Resources · case study · 01
Six vendors rejected. One approved in seventeen days.
A Fortune 500, NYSE-listed insurance carrier with 215+ locations rejected six AI hiring vendors over 18 months. They approved Nodes in 17 days — then ran a controlled study across 747 scored agents. The results challenged every assumption about what makes a good hire.
CustomerFortune 500 carrier
NYSElisted · anonymized
Applications / yr1.5M
ATSAvature
Study samplen=747 · 9+ mo tenure
Statuslive · phase 2 in deploy
Chapter 01 · the challenge
The best candidates were buried in the 98.5% nobody read.
1.5 million applications a year through an Avature ATS. Recruiters could manually review about 1.5% of them. Keyword filters — insurance experience, sales background, degree — had never been validated against actual production. The pattern recognition that separated good hires from bad was retiring with the managers who built it.
01 · volume
1.5% / 1.5M
The best candidates were buried in the 98.5% never reviewed.
Recruiter capacity was the bottleneck, not candidate quality. The 215-location network surfaced a narrow top-of-funnel and missed the rest.
02 · unvalidated
6 filters
Keyword filters, never validated against production.
Insurance experience, sales background, customer service, leadership, communication, teamwork. Applied first, measured never. No one knew which filter correlated with a good hire.
03 · institutional
retiring
The judgment their best managers built over decades was disappearing.
The pattern recognition for "this person will produce" sat in human memory, never documented. As managers retired and candidate volume grew 10–14×, institutional knowledge was vanishing faster than it could be transferred.
Chapter 02 · what we found
Your #1 filter eliminates 80% of your award winners.
Once deployed, Nodes connected ATS screening criteria to production milestones at the individual agent level. We tested each of the six standard keywords against RPA (Rookie Production Award). Insurance experience — the filter most employers apply first — has zero predictive power for RPA (p=0.56) and eliminates 80% of the people who eventually win it.
keyword.ablation · rpa achievement · n=535
keyword
rpa with
rpa without
ratio
p-value
% winners eliminated
Insurance experience
10.9%
9.0%
1.20×
0.56 null
80%
Sales experience
13.8%
6.0%
2.16×
0.007 sig
34%
Customer service
9.3%
9.4%
0.99×
1.00 null
42%
Leadership
9.2%
9.6%
0.96×
0.87 null
30%
Communication
10.2%
8.8%
1.17×
0.65 null
58%
Teamwork
8.3%
9.6%
0.87×
0.85 null
84%
cumulative.funnel · rpa winners remaining
No filter appliedstarting pool
100%
+ Insurance experiencecumulative
20%
+ Sales experiencecumulative
18%
+ Customer servicecumulative
12%
+ Leadershipcumulative
8%
+ Communicationcumulative
6%
+ Teamworkall 6 filters applied
2%
screening for the "ideal hire" with all six standard criteria
98% of winners · never interviewed
Chapter 03 · predictive accuracy
Traditional screening performs identically to a coin flip.
AUC is the standard measure of a binary predictor: 0.5 is random, 1.0 is perfect. The six-keyword composite performs identically to a coin flip on both milestones. Nodes Fit Score is the strongest single predictor (AUC 0.618, p=0.006). Combined with the one traditional signal that works — sales experience — it reaches 0.644.
predictor.auc · rpa achievement
higher = better
Random — coin flip
baseline · no predictive signal
0.500sna
0.500rpa
Traditional fit · 6 keywords
p=0.25 · not significant
0.512sna
0.548rpa
Nodes Fit Score
p=0.006 · strongest single predictor
0.567sna
0.618rpa
Sales keyword + Nodes
combined signal · ats + behavioral
0.576sna
0.644rpa
5.37×
sales + high nodes vs no sales + low nodes · rpa rate 15.3% vs 2.9% · p=0.0007
score.quintile · sna & rpa rate
monotonic lift
Q1n=157 · lowest
15.9%
5.1
baseline
Q2n=145
14.5%
8.3
+1.6×
Q3n=163
19.6%
8.0
+1.6×
Q4n=144
18.1%
9.7
+1.9×
Q5n=138 · highest
22.5%
13.0
+2.55× Q1→Q5
SNA rate · first production milestone
RPA rate · sustained production award
Hires scoring ≥72 achieve top-performer milestones at 2.47× the rate of those below.
measured across 747 nodes-scored agents · 9+ months tenure · p=0.006 · out-of-sample
Chapter 04 · the constant
Each day faster to first production = $54.35 more in annual output per person.
Across 10,362 hires from 2022–2025, we tested whether speed-to-milestone predicts annual output. Without scoring (n=1,011, 2022–2024), the correlation was noise (r=+0.045, p=0.15). With scoring active (n=679, 2025), every bucket step slower meant lower production — the fastest 30-day hires produced 1.8× the slowest 121+ day hires. Scoring creates the relationship between speed and output that did not exist before.
- Median speedup 62 vs 109 days · 47 days faster
- At 2,000 hires / yr $5.11M additional production
- At 500 hires / yr $1.28M additional production
- Compounds forward each quarter · scored → faster → scored
Chapter 05 · three hires
Three hires your ATS would have missed.
The behavioral signals that separate top producers from underperformers are invisible to keyword matching. Adaptability, resilience, relationship building — none of them searchable by industry-experience or job-title match. Two candidates the ATS would have skipped became top producers. One "perfect resume" did not.
agent A · anonymized
fit · strong
Store worker, disc golf shop.
No insurance experience. No sales experience.
No insurance experience. No sales experience.
resume verdict
ats · skip
nodes score
86
strong fit
production result
SNA + RPA achieved.
key behavioral signals
Adaptability 0.8
Technical Aptitude 0.7
Self-Motivation 0.8
would have been skipped
✓ top producer
agent B · anonymized
fit · spotlight
Restaurant server, 17 years.
No insurance experience.
No insurance experience.
resume verdict
ats · skip
nodes score
92
spotlight
production result
SNA achieved. Top producer.
key behavioral signals
Client Relationship 0.9
Resilience 0.8
Communication 0.8
would have been skipped
✓ top producer
agent C · anonymized
fit · below core
Insurance + sales experience.
"Perfect" resume on every keyword.
"Perfect" resume on every keyword.
resume verdict
ats · hire
nodes score
39
below core
production result
No SNA. Did not produce.
key behavioral signals
Resilience 0.2
Adaptability 0.3
Relationship Building 0.2
would have been hired
× did not produce
Chapter 06 · what the fit score analyzes
A multi-dimensional profile for every candidate.
Each Fit Score is the weighted output of 28+ individual assessments across four categories, calibrated against the carrier's validated top-performer persona by location and role. A warehouse worker's "equipment repair" maps to technical aptitude; a restaurant server's "17 years customer-facing" maps to relationship building. Keywords can't do that math.
category · 01
6
Behavioral
Patterns derived from career history — how someone navigates transitions, handles responsibility, and operates under pressure.
- Teamwork
- Leadership
- Adaptability
- Communication
- Problem-Solving
- Resilience
category · 02
16+
Skills
Scored against the carrier's persona. Signals mapped from adjacent roles, not just job-title matches.
- Sales Acumen
- Closing Skills
- Consultative Selling
- Persuasive Communication
- Objection Handling
- Client Relationship Building
- Self-Motivation
- Technical Aptitude
- Ethical Conduct · and more
category · 03
6
Cultural Fit
Values alignment from career history and role patterns. Predicts cultural integration, not just performance.
- Integrity
- Collaboration
- Accountability
- Drive for Results
- Continuous Learning
- Customer Centricity
category · 04
3
Career Analysis
Trajectory patterns. Are they growing in responsibility? Are their roles getting more complex over time?
- Promotion Rate
- Complexity Growth
- Achievement Density
Chapter 07 · results
Year one, in numbers.
Eleven metrics your compliance team is already asking about. Every line here traces to a customer-side production record, not a vendor-reported figure.
results.table · year one · production record
Candidates processed730,000+ · across 215 locations
Contracted hires · 20252,335
Top-performer lift2.47× · high-scored hires · p=0.006
Resume screening reduction40% fewer manual screens
Interview reduction2 fewer interviews per hire
Time-to-hire127 → 38 days · 70% reduction
Year-one customer savings$1.58M
Deployment time34 days · contract → production
Legal approval17 days · fastest in company history
Prior vendor rejections6 vendors blocked over 18 months
Data egressZero.
CertificationsSOC 2 Type I + II · held
samplen=747 · 9+ mo tenure
milestonesSNA · RPA · binary
temporal controlorthogonal to contract month · all p>0.10
p-valuestwo-tailed
sourcecustomer ATS + HRIS
Chapter 08 · after deployment
Risk addressed. Then it compounds.
Every enterprise buyer evaluates three risks before deploying AI in hiring — implementation, adoption, accuracy. The carrier's own evidence answers all three. After that, the system gets sharper with every hire: success profiles retrain on customer outcomes, inside the customer VPC.
risk
the concern
carrier's evidence
01Implementation
What if deployment takes 6 months and disrupts our workflows?
34 days to production. 17 days legal. Zero workflow change. Native Avature integration — recruiters never left the ATS.
02Adoption
What if recruiters ignore it, or it adds steps?
Sustained usage across 215 locations for 8+ months. Adoption driven by less noise in shortlists, not more tools.
03Accuracy
What if predictions are wrong and we face bias claims?
Validated against SNA and RPA milestones (p=0.006). Full audit trails, explainable scores, bias monitoring. EEOC / OFCCP aligned.
compounding · what gets sharper with every hire
inside customer vpc
stage · 01
Months 1–3
Success profiles calibrated. First decision traces captured.
- 28+ dimensional scoring active
- Screening begins across 215 locations
- Time-to-hire starts dropping
stage · 02
Months 4–8
Outcome data validates predictions. Models retrain on customer performance data.
- Accuracy improves without data egress
- Production milestone data feeds back into scoring
- Persona calibration sharpens per role
stage · 03
Year 2+
Queryable library of hiring reasoning. Success profiles extend to workforce development.
- Phase 2 · CRM transcript ingestion
- Persona-based sourcing scales
- Intelligence gap becomes durable
What to take into the room
Want this methodology redlined against your vertical?
A 30-minute architecture walk with your compliance team. We bring the data-flow diagram, a sample audit trail, and the on-job correlation pack — printed, redlined against your requirements.
"We are screening thousands in hours instead of weeks, and the quality is measurably better. Very excited for this partnership to evolve even more."
— Director of field sourcing · Fortune 500 carrier
What your team leaves with
- Data-flow diagram pre-aligned to your VPC
- Sample audit trail for a scored candidate
- On-job correlation pack for your vertical
- SOC 2 Type I + II report under MNDA
- Fit Score methodology · 28+ dimensions
- A redlined copy of this case study