This client's recruiting platform was losing deals to newer competitors with AI features. Their enterprise clients were demanding intelligent resume screening and matching automation. The engineering team had attempted to add ML features before, but the project was abandoned after 6 months—the model worked in testing but recruiters didn't trust it, so adoption was near zero.
We started by understanding why the previous ML attempt failed. The core issue: the model gave recommendations without explaining why. Recruiters couldn't tell if a recommendation was good or bad without doing all the work themselves, so they just ignored it. We proposed explainable AI: every recommendation would include the specific factors that drove it.
The new system uses fine-tuned embeddings to match candidates to job requirements, but more importantly, it shows recruiters why each match scored the way it did: "Strong match on required skills (Python, AWS), 3 years below experience preference, salary expectations 15% above budget." Recruiters can adjust weights on these factors and see scores update in real-time.
Average time-to-match decreased from 12 days to 5.4 days (55% improvement). We projected 60%, but some recruiter workflows were harder to change than expected. Recruiter productivity increased 1.8x (candidates processed per day). Match quality scores (based on client feedback) stayed flat—important because faster shouldn't mean worse. Adoption took 3 months to reach 80% of recruiters.
Unified patient data across 12 legacy systems. Reduced duplicate records by 73%—not the 89% we projected, but enough to materially reduce medication errors.
Built an ML-powered routing system that reduced fleet fuel costs by 18%. Took longer than planned because integrating with their dispatch system was harder than scoped.
Let's discuss how we can help you achieve similar results.