Your HR team is buried in resumes, surprised by attrition, and onboarding people with a Google Doc.
AI for HR for mid-market companies. Resume screening, attrition prediction, onboarding automation, workforce planning. Deployed in 3-6 weeks. $5-$20K.
The Problem
HR teams built for 50 people don't scale to 300.
Resume screening that rejects candidates you need: Your recruiter got 400 applications for a senior developer role and spent 12 hours filtering by keywords. She rejected a candidate who built distributed systems because his resume said 'backend architecture' instead of 'microservices.' Keyword matching filters out non-traditional candidates, career switchers, and anyone who describes their work differently than your job description.
Attrition that hits you three months too late: Last quarter, two senior engineers and a product manager resigned in the same month. The signals were there for months: declining engagement scores, a manager whose reports keep leaving, compensation below the 40th percentile. Nobody connected the dots because the data lives in four systems and gets reviewed quarterly in a spreadsheet that is already outdated. Replacing a mid-level employee costs 1.5-2x their annual salary.
Onboarding that takes six months to produce a useful employee: Your new hire's first week: three hours on forms that could have been pre-populated, a day reading a wiki last updated in 2023, two weeks waiting for system access because IT provisioning stalls on an email chain. Companies with structured onboarding get new hires productive in half the time. Yours is a Google Docs checklist the hiring manager forgets to share until day three.
Workforce planning based on headcount, not capability: Your leadership approved 15 new hires based on headcount requests. Nobody asked whether reskilling existing employees could close the gap at a fraction of the cost. Nobody mapped which roles will be partially automated within 18 months. Without capability visibility, you are hiring for today's org chart instead of next year's operating model.
Our Approach
Audit the mess. Build the fix. Measure what changed. We don't sell HR software licences. We audit your people operations, find where AI cuts the most manual work, and build systems that plug into your existing HRMS, ATS, and communication tools. You own everything we build.
Phase 1 — HR Process Audit (Days 1-3): Interviews with HR leads, hiring managers, and team leads across departments. We map every workflow: hiring pipeline, onboarding sequence, performance review cycle, exit process. We pull data from your HRMS, ATS, and whatever side spreadsheets your team is quietly maintaining. We measure time per process, error rates, and where decisions run on instinct instead of data. Output: where AI saves the most time and money in your HR function. Deliverable: HR process audit with workflow maps, time/cost analysis, data quality assessment, and prioritised AI opportunity list
Phase 2 — Model Design & Data Preparation (Week 1-2): For each AI system, we design the model architecture, define input data requirements, and prepare your historical data. Resume screening gets a scoring model trained on your successful hires, not keyword lists. Attrition prediction pulls data from HRMS, engagement surveys, compensation benchmarks, and manager feedback into one place. We clean the data, handle gaps, and validate that inputs are solid. You review and approve before we build. Deliverable: Model architecture documents, data pipeline specs, integration plan with existing systems, and validation criteria
Phase 3 — Build, Train & Test (Week 2-4): We develop the AI systems, train models on your data, and test against known outcomes. Resume screening is backtested against 12 months of hires: does it surface candidates your team selected while catching strong ones your keyword filters missed? Attrition prediction validates against known departures. Every system shows its reasoning. No black boxes. Deliverable: Trained and tested AI models with accuracy benchmarks, explainability reports, and integration with your existing HR tools
Phase 4 — Deploy & Handover (Week 4-6): Production deployment with one-week parallel testing. Resume screening runs alongside your existing process so recruiters can validate outputs. Attrition dashboards go live with historical context showing how the model would have flagged past departures. Full handover: training for every user, documentation, monitoring dashboards, and a documented escalation path for edge cases. Deliverable: Production-deployed AI systems, HR team training, full documentation, monitoring dashboards, and performance tracking
Deliverables
Audit & Design (Week 1-2)
- HR process audit with workflow maps across hiring, onboarding, retention, and workforce planning
- Data quality assessment and preparation plan for model training
- Model architecture and integration design for each AI system in scope
- Prioritised roadmap with expected ROI for each application
Build & Test (Week 2-4)
- Resume screening system that scores candidates on capability fit, not keyword matches
- Attrition prediction model with risk scores, contributing factors, and recommended interventions
- Onboarding workflow engine integrated with your HRMS and IT provisioning systems
- Back-testing and validation reports showing model accuracy against your historical data
Deploy & Handover (Week 4-6)
- Production deployment with parallel testing alongside your existing HR processes
- Training for HR team, hiring managers, and leadership (recorded)
- Full documentation: model logic, data inputs, decision explanations, and maintenance guides
- Monitoring dashboards for model accuracy, screening efficiency, and attrition risk alerts
Who This Is For
Right for you if: Mid-market company (50-500 employees, $2M+ revenue) where hiring volume has outgrown your HR team's ability to screen well. You are rejecting good candidates or wasting interview slots on bad ones.. You have had unexpected attrition in the last 12 months and want a system that identifies flight risk before the resignation letter arrives.. Your onboarding takes longer than 30 days to get a new hire productive, and the bottleneck is process, not people.. You have an HRMS or ATS with at least 12 months of historical data..
Not right if: Fewer than 50 employees or fewer than 20 hires per year. The ROI on AI screening does not justify the investment at that scale. Start with process automation instead.. No HRMS or structured employee data. AI models need historical data to train on. We can help you set up the data foundation, but that is a different engagement.. You want a plug-and-play HR software product. We develop custom AI systems for your data and processes, not software licences..
Use Cases
BPO / High-Volume Hiring: A 400-person BPO hired 60-80 customer service agents per month. Their two-person recruitment team spent 14 hours per batch screening 500+ resumes with keyword filters in their ATS. They shortlisted on exact phrase matches, rejecting candidates with relevant experience but different vocabulary. Time-to-fill averaged 28 days. — Built a resume screening model trained on 18 months of hiring and performance data. The model scored candidates on capability signals (communication patterns, tenure stability, domain exposure) rather than keyword matches. Integrated with their existing ATS so recruiters saw ranked candidates with explanations instead of a raw applicant list.. Outcome: Screening time dropped from 14 hours to 45 minutes per batch. Offer-to-join ratio went from 55% to 78%. Time-to-fill dropped from 28 to 16 days. Two recruiters now handle the volume that previously required four.
Tech / SaaS: A 180-person SaaS company lost 11 engineers in a single quarter, including three tech leads. Exit interviews cited 'better opportunities' but leadership suspected deeper systemic issues. No early warning system, no way to identify which teams or roles were at highest risk. — Built an attrition prediction model pulling data from their HRMS, quarterly engagement surveys, compensation benchmarks, promotion history, and manager span-of-control metrics. The model generated individual risk scores updated monthly, with contributing factors explained for each flagged employee. Built an intervention dashboard for HR leads showing risk by team, role, and tenure band.. Outcome: Model identified 85% of subsequent voluntary departures at least 60 days before resignation. HR intervened on 14 high-risk employees in the first quarter after deployment, retaining 9. Estimated savings of $55,000 in avoided replacement costs in six months.
Professional Services / Consulting: A 250-person consulting firm onboarded 8-10 new hires per month. The process involved 23 manual steps across HR, IT, finance, and the practice lead. Average time-to-productivity was 12 weeks. New hires consistently reported feeling lost during their first month, and 18% left within 90 days. — Mapped and automated the full onboarding workflow: pre-joining document collection and verification, IT provisioning triggered automatically on offer acceptance, role-specific learning paths based on the hire's experience profile, and a 30-60-90 day milestone tracker with automated check-ins. Added an AI assistant that answered common new-hire questions from the firm's internal knowledge base.. Outcome: Time-to-productivity dropped from 12 weeks to 6. 90-day attrition fell from 18% to 5%. HR time on onboarding admin cut by 70%. New hire satisfaction in the first-month survey went from 3.2 to 4.4 out of 5.
Results
A recent AI for HR project
BPO: resume screening and attrition prediction: Screening time cut by 95%, $47K in annual savings. A 400-person BPO with $14M in revenue had two problems: a hiring pipeline that could not keep pace with demand, and 42% attrition eating the gains from every successful hire. Keyword-based screening rejected 30% of candidates who would have been strong hires based on performance data. We built two systems: a resume screening model trained on 18 months of hiring and performance data, and an attrition prediction model pulling six data sources. Back-tested against 2,400 historical applications and 14 months of departure data. Total investment: $10,000. Within six months, screening time dropped from 14 hours to 45 minutes per batch, attrition fell from 42% to 31%, and estimated annual savings from reduced replacement costs and better hiring efficiency reached $47,000.
Frequently Asked Questions
How much historical data do we need?
Resume screening needs at least 12 months of hiring data with outcome tracking (who was hired, who performed well, who left early). Attrition prediction needs 12-18 months of HRMS data: engagement scores, compensation history, departure records. If your data is incomplete, we supplement with industry benchmarks, though accuracy improves with volume. The audit assesses data readiness before we commit to model specs.
Will the AI screening model introduce bias?
Keyword-based screening is already biased. It favours candidates who mirror your job description's vocabulary and penalises non-traditional backgrounds, career switchers, and anyone from less well-known institutions. Our models train on performance outcomes, not resume formatting. We run bias audits during testing, checking for disparate impact across gender, age, institution type, and career background. Every score comes with reasoning your recruiters can review and override.
Does this replace our ATS or HRMS?
No. Everything layers on top of your existing systems. The AI integrates with your ATS (Greenhouse, Lever, Freshteam, whatever you use) and your HRMS. Your HR managers keep working in the tools they know. The AI layer adds ranked candidate lists instead of raw applicant dumps, attrition risk flags in existing dashboards, and automated onboarding triggers from your current systems.
How long before we see results?
Resume screening improvements are visible immediately: screening time drops within the first batch. Attrition prediction takes 2-3 months to validate because you need to see whether the model's risk flags match actual departures. Onboarding improvements show up in time-to-productivity metrics within 60-90 days of the first cohort going through the new process. We set up measurement dashboards during deployment so you are tracking from day one.
What does this cost?
$5-$20K depending on scope. One area (screening, attrition, or onboarding) runs $5-$8K. Two or three areas with deeper integration runs $10-$20K. The process audit in week one gives you a precise scope and price before you commit to the full build. Every engagement includes projected ROI based on your actual data.
Can the attrition model predict departures in specific teams or roles?
Yes. The model generates individual risk scores, but the dashboard rolls up to team, department, role, tenure band, and manager views. Your HR leads can spot patterns: a manager whose reports consistently flag as high risk, or a role where below-market compensation is driving departures. The model tells you both who might leave and why.
What happens after handover? Do we need a data team to maintain the models?
The models deploy with monitoring dashboards that track accuracy over time. If performance drifts (which happens as your workforce composition changes), the dashboard alerts you. We include a 90-day performance review with every engagement. After that, periodic retraining is available through our AI Operations retainer. Your HR team works with the outputs (scores, dashboards, alerts) and never needs to touch the model internals.
We already use an HR platform with built-in AI. How is this different?
Most HR platforms offer generic AI trained on broad datasets. They do not know your company, your culture, or what a good hire looks like in your context. We build models trained on your data: your hiring history, your attrition patterns, your onboarding sequence. Generic tools give you a rough direction. Custom models give you accuracy you can make decisions on.
