The salary is the visible part. It’s also the small part.

The arithmetic everyone does

Studies put the cost of a bad hire anywhere from several months’ salary to multiples of the annual package, depending on seniority and how it’s counted. The exact multiplier matters less than the categories, so let’s do the honest arithmetic for a small business, using a simple example: a mid-level hire who doesn’t work out and leaves (or is let go) after six months.

The direct costs:

• Six months of salary, taxes and benefits for underperformance you paid full price for.
• Recruitment costs, twice. The ad, the hours screening and interviewing, any agency fee. Everything you spent finding this person, you now spend again finding their replacement.
• Onboarding and training time, the weeks your team invested in getting them up to speed, now written off.

For an SME, this alone typically lands somewhere around a year’s salary in total impact. Painful, but survivable. The trouble is that the direct costs are rarely the biggest line.

The costs that don’t show up in a spreadsheet

Team drag. In a ten-person company, one person is 10% of the workforce. Colleagues cover the gaps, quality dips, and, the truly expensive part, your best people notice. Nothing demoralises a strong performer like watching underperformance be tolerated because “we just hired them”. Some of the worst bad-hire outcomes aren’t the person who left; they’re the good person who left because of them.

Customer exposure. In a small company, almost everyone touches customers. A bad hire in a client-facing seat doesn’t just underperform internally, they underperform in front of the people who pay you. A damaged client relationship can outlast the employment by years.

Founder attention. The scarcest resource in any SME is the attention of whoever runs it. A struggling hire consumes it voraciously: extra check-ins, redone work, difficult conversations, the slow decision about whether to let go. Every one of those hours came out of sales, product, or strategy. This is the cost that never appears anywhere, and it’s often the largest.

The opportunity cost of the seat. For six months, the role was filled but not done. Whatever that role was supposed to unlock, the sales pipeline, the delivery capacity, the product progress, didn’t happen. You paid for the seat and lost the output.

Why SMEs carry more of this risk, with less process

Here’s the uncomfortable asymmetry. A corporation with 5,000 employees absorbs a bad hire statistically. An SME absorbs it personally: 10% of the team, a real client, the founder’s quarter.

And yet it’s the corporation that has the structured interviews, the scorecards, the dedicated recruiters, and the SME that hires under pressure, from a pile of CVs nobody had time to read properly, in an interview squeezed between two other jobs.

The companies with the least margin for hiring error are the ones making decisions with the least process. That’s not a criticism of SME owners; it’s a time problem. But it is fixable.

Reducing the risk without a recruitment department

The risk of a bad hire is highest at two moments, and both can be defended cheaply:

1. The screening. When the pile is too big to read properly, good candidates get missed and the shortlist forms by luck and fatigue, which means the interview stage starts from a weakened field. Fixing this is a process problem (we’ve written a practical screening framework) and, at volume, a tooling problem, the part our honest guide to AI in recruitment walks through in full.
2. The decision. Most bad hires announce themselves before the contract is signed: in the urgency to fill the seat, in interviews where you talked more than you listened, in the feeling of having to convince yourself. We’ll publish a piece on exactly those warning signs soon.

A bad hire is one of the most expensive mistakes a small business can make, and one of the few expensive mistakes that improves dramatically with twenty minutes of preparation and an honest process.

The cheapest bad hire is the one you didn’t make.

The first evening, they read carefully. Ten CVs, proper notes. The second evening, twenty more, faster. By the weekend, the remaining seventy get a ten-second glance each, and whoever applied last week is competing against a tired brain, not against other candidates.

Nobody designed this. It’s what happens when attention runs out before the applications do. The good news: most of the damage is avoidable with process, not heroics. Here’s the framework.

1. Define “good” before you open a single CV

The most expensive screening mistake happens before screening starts: reading applications with no written criteria. Without them, the first impressive CV becomes the benchmark, and everyone after is compared to that person instead of to the role.

Before opening the pile, write down:

• 3 to 5 must-haves. Things the job genuinely cannot be done without. Be brutal here: every “must-have” that’s really a nice-to-have silently filters out good people.
• 3 to 5 nice-to-haves. Real differentiators, weighted less.
• 1 to 2 disqualifiers. Hard constraints (location, language, legal requirements).

This takes twenty minutes and changes everything after: screening becomes checking against a list instead of forming an impression, which is faster, fairer, and far more consistent at CV number 80.

2. Do a two-pass read, not one deep read

Reading every application deeply, in arrival order, is how the queue-position lottery happens. Do two passes instead:

• Pass one, sort, don’t judge (30 to 60 seconds each). Against your must-haves only: yes, maybe, no. No notes, no agonising. The goal is to shrink the pile, not to rank it.
• Pass two, read properly (the “yes” pile, then “maybe” if needed). Now give real attention to a pile that’s a fraction of the size, with your criteria next to you.

Two passes feel slower. They’re dramatically faster, and every application gets looked at against the same bar, whether it arrived first or last.

3. Batch the work; don’t graze on it

Screening in stolen five-minute gaps between other tasks is the worst of both worlds: it feels constant and produces inconsistent judgements. Block two or three fixed sessions in the calendar and do nothing else in them. At SME volume, pass one over a hundred applications is a single afternoon, when it’s actually an afternoon and not three weeks of grazing.

4. Reply fast, especially to the “no” pile

Speed is part of screening, not an afterthought. The strongest candidates are in other processes and are gone within days, so a slow reply is a rejection you didn’t mean to send. And every candidate you ignore tells their network about it; in an SME’s local market, that reputation compounds fast.

A same-week acknowledgement to everyone and a clear “no, thank you” to pass-one rejections cost minutes with templates, and put you ahead of most companies competing with you for the same people.

5. Keep score, or fatigue keeps it for you

For the shortlist, use the same simple scorecard per candidate: your criteria, a score, one line of evidence each. It keeps the comparison honest when interviews are spread across two weeks, and it’s the only defence against the classic trap: hiring the most recent good candidate instead of the best one.

Where software fits (and where it doesn’t)

Everything above works with a spreadsheet and discipline. What software, including, yes, ours, changes is the economics of pass one: reading everything without fatigue, sorting against your criteria, and showing the reasoning so you can check it. The judgement in pass two, and the decision itself, stay with you. That division of labour, machine reads and human decides, is the principle behind our honest guide to AI in recruitment for SMEs.

But tool or no tool, the principle stands: a candidate’s chances should depend on their fit, not on where they landed in your inbox.

If you run or hire for a small business, you’ve heard both stories about AI in recruitment.

In one, AI reads a thousand CVs in a minute, removes human bias, and hands you the perfect candidate. In the other, it’s a black box that rejects good people for bad reasons and turns hiring into a lottery run by a machine.

Both stories are wrong, and both are sold hard. So here’s the guide I wish more SMEs had before signing anything: what AI genuinely does well in hiring, what it should never do, and how to tell serious tools from shortcuts.

Full disclosure: I co-founded Venora, an AI-assisted recruitment platform for SMEs. You should read this guide with that in mind, and notice that it will argue against several things AI vendors love to promise.

What AI genuinely does well

There is one part of hiring where software is honestly better than people: the first read at volume.

• It reads everything, without fatigue. A human reviewer gives full attention to CV number 8 and a skim to number 80. A machine gives the 200th application the same read as the 1st. At volume, consistency beats brilliance.
• It organises the pile. Sorting applications by likely fit means your limited attention goes first where it’s most useful, instead of being spent in whatever order the inbox dictates.
• It can show its reasoning. A well-built system explains why a candidate looks like a match: which requirements they meet, where the evidence is, what’s uncertain. That turns screening from a gut-feel skim into something you can inspect and challenge.
• It buys back time. For an SME, this is the real product. The hours spent on first-pass reading are hours not spent interviewing, checking references, or running the business.

Notice what all of these have in common: they happen before the decision.

What AI should never do

Hiring is a decision made with context, risk and judgement, whether this person will thrive in this team at this moment. AI has none of those things. It has patterns.

When a system moves from reading to deciding, rejecting candidates on its own, producing a shortlist nobody reviews, two things break:

1. Failures become invisible. If the model undervalues career changers or anyone whose path doesn’t match past hires, you’ll never see those candidates, and never know you didn’t.
2. Nobody can answer “why”. Not to the rejected candidate, not to yourself when a hire fails, and not to a regulator. European law already classifies recruitment AI as high-risk for exactly this reason.

The rule of thumb we build by, and the one I’d give any SME: automate the reading, never the decision. I’ve made the longer version of that argument before: the right role for AI in hiring is the passenger seat, with a human at the wheel.

Do you even need AI?

An honest guide has to include this section. You probably don’t need AI in recruitment if:

• You receive a handful of applications per role and can genuinely read them all.
• You hire once or twice a year and the process works.
• Your bottleneck isn’t screening, it’s attracting candidates in the first place (that’s a job-ad and employer-brand problem; no screening tool fixes it).

You probably do benefit if applications arrive in the dozens or hundreds, hiring is one hat among many for whoever handles it, and good candidates are slipping through because replies go out late or the pile never gets fully read.

Five questions to ask any vendor (including us)

1. Does the system explain why it recommends each candidate? If the answer involves the word “proprietary” and no screenshot, walk away.
2. Can a human always override it, and does the final decision sit with a person by design?
3. What data does it learn from? If it learns from your past hiring, ask how it avoids inheriting your past biases.
4. What happens to candidate data? Where it’s stored, how long, and under which legal basis. GDPR applies to candidates too.
5. What does it look like when the system is wrong? Every vendor has a demo of success. Ask for the failure mode. Serious teams have an answer; the rest have marketing.

The bottom line

AI won’t choose your next hire, and you shouldn’t want it to. What it can do, done right, is make sure every application actually gets read, show you its reasoning, and hand the decision back to you with the noise removed and the hours returned.

That’s the standard to hold any tool to. It’s the one we hold ourselves to.

I have spent more than twenty years watching technology arrive in places people swore it never would. Payroll, accounting, customer service, logistics. Each time the pattern repeats. First disbelief, then a rush, then the slow and honest work of deciding what the machine should really do and what has to stay with us. Recruitment is living that moment right now, and I want to be clear about where I stand.

HR is having its inflection point

For decades, hiring has been stubbornly manual. A recruiter opens an inbox with three hundred CVs for a single role and starts reading, one document at a time, late into the evening. By CV number ninety, attention drifts. Strong candidates slip through, not because anyone is careless, but because human attention is finite and the volume is inhuman. This is exactly the kind of problem AI was built to help with.

Modern models can read every CV against the real requirements of a role, in minutes, without getting tired at number ninety. They can surface the candidate the recruiter would have missed, explain why someone fits, and turn hours of screening into minutes of review. That is not science fiction. At Venora, it is what we ship today.

The driver-assistance analogy

Here is the part I care about most, and the reason I keep reaching for a comparison from a very different industry: driving.

When advanced driver assistance first appeared, it did not replace the driver. It watched the lane, kept a safe distance, braked when you were distracted, and warned you about the car in your blind spot. It made good drivers safer and long journeys easier. What it did not do, what it still does not do responsibly, is take full ownership of the decision. Your hands stay on the wheel. Your eyes stay on the road. When something matters, you decide.

AI in recruitment is at exactly that stage. It is assistance of the highest quality: it sees more than you can, it never tires, it flags what you might miss. But it is not the one who should sign the offer. Hiring decisions change people’s lives, and they carry context a model simply cannot see: the team that needs a certain kind of energy, the candidate whose unusual path is precisely their strength, the judgement earned over a hundred interviews.

Why human validation is not optional today

There are three reasons I insist on human validation, and none of them are nostalgia.

• Fairness. Models learn from data, and data carries the bias of the world that produced it. A human in the loop is how you catch the moment the machine quietly penalises the wrong thing.
• Accountability. When you advance or reject a candidate, someone has to be able to stand behind that choice and explain it. A black box cannot be held accountable; a person can.
• The law increasingly agrees with me. Under the EU AI Act, recruitment is a high-risk use of AI, and meaningful human oversight is not a nice-to-have; it is a requirement.

We did not bolt human oversight on to satisfy regulators. It has been the core of our product from day one.

The goal was never to remove the recruiter from the room. It was to give the recruiter a room that scales.

What AI should own, and what it should not

I am not arguing for caution as an excuse to do less. The opposite. AI should aggressively own the parts of hiring that exhaust people and add no human value: reading every CV, matching against requirements, ranking, comparing, drafting, scheduling, remembering. Give all of that to the machine, and give it gladly.

What stays with us is judgement. The final call. The conversation. The decision to take a chance on someone. The machine proposes; the human disposes. When AI does the reading and the recruiter does the deciding, both are doing what they are best at, and the candidate gets a fairer, faster, more human process than either could deliver alone.

Where this goes next

Will the balance shift over time? Of course. Driver assistance did not stand still, and neither will we. As these systems earn trust, through transparency, through being right in ways you can verify, through years of people checking their work, they will be handed more. That is how trust is supposed to work: earned in increments, not demanded up front.

But I do not believe the destination is a fully autonomous hiring machine that decides who works where while people watch from the sidelines. I am not even sure that world would be a better one. The future I am building toward is quieter and, I think, better: every recruiter with a tireless, transparent copilot that does the heavy lifting and always, always, leaves the final word to a person.

That is the future of HR and AI as I see it. Not replacement. Assistance, with a human at the wheel.