 3: Pilot on One Role (2 Weeks)

We launched the stack on a single, high-volume role: Senior Full-Stack Engineer. The dedicated recruiter used the tools exclusively, while we shadowed and measured every step.

1. Day 1: Use AI sourcing to generate a list of 50 potential candidates from LinkedIn, GitHub, and Stack Overflow. Spend 30 minutes vetting the top 10 manually.
2. Day 2: Send personalized outreach using Gem's AI templates; manually review each message before sending (20 minutes total).
3. Day 3-7: GoodTime automatically schedules calls with candidates who reply; recruiter only steps in to handle declines or reschedules.
4. Day 8-12: Ideal screens incoming applications, ranking them green/yellow/red based on the intake form; recruiter reviews greens first and ignores reds (saves 4 hours/week).
5. After 2 weeks, compare time metrics to baseline.

After one week, the recruiter told me, 'I feel like I'm actually recruiting again instead of doing data entry.'

Phase 4: Scale and Lock in Improvements (Ongoing)

The pilot cut the senior full-stack role's time-to-fill from 44 to 21 days. We then rolled out the stack to all technical roles, with weekly calibration meetings. For deeper process improvements, check our [workflow automation guide](INTERNAL:workflows/ats-automation).

1. Create a standard operating procedure (SOP) document for each tool.
2. Hold weekly 15-minute stand-ups to review AI-sourced candidate quality and adjust search parameters.
3. Gradually give recruiters more autonomy to refine AI prompts as they see what works.
4. Use a dashboard (we built a simple Google Sheet with ATS data) to track ongoing time-to-fill and response rates.
5. After 90 days, sunset any manual sourcing or scheduling that the AI had made redundant.

The Results: Before vs. After

• Metric: Time to source 3 qualified candidates | Before (Manual): 6 days | After (AI-Enabled): 1 day
• Metric: Outreach response rate | Before (Manual): 11% | After (AI-Enabled): 23%
• Metric: Scheduling coordination per interview | Before (Manual): 8 emails | After (AI-Enabled): 0 emails (auto-booked)
• Metric: Resume screening per 100 applications | Before (Manual): 5.5 hours | After (AI-Enabled): 30 minutes
• Metric: Offer acceptance rate | Before (Manual): 78% | After (AI-Enabled): 88%
• Metric: Time-to-fill (median) | Before (Manual): 45 days | After (AI-Enabled): 19 days

Copy-Paste Scripts to Start Tomorrow

AI Sourcing Prompt: "Find me Senior React developers within 30 miles of Austin, TX who have contributed to any React-related open-source projects in the last 12 months, currently work at a company with under 500 employees, and have not held a role at a FAANG company in the last 2 years. Exclude candidates who are likely over 15 years of experience (judge by graduation year). Rank by recent GitHub activity."

AI-Powered Outreach Message: "Subject: quick question about {project_name} / Hi {first_name}, I noticed your work on {project}. We're building something similar at {company_name} and I'd love to get your take over a 15-minute call—no pitch, just curiosity. Are you open to a quick chat this week? / Best, {recruiter_name}"

AI Interview Scheduling Command: "Schedule a 30-minute video call with {candidate_name} and {hiring_manager_name} next week, avoiding Mondays and after 4pm. If no mutual slots, expand to the following week and offer 3 options. Copy me on the invite."

Limitations and Lessons Learned

No AI stack is perfect. Here's what we'd do differently and where you should tread carefully.

• AI screening tools still need human review: Ideal had a 5% false-positive rate where great candidates were marked red because their resumes didn't match keywords exactly. We built in a weekly manual audit of "red" candidates.
• Hiring manager availability remains a bottleneck: The AI scheduler can't force a busy VP to open their calendar. If time-to-fill plateaus, address stakeholder responsiveness first.
• Personalization can feel robotic if over-automated: We found that recruiters should spend 2-3 minutes tweaking each AI draft. The open rate spike came from combining AI efficiency with human nuance.
• Tools need constant tuning: What worked for React developers didn't work for DevOps roles. AI search strings and intake forms must be role-specific.

I believe the biggest insight was that AI forced us to get crystal clear on job requirements upfront. When you feed an AI vague criteria, you get noise. When you define exactly what 'must-have' means, the system hums.

Summary: You Can Cut Time-to-Fill This Quarter

You don't need a massive budget. This entire stack cost under $2,000/month in tool subscriptions and saved 60 hours of recruiter time per month. The key is to audit first, implement in phases, and never let a machine make the final call on a human. Ready to try? Start with Phase 1 tomorrow—measure your baseline and pick one role to pilot. The scripts above are your first step. If you want the full SOP template we used, subscribe to RecruitHacker for the download.