The economics of replacing one part-time admin with one agent.

This post is for anyone running a small business who has a part-time admin doing repetitive work. Inbox triage. Sheet entry. Calendar coordination. Lead enrichment. Customer follow-up. If you’ve looked at AI agents and wondered whether the math actually works for your specific situation, this is the math.

I’ll show what a typical 20-hour-per-week admin costs fully loaded, what a Hapex agent costs to do the same workflows, and where the comparison breaks down. Then I’ll address what to actually do with the human when the agent is doing 70% of their work.

The fully loaded cost of one part-time admin

The number you see on the payroll line is not the number you pay. Here’s the breakdown for a 20-hour-per-week admin at $20/hour (slightly above minimum, slightly below the typical office admin rate):

• Wages: $20/hr × 20 hr/week × 52 weeks = $20,800/year
• Payroll tax (FICA + Medicare): 7.65% = $1,591/year
• State unemployment + workers comp: roughly 2% = $416/year
• Tools they need to do the job: Google Workspace seat ($14), occasional Zoom ($16/seat), Slack ($8/seat) = ~$456/year
• Management overhead: at least 2 hours/week of your time supervising, training, answering questions, reviewing work. At a conservative founder hourly value of $75 = $7,800/year of your time

Direct cash out: about $23,300/year. Add the founder time you spend managing them and the total economic cost is closer to $31,000/year, or about $2,600/month.

This is the part-timer doing roughly 1,040 hours of work per year. The hourly all-in cost to the business is about $30/hour delivered, even though they see $20.

What that 20-hour week actually covers

I’ve sat with enough small business owners to know what part-time admins actually do. The realistic breakdown for 20 hours:

• 4-6 hours: inbox triage, replying to routine inquiries, forwarding to the right person
• 3-4 hours: data entry into a CRM, spreadsheet, or accounting system from emails, forms, or paper
• 2-3 hours: scheduling, calendar coordination, reminder-sending
• 2-3 hours: customer follow-up emails to specific lists or segments
• 2-3 hours: generating routine reports (sales summary, inventory snapshot, weekly metrics)
• 2-3 hours: miscellaneous one-offs that show up that week

The first five categories — about 13-19 hours of the 20 — are exactly what AI agents do well. The last category is what humans still do better.

What the agent side of the comparison costs

You don’t buy these as separate agents anymore. One Hapex Operator runs all of them for your business, and it routes each job to the right model automatically: fast models for routine triage, genius-tier models for the harder judgment work. You subscribe to a plan based on how much you run, not by picking a model per task. For the workload above:

• Email triage — the Operator covers 4-6 hours of triage work
• Lead enrichment + follow-up — covers 5-7 hours of data entry, follow-up drafting, sheet updates
• Daily briefing + reporting — covers 4-6 hours of calendar coordination and weekly report generation

All of that fits inside one plan. Hapex Plus is $49/month for one business, all 75+ capabilities, up to 200 automation runs per month. If the workload above runs more than that, Hapex Pro is $79/month for up to 600 runs, smarter models via priority routing, and multiple connected accounts. Call it $49 to $79/month, or about $588 to $948/year. Annualized that’s under 4% of what the human costs to do the same work, before counting the management overhead you stop paying.

Even at the top, Hapex Max is $125/month, or $1,500/year, for unlimited runs (fair use) on genius-tier models. Still under a twentieth of the loaded admin cost. And if your volume is spiky, Hapex Flow has no monthly fee at all: you buy credits and spend one per run.

Where the math doesn’t work

I’m not going to pretend this comparison always wins. Here are the cases where the agent loses:

Your admin’s job is mostly that 20% miscellaneous category. If most of what they do is novel one-offs, judgment calls, customer relationships, or physical-world tasks (filing receipts, answering the phone in person, restocking the showroom), the agent saves you very little. Buy the agent for the repetitive 80% only if that 80% actually exists.

You under-pay your admin. If your part-timer is at $14/hour and you don’t pay any of the soft costs, the all-in is closer to $16k/year. The agent comparison still wins, but not by 5x — more like 5-6x, which is enough that some owners decide the human is still worth keeping for the relationship.

You don’t have a defined workflow for them today. If your admin’s job is "do whatever comes up," an agent will frustrate you. Agents work against described workflows. If you can’t describe what the work is, you can’t automate it — and you probably should have written it down for the human too.

What you do with the human

This is the part most owners don’t think through, and it’s the part that determines whether the agent is a real win or a layoff dressed up as efficiency.

The realistic options when your agent absorbs 70% of an admin’s work:

1. Move them up. The repetitive work was probably the lowest-value thing they did. Use their freed-up hours for the actually-hard work you’ve been doing yourself: client calls, follow-ups that need a human voice, real customer service, sales coordination. The all-in number stays the same; the work the business gets out of it goes up.
2. Reduce their hours. If they’re a parent or student who took the 20-hour role specifically because it was 20 hours, drop them to 10. Same hourly, half the cost, work still covered between them and the agent. They keep the role they wanted.
3. Let them go. If the work was the entire job and you can’t find higher-leverage work for them, this is the honest path. Give notice, write a useful reference, help with the transition. Don’t pretend the agent is augmenting them when it’s replacing them.

Most owners I’ve talked to land on option 1. The work the human was doing wasn’t the highest-value work; it was the most-urgent work. Agents handling the urgent stuff frees the human to do the important stuff. That’s the upside the comparison numbers don’t show.

Try the math on your own situation

Two quick questions before you commit:

1. What are the five most-repetitive tasks your admin (or you) does every week? Be specific: "send weekly Slack standup" not "communications."
2. Of those five, which ones could you describe to a brand-new hire in writing without it taking more than three paragraphs?

The ones that fit in three paragraphs are the candidates for your Operator. Take them to build.hapex.ai, paste the descriptions, and see which plan covers your run volume. Activation only happens after you run a live test against your real data.

If you want to see what a working agent looks like before you start, the use-cases gallery has five concrete examples with sample output for each.

— Shameel Khairi, Hapex AI