Implementation
Your AI hiring strategy is only as good as the exceptions
Everyone designs the happy path. Whether AI screening scales depends on what happens to the candidates who fall off it.
November 19, 2026
Simon Ovadia

The short version
Most AI hiring designs describe the candidate who applies, completes the AI screening and moves forward. But the real work of scaling sits with everyone else. What happens if a call drops, or a status isn't clear? What if regulation changes and the process has to adapt? Once recruitment automation takes over a step, the exceptions a recruiter used to absorb have to be designed on purpose.
From a high level, it's simple. A candidate applies, the AI screens or assesses them, and the recruiter moves the right people forward. On a slide, that's three boxes and two arrows. Shown on a call, everyone in the meeting nods, and rightly so, because that's what will happen to most candidates.
But it's the exceptions my clients write to me about once they go live.
The happy path
Step 1
Candidate applies
Step 2
AI screens or assesses
Step 3
Recruiter moves the right people forward
What happens when three candidates share one ATS status?
One professional services firm I work with went live across more than twenty countries at once. The assessments worked, the ATS integration worked, and results flowed back to recruiters as planned.
Then recruiters noticed that candidates who should have been distinguishable all looked the same in their ATS. The ATS status lumped them all together.
One ATS status. Three different candidates.
Knocked out by an eligibility question
For the recruiter
A clear no.
Let the deadline pass
For the recruiter
Might be a strong candidate who was on holiday.
Completed everything, scored below the threshold
For the recruiter
Made the effort and deserves a considered answer.
For a recruiter, those are three different situations. Without the distinction, the recruiter either treats all three the same or leaves the ATS to check each one by hand, which is exactly the manual work that was supposed to disappear.
Nothing broke. Every component did what it was built to do. But the client was right to raise each instance.
What if the candidate did nothing wrong in voice screening?
With voice screening, a call sometimes doesn't end the way it should. The connection drops, or there's a technical issue on either side.
We keep a separate status for this, distinct from the candidate who ran out of time and the candidate who didn't meet the bar. The distinction makes sense the moment you think about what happens without it.
Without a separate status, someone who lost signal on a train is recorded the same way as someone who failed.
With it, the recruiter knows this person simply needs another chance. It's a small design choice that makes a real difference to candidate experience.
Not every exception can be designed from HQ
Some exceptions only become visible once you're in the market.
Channel
WhatsApp isn't everywhere
With an airline client, we proposed WhatsApp voice calls for screening. It's how people communicate. But WhatsApp calling doesn't work in every region they operate in, so we moved to a different calling channel there.
Regulation
Pay transparency changes the question
At a global technology company, an open question about salary expectations no longer fit EU pay transparency rules. It's being reworded into a yes or no: does the range in the job description work for you?
Works councils
Internal moves need a say first
When assessments are used for internal moves, the candidates are employees. In many European countries, that means the works council has a say before anything starts.
None of these are failures of the technology. They're things only the local team could have known.
What if the ATS itself changes?
One of my clients approved a rollout into two new regions this year. The start date now depends less on us, or on their TA team, than on when their migration to a new ATS is completed.
Here the exception isn't a candidate at all. An ATS integration can't be treated as a milestone you tick off once. The systems around the AI keep changing, and the process has to survive that.
Why do exceptions matter more with AI?
Before AI, recruiters handled exceptions without anyone necessarily taking note. They made a phone call, gave someone an extra day, or left a note for the hiring manager. Their judgment was enough, and nobody had to design a process for every single situation.
With recruitment automation, that quiet judgment disappears unless it's put back in on purpose. This matters most in high-volume recruiting, where an AI agent handles thousands of candidates at once and a small gap is repeated thousands of times.
AI hiring go-live checklist: what I'd do before launch
For the people side of go-live, see Fabi's piece on change management for AI hiring.
I still like that high-level slide with the boxes and arrows. But it's just as important not to lose sight of the trees in the forest.
Key takeaways
- The happy path is where most candidates go, but the exceptions are where recruiters decide whether they trust the process.
- Many exceptions are local: calling channels, regulation and works councils can't be designed centrally.
- Recruitment automation removes the informal judgment that used to handle edge cases, so it has to be designed back in.
Want AI screening that's built for the exceptions too?
See how Maki handles AI screening at scale, from dropped calls to ATS statuses recruiters can act on.
Request a demoFAQ
What is the happy path in AI hiring?
The happy path is the journey most candidates take: they apply, complete the AI screening or assessment, and the recruiter moves the right people forward. Most AI hiring designs describe only this path, which leaves out the candidates whose journey doesn't go to plan.
What are exceptions in AI screening?
Exceptions are the cases that fall off the happy path: a voice screening call that drops, a missed deadline, an eligibility knock-out, a candidate who needs an adjustment, or a local rule that changes how a question can be asked. Recruiters used to handle these informally. With AI, each one needs a defined status and an owner.
Why should ATS statuses separate different candidate outcomes?
A candidate who was knocked out, one who missed the deadline and one who scored below the threshold each need a different response. If the ATS integration puts them all under one status, recruiters either treat them the same or check each one by hand.
How do you prepare an AI hiring process for go-live?
Ask "What happens when...?" until no one has new answers, give every exception a visible status and an owner in the recruiter's own system, and review exceptions regularly as data on where the process doesn't fit reality.
Page reviewed 19 November 2026.
See what Maki Agents can do for you



