Skip to main content

AI in Recruitment: From CV to Offer in 48 Hours Instead of 2 Weeks (Case Study)

2025-10-23

Recruitment in 2025: Candidates Wait 3 Days Max. Your Process Takes 14 Days. You Are Losing the Best Talent.

The recruitment problem today is not "there are no candidates." The problem is: your process is too slow. A top candidate receives 3 offers in a week. If you need 2 weeks to go from CV to offer, they have already signed elsewhere. Below is a case study of how an IT company (25 people) cut their process from 14 days to 48 hours using AI. Without sacrificing quality. In fact -- quality went UP.

Before: The Typical 14-Day Process (That Lost 60% of Candidates)

Timeline before AI:

Days 1-3: CV Screening

  • 100 applications come in
  • HR spends 5 minutes per CV = 500 minutes = 8.3 hours
  • Selects 20 "maybes"
  • Problem: Subjective, inconsistent criteria. "I like this one" vs "I don't"

Days 4-7: First Screen Call

  • HR calls 20 candidates (30 min each) = 10 hours
  • Handwritten notes, scattered across email/Slack
  • Selects 10 to "pass to hiring manager"
  • Problem: 50% of time spent "qualifying out." Wasting time on people who don't fit

Days 8-11: Technical Interview

  • Hiring manager + senior dev talk to 10 candidates (1h each) = 20h (2 people)
  • Selects 3 "finalists"
  • Problem: Scheduling nightmare. Candidate available Tuesday, senior dev Thursday, hiring manager Friday. Coordination takes 2-3 days

Days 12-14: Decision + Offer

  • Internal deliberation
  • Salary negotiation
  • Offer letter
  • Problem: By day 14, 60% of top candidates have already signed elsewhere

Metrics before:

Metric Value
Time to Offer (average) 14 days
Offer Acceptance Rate 45% (55% reject because already signed elsewhere)
Candidate Experience Score (1-5) 2.9 ("too slow", "no feedback")
HR Time / Hire 28h
Quality of Hire (6-month retention) 72%

After: The 48-Hour AI-Powered Process (5 Steps)

Step 1: AI CV Screening (10 Minutes Instead of 8 Hours)

Stack: Lever ATS + OpenAI API custom integration.

How it works:

  1. CV arrives in Lever
  2. Webhook triggers AI analysis against defined criteria (required skills, years of experience, red flags)
  3. Scoring 0-100 + 1-paragraph summary
  4. Auto-sort: >80 = "Interview", 60-80 = "Maybe", <60 = "No"

Criteria (example for Senior Developer):

  • Must-have: 5+ years experience in [tech stack], portfolio/GitHub
  • Nice-to-have: Open source contributions, previous startup experience
  • Red flags: Job hopping (<1 year per job), unexplained gaps, no technical projects

Sample output per CV:

Score: 87/100
Summary: Strong technical background (7 years React/Node.js). Active GitHub (200+ contributions). Previous role at similar-stage startup. No red flags. Recommended: Interview.
Key highlights: Led team of 3, shipped 2 major features, AWS certified.
Concerns: None major. Slightly short on backend experience (3y vs 5y required) but compensated by strong frontend.

HR review: 30 seconds per candidate (reads summary, checks score). Decision: interview / no. Total time: 100 CVs x 30 sec = 50 minutes.

Time saved: 7.5 hours.

Accuracy vs manual screening: Tested on 200 CVs. AI + HR review agreed with "pure HR review" 92% of the time. The 8% disagreement consisted of edge cases that would have required discussion anyway.

Step 2: Automated Pre-Screening Questionnaire (Replaces Phone Screen)

Problem with phone screens: 50% of time spent on questions that can be asked asynchronously. "Why do you want to change jobs?", "What are your salary expectations?", "When can you start?"

Solution:

  1. Candidates scoring >80 receive an email: "Congratulations! Next step: Please fill out this 10-min questionnaire"
  2. Questionnaire (built in Typeform): - 5 qualifying questions (salary range, availability, work arrangement preferences) - 3 behavioral questions (video answer, 2 min each): "Describe a challenging technical problem you solved" - 2 technical questions (code snippet review, multiple choice)
  3. AI analyzes video answers (transcription + sentiment + keywords)
  4. Scoring + flag if anything concerning

Candidate time: 15 minutes (asynchronous, at a time they choose).

HR time: 3 minutes review per candidate (reads AI summary + watches flagged parts of video).

Pass rate: From 20 after CV screening, 12 pass the questionnaire.

Time saved: 20 candidates x 30 min phone screen = 10h reduced to 20 x 3 min review = 1h. Savings: 9 hours.

Step 3: AI-Powered Scheduling (Zero Back-and-Forth)

Old way: Email ping-pong. "Are you available Tuesday?" "No, Thursday?" "Thursday the senior dev is on vacation, Friday?" ... 5-7 emails, 2-3 days.

New way: Calendly + AI optimization.

  1. Candidates who pass the questionnaire receive a Calendly link: "Book your interview"
  2. Calendly shows available slots (auto-synced with the interviewing team's calendars)
  3. AI optimizes slot suggestions: "Prefer slots where all interviewers are available" + "Avoid booking 3 interviews back-to-back (interviewer burnout)"
  4. Candidate clicks slot, auto-booked, calendar invites + prep materials sent

Time: Candidate books in 2 minutes. Zero HR involvement. Interview typically scheduled within 24 hours of passing the questionnaire.

Step 4: Structured Interview + AI Note-Taking

Interview setup (60 min):

  • Hiring manager + technical lead
  • Structured questions (same for all candidates -- eliminates bias)
  • AI notetaker (Otter.ai or Fireflies.ai): Joins call, transcribes, extracts key points

Post-interview (5 minutes):

  • AI generates summary: candidate answers to each question + highlights + concerns
  • Interviewers review, add notes, score (1-5 per criterion)
  • AI aggregates scores + flags major disagreements between interviewers

Time saved: Zero time spent writing post-call notes (previously: 15 min per interviewer). AI does it instantly.

Step 5: Fast Decision + Personalized Offer

Day 2 (after all interviews):

  1. Decision meeting (30 min): Review AI-generated comparison report (all candidates side-by-side, scored)
  2. Choose finalist
  3. AI generates personalized offer letter (templates + candidate-specific details auto-filled)
  4. Hiring manager reviews (5 min), approves
  5. Sent within the hour

Timeline:

  • CV submitted: Day 0, 9am
  • AI screening: Day 0, 9:10am
  • Questionnaire sent: Day 0, 9:15am
  • Candidate completes: Day 0, 6pm
  • Interview scheduled: Day 1, 2pm (auto-booked by candidate)
  • Interview happens: Day 1, 2-3pm
  • Decision meeting: Day 2, 10am
  • Offer sent: Day 2, 11am

Total elapsed: 50 hours (2 days + 2h). Rounded for marketing: 48 hours.

Results After 6 Months (Hard Data)

Metric Before After Change
Time to Offer 14 days 2 days -86%
Offer Acceptance Rate 45% 78% +73%
Candidate Experience Score 2.9/5 4.6/5 +59%
HR Time / Hire 28h 6h -79%
Quality of Hire (6m retention) 72% 88% +22%
Cost per Hire $2,100 $800 -62%

The key takeaway: Quality of Hire went UP, not down. The fear was that "a fast process = worse candidates." The reality: better candidates, because you stop losing them to competitors.

Tech Stack Used (Copy This)

ATS: Lever (alternatives: Greenhouse, Workable)

AI Screening: Custom integration Lever + OpenAI API (GPT-4)

Questionnaire: Typeform + video answers

Scheduling: Calendly (alternative: Cal.com)

Interview Notes: Otter.ai (alternative: Fireflies.ai)

Offer Generation: Custom templates in Lever + auto-fill with candidate data

Total monthly cost (for 10 hires/month):

  • Lever: $600/month
  • OpenAI API: ~$150/month (CV screening + analysis)
  • Typeform: $50/month
  • Calendly: $15/month
  • Otter.ai: $40/month
  • Total: $855/month

Savings: HR time saved = 22h/hire x 10 hires x $20/h = $4,400/month. ROI: 5.2x.

Implementation Roadmap (60 Days)

Weeks 1-2: Setup Infrastructure

  • Purchase licenses (ATS, Calendly, Otter, Typeform)
  • Setup OpenAI API + test on 20 sample CVs
  • Build scoring criteria (what is must-have, nice-to-have, red flag)
  • Create question bank (structured interview questions)

Weeks 3-4: Build Integrations

  • Lever webhook to OpenAI API auto-scoring
  • Typeform to AI analysis pipeline
  • Calendly sync with team calendars
  • Otter.ai auto-join calls

Weeks 5-6: Pilot (1 Role)

  • Run the process for 1 open position
  • Collect feedback from team + candidates
  • Iterate (adjust criteria, questions, flow)

Weeks 7-8: Rollout (All Roles)

  • Train HR team on the new process
  • Document the playbook
  • Monitor metrics weekly

Common Concerns (and How We Addressed Them)

Concern 1: "AI will discriminate against candidates"

Response: AI screening is actually more fair than human screening. Humans have unconscious bias (name, school, CV appearance). AI scores on objective criteria. PLUS: We audit AI decisions monthly. If we see a pattern (e.g., everyone from university X gets lower scores without reason), we fix the criteria.

Concern 2: "Candidates don't like talking to AI"

Response: AI doesn't talk to candidates. AI only analyzes data. Every interview is with a human. Candidate feedback: "The process was fast and transparent" -- not "I felt like a number."

Concern 3: "What if AI misses a great candidate?"

Response: Human in the loop. HR reviews every score above 60. Edge cases (e.g., non-traditional background) are flagged for manual review. In 6 months: zero cases where "AI rejected someone who should have passed."

What's Next? Your Action Plan

If you want to replicate this process:

  1. Measure your current baseline (time to hire, cost, acceptance rate)
  2. Choose 1 bottleneck to attack (CV screening? Scheduling? Notes?)
  3. Pilot on 1 role for 1 month
  4. Measure improvement
  5. If positive, roll out to all roles

Need help building an AI recruitment process? Book a consultation -- we will walk through your current process, identify bottlenecks, and design an AI-powered flow. Also check our case studies of other companies that cut time-to-hire by 70-90%.

FAQ

Does this work for non-tech roles?
YES. We tested it for Sales, Marketing, Operations, and Customer Success. The process is identical -- only the criteria differ.

Is this legal (GDPR)?
YES, if you do it right. You must: 1) Inform candidates that AI is used, 2) Provide an opt-out option, 3) Allow candidates to request human review. Consult with a compliance attorney. More: ICO AI Guidance.

What if we don't have a developer for integration?
Most of this can be done no-code: Lever has built-in AI on higher plans. Zapier can connect the rest. If you need custom work, hire a freelancer (cost: $750-$2,000 one-time).