What an AI Recruiting Automation Company Should Prove in a Year
Most pitches from an AI recruiting automation company come with a demo. Mariia Dykun, Head of People Operations at Kindgeek, has something a demo can't offer: a year of production data. Her recruitment team put Easyflow's agents into daily use twelve months ago, kept four of them running, and killed two on purpose. A demo shows what an agent can do. A year of real hiring shows what a team still bothers to use.
What Should an AI Recruiting Automation Company Prove?
An AI recruiting automation company should prove that its agents are still used after launch, save measurable recruiter time, integrate with the existing ATS and hiring tools, transfer ownership to the client, and handle candidate disclosure, bias risk, and human review clearly. A demo is not enough; a 12-month production reference is the stronger proof.
This piece is built from a conversation with Dykun about exactly that year, agent by agent: what stuck, what got shut down, and what changed for her team along the way. A few things stood out before we even got into the details.
Key Takeaways
Four of the six agents Kindgeek deployed made it through twelve months of real hiring. Two got shelved once they stopped earning their keep.
The agent that reformats CVs into Kindgeek's corporate template alone freed up roughly a quarter of the sourcer's working time.
The hardest conversations weren't technical. Telling a candidate a bot helped judge their application, and keeping bias out of the pipeline, are still the two things Mariia calls tough, a year in.
Convincing a skeptical CFO took one thing: hours saved per week.
It doesn't come across as a polished marketing pitch, and that's what makes it genuinely valuable. So let's dive deeper into what this honest conversation uncovered.
What AI Recruitment Automation Actually Builds
In the world of recruitment, many buyers find themselves caught in a web of confusion, believing that AI recruitment automation is just another form of AI recruiting software. However, the reality is far more complex, and this distinction often becomes painfully clear only after they've invested six months in a subscription that fails to integrate with their ATS.
Before working with Easyflow, most of Kindgeek's recruiting automation was built in-house. The head of People Operations tested tools herself, picked up ideas at HR conferences, and pieced workflows together in Zapier. It worked, to a point.
“Once we reached, for example, the stage of adding steps with pre-written code, that's when it started to stall,”
— Mariia Dykun, Head of People Operations at Kindgeek
Bringing in a dedicated team changed the pace of the work more than it changed the ambition:
“We felt a really great relief, because personally, I saw that it takes a lot of time to research how it works, find something, and explore deeper technical moments.”
— Mariia Dykun, Head of People Operations at Kindgeek
That gap between testing a tool yourself and having a team build it properly is roughly the difference between AI recruiting software and an AI recruitment automation company.
AI recruiting software, the kind bolted onto Greenhouse, Workday, or one of a dozen newer platforms, sells you a seat. You configure it inside somebody else's product, and their roadmap decides what changes next quarter.
A staffing agency sells you a placement: useful, but it doesn't leave behind a system.
An AI recruiting automation company builds the agent to fit whatever ATS, calendar, and messaging tools a team already runs, then transfers the source code, documentation, and a knowledge-transfer session once it's live.
Gartner's HR practice called this shift directly heading into 2026, naming recruiter AI agents as one of the technologies moving from pilot to mainstream, alongside generative AI and interview intelligence tools. The four agents Kindgeek is still running a year later are what that looks like once the pilot phase is over and nobody's watching the demo anymore.
For a CHRO exploring an AI recruiting software alternative, the key consideration is ownership: who retains the outcomes after the contract concludes, and whether “AI recruiting agent development” involves meaningful integration within your existing systems, or simply adds another layer of management.
Automation Company | Recruiting Software | Recruiting Agency | |
|---|---|---|---|
What you get | Custom agent built into your existing ATS and tools | A seat inside the vendor's platform | A filled role |
Pricing model | Fixed scope, priced by complexity | Per-seat subscription | Placement fee |
Who owns it after launch | You, with source code and documentation | The vendor, indefinitely | Not applicable |
Roadmap control | Yours | Theirs | Not applicable |
What to watch for | Confirm ownership and handover terms are written into scope | Check what breaks when you outgrow platform defaults | Check what happens once the search ends |
The Agents Still Running After a Year
After countless hours spent sifting through resumes and drafting onboarding documents, four agents emerged from the demo phase to become a lifeline in Kindgeek's daily recruiting workflow. They quietly tackled the repetitive tasks that had drained the recruiters' energy. Each agent took on a specific, named piece of manual work, alleviating the exhaustion that had long weighed on the team's shoulders.
The CV formatting agent takes a candidate's original resume and rebuilds it in Kindgeek's corporate template, matching branding and colour without a recruiter opening a design file. Before it existed, that task alone ate close to a quarter of the sourcer's working time.
The pre-onboarding document agent drafts every document a new hire needs before day one: service agreements, NDAs, privacy policy paperwork, populated automatically with the candidate's own data.
The AI interview analyzer pulls a transcript from Fathom after each interview stage and returns a structured feedback summary within seconds, covering screening calls, technical interviews, and finals alike. Every candidate is asked for consent before the interview starts, confirming they're comfortable with a tool that records the call and takes notes.
And a Telegram bot handles job-posting distribution across channels and sends welcome messages when a vacancy opens. It's the same category of tool behind Easyflow's Telegram recruitment agent for another client, which scaled outbound candidate outreach 10x without adding headcount.
“We already have agents, which are a really important and natural part of our workflow,” Mariia Dykun said. A year in, her team isn't asking whether to use them. They're asking for more: “Everyone on the team is actually asking, ‘set it up for me, please,’ or if someone doesn't have it set up yet, ‘let me into the Claude environment, I don't want to do this manually anymore.’”
The builds were specifically designed for Kindgeek's People Operations routine, yet they can be customized to suit any organization and its processes. Easyflow's recruitment agent catalogue documents comparable time savings for the same categories of work.
The following data, derived from Easyflow's agent benchmarks, illustrates the significant time savings achieved through AI automation in recruitment processes, providing insights that any effective AI hiring automation firm should present before contract finalization.
Agent | Manual task replaced | Results shown | Status after 12mo |
|---|---|---|---|
CV formatting agent | Reformatting resumes into corporate template | Cuts CV/profile review from ~5hrs to 2 | Running daily, no longer touched by sourcer |
Pre-onboarding document agent | Manually drafting service agreements, NDAs, privacy docs | New-hire setup from ~3hrs to 30 min | Running for every accepted offer |
AI interview analyzer | Reviewing transcripts, writing feedback by hand | Interview analysis from up to 12.5hrs to 10 | Running after every interview stage |
Job-posting bot | Manually posting vacancies across channels | Multi-platform publishing from 1hr to 15 min | Running daily |
These four agents are a slice of a much larger map. Easyflow's recruitment automation playbook breaks the full hiring lifecycle into twelve agent blueprints, sourcing through onboarding, with before/after time comparisons for each stage.
The Two Agents They Killed
The Kindgeek's AI recruitment automation experience reveals not only what was embraced but also what was left behind in the pursuit of efficiency.
The first was an English-level assessment agent, built to screen candidates on language ability at the first-round stage. On paper it worked: it scored grammar, vocabulary, and structure competently. In practice, it optimised for the wrong thing.
“We noticed that agent makes really complex evaluations, including grammar parts, vocabulary and so on. Candidates with really great pronunciation, with really great confidence, they can have really lower level of grammar parts, but the agent gives them a great level, for example, upper intermediate and above.”
— Mariia Dykun, Head of People Operations at Kindgeek
For roles where a candidate needs to sound confident on client calls rather than pass a grammar exam, that mismatch was disqualifying. The agent got shelved.
The second was a recruitment analytics agent, meant to consolidate reporting across the vacancies. It's still done by hand in Excel today, not because the agent failed technically, but because of the mix of sources it needed to summarize. The listings were too fragmented to be worth automating yet. Both partners are still looking for a solution, coming back to this topic in all-hands meetings.
That's fit-testing working the way it should. A vendor worth hiring will tell you which of its own agents didn't earn their keep, not just recite the ones that did. A sales deck with zero casualties in it is the one worth questioning.
What Changed on the Team
Mariia Dykun is careful not to overclaim what the agents changed. “I can't say that the main reason was the introduction of AI agents,” she said, “but I can definitely state that one of the reasons was the introduction of agents.”
Before Easyflow, Kindgeek's recruitment team ran four full-cycle recruiters and one sourcer, and the sourcer's day was split between finding candidates and clearing administrative work.
“About 25% of the sourcer's time was spent on creating corporate CVs. We were able to redistribute the sourcer's working time to higher-quality sourcing, namely deep diving into vacancies, expanding the search grid.”
— Mariia Dykun, Head of People Operations at Kindgeek
Telling her team that AI wouldn't take their jobs turned out to be the easy part: “Easy. Our team is really mature and has a great vision about AI, about their future and career growth.”
The harder questions weren't about the team's own jobs. They centered on the people on the other side of the process, and how much the pipeline could be trusted to be fair.
Candidate Trust, Bias, and Governance in AI Recruiting Automation
When to Tell Candidates AI Was Used
Telling a candidate that a bot had helped judge their application landed differently than the internal questions did:
“Tough. A lot of candidates are worried about that, and we have a lot of different legal restrictions regarding it. I believe it's a really sensitive topic.”
— Mariia Dykun, Head of People Operations at Kindgeek
Gartner flagged this directly as an open question heading into 2026, recommending that talent acquisition leaders clarify AI use for candidates and offer an opt-out where practical, specifically to protect trust rather than just compliance. In practice at Kindgeek, that disclosure happens before the interview analyzer ever runs: candidates are asked directly, before the call starts, whether they're comfortable with a tool that records the call and takes notes.
Why Human Review Still Matters
Mariia draws a clear line around where the agents stop.
I don't want to rely only on agents or on AI. The agent is a helper.
Recruiters compare their own read of a candidate against what the AI surfaced, because it often catches signals a rushed reviewer might miss, but the final call stays human.
How to Monitor Bias in Recruiting Agents
Keeping bias out of the pipeline got a blunt verdict from an experienced Head of People Operations: “Probably tough.”
Kindgeek's own experience shows what that looks like in practice. Its English-level assessment agent, discussed above, overweighted grammar and vocabulary over pronunciation and confidence, a bias toward one dimension of communication ability that quietly disqualified candidates who were strong on the criteria that actually mattered for the role. It got shelved once the pattern surfaced. That's closer to bias monitoring in practice than any dashboard: someone checking outcomes against what the role actually needs and being willing to turn the agent off when they don't match.
What Final Hiring Decisions Should Stay Human-Owned
Sourcing quality, candidate judgement, and final hiring decisions stayed with Kindgeek's recruitment team throughout the rollout. None of the four agents still running make a hiring call on their own; they format documents, surface information, and draft communication faster than a person could, then hand the decision back.
Kindgeek's answers suggest disclosure and bias are less a policy problem to solve than a standing discomfort a team learns to live with, and probably should.
How to Measure ROI From AI Recruiting Automation
Kindgeek's numbers point to a pattern worth being able to show, not just claim. Here's where the hours actually come from.
Recruiter Hours Saved
Getting a skeptical CFO to sign off on ROI was easy for Dykun, once the pitch came down to a specific number: “The most important thing is to show numbers, for example how much time our team spends every day, and how the agent can help with that.” At Kindgeek, that number started with the CV formatting agent alone: roughly a quarter of the sourcer's working time, back in the team's hands within weeks of launch.
Time-to-Shortlist
Sourcing is typically where AI recruiting agents show up first, and it's also where the time savings are largest. Easyflow's AI Sourcing Agent, the same category of tool behind Kindgeek's own candidate search, cuts discovery and shortlisting from up to 30 hours to 10 in Easyflow's own published benchmarks, a 66% reduction.
Time-to-Interview Feedback
The same pattern shows up after the interview, not just before it. Easyflow's Tech Interviewer agent, the public equivalent of the AI interview analyzer Kindgeek runs after every stage, cuts interview analysis from up to 12.5 hours to 10, a 20% reduction, by turning a raw transcript into a structured, validated report automatically.
Onboarding Document Time
AI onboarding automation is usually the easiest ROI case to make, because the task itself is entirely paperwork. Kindgeek's pre-onboarding document agent removed manual drafting of service agreements, NDAs, and privacy paperwork entirely. Easyflow's public Onboarding Agent shows the same category of task falling from about three hours to thirty minutes.
Candidate Experience Risk
Not every ROI line item is a time saver. Disclosure and bias, the two things Dykun still calls tough, are the cost side of the same ledger: get them wrong and the hours saved upstream get spent again downstream, in complaints, legal exposure, or candidates who quietly opt out of the process.
Adoption After 12 Months
The clearest ROI signal isn't a single metric. It's that four of six agents are still running a year later, without anyone on Kindgeek's team being reminded to use them. A tool nobody has to be pushed to keep using has already paid for itself in a way a pilot-stage demo can't.
What to Ask Before You Sign
Most organizations evaluating recruiting AI right now are still in the pilot stage, not running it in production. McKinsey's November 2025 State of AI survey found that 88% of organizations regularly use AI across the business. However, only 23% are scaling an agentic system in any single function, and adoption in most companies still hasn't moved beyond experimentation. A vendor with a genuine 12-month reference is unusual enough that it's worth treating as a filter on its own.
Before signing with any recruiting automation vendor, a short list of questions does more work than a feature comparison:
Question to ask | Why it matters | What to watch for |
|---|---|---|
Which agents have you decommissioned, and why? | Shows whether the vendor tests fit honestly | A clean record with zero killed projects |
Who owns the code and documentation after handover? | System vs. rented access | Vague “ongoing support” language |
Does this integrate into our existing ATS? | Value collapses if it's another platform to manage | Integration treated as a roadmap item |
Can you show a reference at 12 months, not 90 days? | Early results are common; retained use isn't | Case studies that stop at launch |
Fixed scope or ongoing subscription? | Fixed scope means paid to finish, not to keep billing | Per-seat pricing dressed as a project fee |
The 12-Month Test for an AI Recruiting Automation Company
A demo tells you what an agent can do. A year of use tells you what a team actually kept. Kindgeek's recruitment pipeline now runs on four agents nobody has to be reminded to use, built around the tools the team already had, with two other agents shelved on purpose once they didn't hold up under real hiring conditions.
Mariia Dykun, Head of People Operations at Kindgeek, doesn't treat any of this as finished business. Asked whether recruiters will still exist in five years, she didn't give a clean yes:
“Tough. But we know that you should learn to use AI tools, because if not, someone else will learn that and will be better in 5 years than you.”
That's the standard worth applying for before signing with any recruiting vendor: not a polished pilot, but a reference that's still running twelve months in, honest about what didn't make the cut.
Proof point | What it shows |
|---|---|
Agents still running after 12 months | The workflow created retained value |
Agents decommissioned honestly | The vendor tests fit, not just demos |
Time saved per week | ROI can be defended to CFO |
ATS and calendar integration | The system fits existing work |
Human review stays in place | Hiring judgment is not fully outsourced |
Bias and candidate disclosure are addressed | The system can survive real recruiting governance |
Posted by
Kateryna Shykula
Content Producer
What is an AI recruiting automation company?
It's a firm that builds custom AI agents for specific recruiting tasks, such as CV screening, interview analysis, or onboarding documentation, and integrates them directly into a company's existing ATS and tools. Unlike AI recruiting software, it's typically priced as a fixed-scope project rather than a per-seat subscription, and ownership of the resulting system transfers to the client.
Does an AI hiring automation firm replace recruiters?
Not based on how Kindgeek uses it. The agents took over specific, repetitive tasks such as CV formatting, document drafting, and transcript summarizing. Sourcing quality, candidate judgement, and final hiring decisions stayed with the recruitment team throughout, and Dykun is explicit that the agents are a helper, not a replacement for a recruiter's own read of a candidate.