31% of New Hires Left Within 12 Months

New hire starts Monday. Claire, the operations manager, sent an email to IT the preceding Wednesday: "New starter Monday, please set up laptop and accounts." IT confirmed receipt. Sometimes the laptop was ready Monday. Sometimes Wednesday. Once, it took 9 days. The new hire spent their first week and a half logging into a colleague's machine to complete HR paperwork, which is a first impression that communicates something rather specific about how organised the company is.

Claire adds a new tab to her spreadsheet. Thirty-four rows. One per milestone. IT setup, system access, compliance training, team introductions, client exposure, buddy assignment, 30-day check-in, 60-day check-in, 90-day check-in, probation review preparation, and 24 role-specific training modules. Each milestone assigned to a responsible person: IT for provisioning, line manager for introductions, compliance team for training, buddy for social integration.

She sets calendar reminders. Week 1, week 2, week 4, week 6, week 8, week 12. At each checkpoint, she's supposed to open the spreadsheet, chase anyone behind, and update the colour coding. Green for done. Amber for in progress. Red for overdue. White for not started.

In practice, she checks at week 1 (the new hire is physically present, hard to ignore). She checks at week 4 (the 30-day review reminds her). She checks at week 12 (probation review is imminent).

Weeks 2, 6, and 8 are, in Claire's honest assessment, "aspirational." Between check-ins, milestones drift. The compliance training that should be done by day 14 gets pushed to day 30. The buddy introduction that should happen day 1 happens day 12 (or not at all). The client exposure that should start by week 3 starts at week 6 because the line manager forgot.

Claire is also running office operations, facilities, and vendor management. Onboarding tracking is approximately 20% of her job by importance and approximately 5% of her job by available time, which is a gap that 34 milestones fall into with predictable regularity.

HR tech and professional services firm in Bristol. Thirty-eight employees. £2.8M annual revenue. Hires 15 to 20 new employees per year across consulting, development, and operations.

Claire's loaded cost: £38 per hour. She spends an estimated 8-12 hours per hire across the 90-day onboarding period: chasing, checking, updating, reminding. At 17 hires per year, that's £5,168 to £7,752 annually in Claire's time alone. Line managers spend an additional 4-6 hours per hire chasing onboarding tasks that should have been prompted automatically: £4,080 to £6,120 per year. Total annual cost of manual onboarding tracking: £9,248 to £13,872.

The completion numbers from the prior 12 months: 71% of milestones completed by the 90-day probation review. Meaning 29% were still outstanding when the formal assessment was due. Four of 17 hires reached their probation review without completing mandatory compliance training. Claire discovered this at the meeting. The spreadsheet had said "in progress" for all four. It had said "in progress" for 76 days.

Voluntary turnover within the first 12 months: 31%. Industry average for their size and sector: 18-22%. Exit interviews were consistent. "I didn't feel set up properly." "I was still figuring things out at month 3." "My buddy never really connected." "I didn't know who to ask." The onboarding programme existed. The follow-through didn't. And the gap between having a programme and executing it consistently was costing the firm roughly one in three new hires within their first year.

The spreadsheet captured intent. Thirty-four milestones, properly defined, properly assigned, properly sequenced. On paper, thorough. In practice, the programme depended on Claire remembering to check, responsible people remembering to act, and everyone maintaining focus on onboarding tasks alongside their primary jobs for 90 consecutive days.

The structural problem: onboarding milestones have dependencies. Laptop must be ready before system access. System access must exist before training. Training must complete before client exposure. If the laptop is 5 days late, every downstream milestone shifts. The spreadsheet didn't model dependencies. Claire carried them in her head. When IT was late (roughly 40% of hires), she had to mentally recalculate every subsequent timeline. Sometimes she did. Sometimes she was managing a facilities crisis and the recalculation didn't happen at all.

BreatheHR had a basic onboarding checklist feature. Nobody used it because it didn't connect to IT provisioning, the LMS, or Slack. A checklist that can't see whether the work was done is a list of aspirations.

Four stages. All built on top of the existing stack.

Stage 1: Trigger and plan generation

New hire confirmed in BreatheHR triggers the agent to create a 34-milestone plan with owners, due dates, and dependency chains. Pre-start tasks fire automatically: IT provisioning request sent 5 days before start date with confirmation due 2 days before. If no confirmation, Claire gets an alert triggered by the absence of confirmation, not a reminder she set for herself and might dismiss during a facilities meeting.

Stage 2: Milestone tracking

Monitors completion via integrations where possible. LMS webhook confirms compliance training completion (the agent knows before Claire does). Slack activity confirms buddy introductions happened. Calendar event completion confirms check-in meetings occurred. Where automation isn't possible (subjective milestones: "is the new hire integrating with the team?"), the agent prompts the responsible person at the right time and logs their response.

The dependency chain is encoded. If IT provisioning is late, every downstream milestone shifts automatically. Claire doesn't need to recalculate manually.

Stage 3: Automated prompts and escalation

Responsible people receive prompts at milestone due dates. If no response or completion within 48 hours: escalation to Claire with context. If Claire doesn't act within 24 hours: escalation to the line manager. Dependency blocks trigger immediate alerts (training can't start without system access; system access delayed means training delayed means compliance risk).

Milestones don't drift silently for weeks between Claire's check-ins. They drift for a maximum of 48 hours before someone is asked why.

Stage 4: Onboarding health dashboard and probation readiness

Weekly view: all active hires, milestone completion percentage, overdue items by owner, dependency blocks, and a probation readiness score (percentage of mandatory milestones complete with days until review). Claire reviews in 15 minutes. Line managers see their own hires only.

The probation readiness score addresses the compliance training gap directly. Four hires reaching probation without mandatory training is now structurally impossible because the alert fires at day 15, not day 90.

The dependency chain encoding was the highest-value work in the build. Thirty-four milestones with 12 dependency relationships. When IT provisioning was late for the first hire onboarded with the agent, every downstream milestone shifted automatically and affected owners received updated timelines within an hour. Under the old system, Claire would have discovered the delay at her next spreadsheet check (days later), recalculated manually (if she had time), and re-chased (if she remembered). The agent turned a cascading delay into a managed adjustment.

The LMS integration eliminated the compliance training gap entirely. The agent knows the moment a training module is marked complete. It knows the moment the 14-day deadline passes without completion. Four hires reaching probation without mandatory training is now structurally impossible because the alert fires at day 15.

The buddy introduction failure rate was higher than anyone realised. Under the old system, Claire assumed buddy introductions happened because they were assigned. The agent tracked Slack activity for the first direct message between new hire and buddy. In the first quarter, 3 of 5 new hires had no Slack contact with their assigned buddy in the first week. The prompt to the buddy (triggered by absence of contact) resolved all three within 48 hours. Claire's assumption that buddies were connecting had been, in her words, "optimistic."

The invisible 30% showed up here too. Claire was spending roughly 25% of her capacity on onboarding tracking that didn't appear in her job description. She was the system. The agent replaced the checking, chasing, and calendar-reminder infrastructure. Claire still handles exceptions. The agent handles the 578 milestones per year that were falling between her three check-ins.

£180 per month. £2,160 per year. Against £9,248 to £13,872 in direct tracking costs. And against the 31% first-year turnover that exit interviews attributed to onboarding quality.

Each hire who leaves within 12 months costs the firm an estimated £15,000-£20,000 in recruitment, training, lost productivity, and institutional knowledge. At 31% turnover across 17 hires, that's 5-6 departures costing £75,000-£120,000 per year. If the onboarding agent prevents even two of those departures, the return exceeds the cost by an order of magnitude.

Claire described the change: "I used to find out what hadn't happened at the probation meeting. Now I find out what hasn't happened on Tuesday and it's fixed by Thursday."

If your onboarding has a checklist that gets checked three times in 90 days, your completion rate is lower than your spreadsheet suggests. "In progress" is not a status. It's a holding pattern that can last indefinitely unless someone prompts the responsible person, tracks the dependency chain, and escalates when things drift.

Your operations manager didn't fail at onboarding. The system failed to exist. She was the system. And a human being who is also managing facilities, vendors, and office operations cannot simultaneously be a reliable 34-milestone tracking engine for 17 concurrent onboarding programmes.

The Monday morning bottleneck here isn't a single morning. It's 90 days of chasing per hire, conducted by one person, between three check-ins, across five systems that don't talk to each other. Every new hire creates a fresh 90-day version of it.

This is Blueprint #44 in the AdAI series. Every week we publish the mechanics of a real AI agent build: the bottleneck, the systems, the logic, the results. Free to read. No signup required.

Want the next one in your inbox? Subscribe to AdAI News. One pair of articles per week (one blueprint, one strategic framework). Every Thursday.

by TG
for the AdAI Ed. Team