In the last post, we looked at where to start automating orders, inventory, and customer support in e-commerce. I get similar questions from founders in other industries too. A few days ago, it was a CEO running a staffing agency.
"We get more than 200 resumes a day. Just reading through them is a full-time job, and yet only a single-digit number actually make it to the interview stage." On the whiteboard in one corner of his office, twelve broken matches — open positions that had stalled — were marked with red magnets. Those magnets, he told me, hadn't moved in a month.
Is resume matching really the bottleneck?
The moment you bring up HR automation, the first keyword that comes up is resume matching. The global numbers point the same way. According to industry reports, as of 2026 more than 97% of Fortune 500 companies have adopted AI-based ATS (Applicant Tracking Systems). In Korea, Saramin launched its "Career Matching Agent" this past March, and matching platforms like HRCap's GO ACE reportedly logged a 37% average match rate during their pilot rollout.
Looking at the numbers alone, the area a staffing firm should tackle first seems obvious: resume screening.
But sit alongside the team on the ground for a week or so and the picture shifts. What actually eats up the most time isn't reading resumes. It's everything upstream — pulling the real requirements for the position out of the client — and everything downstream, namely onboarding after the hire. We initially thought "matching first, obviously" too, but once we were inside the operation the answer turned out to be different.
Real matching is about "intent," not "keywords"
Most job orders contain sentences like "3+ years of Python, AWS experience preferred." Run that through a keyword match against resumes and you get 50 candidates. You can't hand 50 candidates to a client. So in the end, the matching is redone by a human.
This is where the semantic matching tech often called ATS 2.0 starts to matter. Instead of plain keywords, it tries to read context — things like "the team is small, so we need someone with full-stack tendencies" or "the previous person quit after a year, so stability is the priority this time." One global vendor claims, in its own benchmarks, screening that's 3x faster than manual review with 87% accuracy. The number comes from the vendor itself, so take it with a grain of salt — but the direction of the trend is clear.
What we actually validated on the ground were automation points more like these:
- As soon as a job order comes in, an LLM summarizes "ambiguous parts that need clarification" — converting them into a list of questions the recruiter can send back to the client
- When resumes arrive, automate the first-pass filter (basic qualification check) and reserve the second-pass match (narrowing down for culture fit) for human review
- Capture interview feedback as text and feed it back as weights into the next round of matching
Onboarding — surprisingly high ROI
The area with the highest ROI in staffing and HR automation, in our experience, isn't matching — it's onboarding.
One client was bringing in 20 to 30 new hires every month. In the first week alone there were over 30 items to deal with: documents to file, training sessions to attend, accounts to provision. Two HR staff were burning half their day on this. At first we thought, "RPA could just take all of this off your plate," but once we were inside it, there were more sticking points than expected.
Things like equipment orders, scheduling security training, and matching mentors all required constant human involvement. So what we ended up building wasn't RPA — it was something closer to "checklist orchestration." Once a start date is set, 30 items are generated automatically, and each one pings the responsible owner over Slack. The items themselves still get handled by people, but the triggering and tracking move to the system.
With that structure in place, the time a person spent on each new hire dropped by about an hour. Thirty hires a month equals thirty hours. "Thirty hours saved" doesn't sound huge on its own, but in practical terms it's two fewer late nights a month for the HR team.
For an industry benchmark, Spotify reportedly cut manual onboarding work by more than 60% with a similar setup. Those are global big-tech numbers, so they don't translate one-to-one to Korean SMBs. But as a ceiling — "this is roughly how far automation can go here" — they're a useful reference.
Areas you shouldn't automate
Honestly, there are also areas in staffing and HR where automation is the wrong move. I once saw a company that automated both offer letters and rejection notices through a chatbot. Efficiency went up, but candidate satisfaction and reapplication rates dropped at the same time. There are moments where the value lies in a person telling another person.
The core judgment call in HR automation is: "If you remove the human from this task, what disappears with them?" If the value of the work comes from human judgment, relationships, or warmth, then automation should aim at freeing up that person's time — not at replacing them. It's not an easy line to draw, but it's worth defining once, right at the start.
So where do you actually start?
The order we most often recommend to people running staffing or HR operations is this.
First, bundle job-order review and first-pass resume filtering under an LLM. The return on time is the highest here. Second, automate the onboarding checklist and notifications — not as RPA, but as a trigger-and-tracking layer. Third, automate interview scheduling and post-interview feedback collection. These three come before any effort to improve the matching algorithm itself.
Wiring up these three areas usually takes 6 to 8 weeks and fits inside a PoC budget of roughly 2 to 4 million KRW per month. At 5years+, we've run automation PoCs across the HR and staffing space for SMBs in both Korea and Japan. The biggest difference we bring is taking the time, up front, to map out "where you shouldn't be automating in the first place." For a free consultation, head to our contact page.
The next post will cover logistics automation — pickup, tracking, customs. Long pipelines with heavy dependencies on external systems, and a very different automation design than HR.