Early on at Simply Noted, I tried to fully automate our new hire onboarding. We built out an email drip sequence, auto-generated training docs, and even set up a bot to answer common questions. On paper it was clean. In practice, new hires felt like they were being processed, not welcomed.

Our retention in the first 90 days actually got worse. People would show up, get dumped into a workflow, and never feel connected to the team. The turning point was when one new employee told me, “I didn’t know who to call when I had a problem because everything came from a system.”

We pulled back and rebuilt onboarding as a hybrid. The logistics, paperwork, scheduling, and compliance stuff stays automated. But every new hire gets a handwritten welcome note from me on day one, a personal call with their direct manager, and a human check-in at 30 and 60 days. That was the lesson. Automation is incredible for process. But it falls flat for moments that need to feel personal.

I run Simply Noted, where we use patented robotics to create real handwritten notes at scale. So I live and breathe automation every day. But I also know exactly where the line is. When something is supposed to make a person feel valued, a workflow will never replace a human being on the other side of it.

“The biggest mistake I made with hiring automation was trusting it to make judgment calls instead of simply improving efficiency.” One of the biggest misses I experienced was trying to automate early-stage candidate screening too aggressively. On paper, it looked efficient. We built workflows that filtered resumes based on keywords, years of experience, and predefined qualifications. The process was certainly faster, but it quietly eliminated people who would have been outstanding hires.

I still remember reviewing a rejected application from a candidate who had taken an unconventional career path after leaving the military. The automation flagged the resume because the job titles didn’t match our search criteria. When I looked at it personally, the leadership experience, problem-solving ability, and adaptability were exactly what our client needed. That candidate ended up becoming one of the strongest performers on the team.

Since then, I’ve treated automation as a tool for reducing administrative work, not replacing human evaluation. Technology is excellent at organizing information, scheduling interviews, and handling repetitive tasks. It struggles when context, potential, or transferable skills matter.

One thing I’ve learned over the years is that “The best hiring decisions are rarely made by algorithms alone; they’re made when technology creates time for people to make better judgments.” Automation should help recruiters ask better questions, not decide who deserves the opportunity to answer them.

If a process removes human curiosity from hiring, it’s probably automating the wrong thing. That’s the lesson I wish I’d understood much earlier.

My smart speaker has misheard me for 2 years and I keep it around because the misses are cheap. We made the same bet on a bot that answered HR questions off our handbook across a team of around 60 working fully remote. The easy stuff about leave balances and public holidays it got right. Nobody messages HR about the easy stuff though.

The questions people actually brought were the tangled ones, like a parent falling sick and nobody sure what counts as leave. A confident wrong answer there costs you more than the hours it ever saved. We moved it back to the simple queue and let a person take the rest. I still can’t tell if that line holds, or if it just needs to learn when to stop talking.

One HR process we tried to automate was the exit process, and honestly, we were a bit disappointed at the results. All the practical bits worked well, like collecting paperwork, making sure equipment came back, and closing off access. But the part we could not replace was the exit interview.

When someone is leaving, they often say things they have been holding in for months. Sometimes it is about workload. Sometimes it is about a manager. Sometimes it is just a small culture issue that has been bothering more people than you realise.

A form can tell you why someone clicked “other.” But it can’t ask the unique questions needed to get to the root of the issue.

So we learned to automate the admin, but we stick with the actual in-person exit interview. Use the tools to keep the process flowing, but do not strip out the one part where useful feedback that can reduce turnover becomes clear.

The worst HR automation failure is in the context of high-volume recruiting processes where there’s immediate SMS outreach automation attached to applying for jobs.

In an example with a restaurant brand and their hiring process, they set up an automated workflow to immediately SMS anyone who applied via their job application form. The problem? They didn’t anticipate bad bots. Their careers website got attacked by sophisticated automated bots that filled out the application form with scraped/stolen consumer data.

The HR automation acted exactly as designed and immediately SMSed hundreds of phone numbers. The candidate pipeline metric jumped from ~50 to over 400 applicants in a matter of hours. The problem? There were zero genuine prospects. Not only was the recruiting marketing budget wasted, but it created SEVERE legal risk.

When an automated HR system SMSes a real consumer with stolen data, that consumer never gave consent to be contacted. This creates direct consumer-protection-type violations of laws like the TCPA (Telephone Consumer Protection Act), which carry penalties of between $500 and $1500 PER INCIDENT.

So imagine if a sophisticated bot farm submits 1000s of fake applications, how much liability did your totally unprotected HR automation tool just create? From what I’ve seen across the industry, the key learning here is that you can’t automate candidate outreach without first automating filters for applicant traffic.

Before the HR system sends any automated texts, the submissions to the form need to be filtered by machine learning-based bot detection to screen out the sophisticated invalid traffic (SIVT).

The point is, it’s more complex than HR folks realize – recent tests run by Veracity Trust Network’s Website Threat Protection platform have shown that typically 38% of web traffic is suspicious. If you aren’t applying strict vetting and filtration of traffic at the top of the applicant funnel, then your automated HR workflows will only accelerate costly mistakes that damage your brand.

Overdoing it with automated systems in Human Resources (HR) can lead to problems, particularly when it comes to matching up mentors and mentees. The assumption is that simply having both parties fill out their profiles will allow the automated system to complete the rest of the matching process, but matching is not simply about data; two individuals may be very similar on paper, yet have different styles of communicating, different availability, different goals, and different expectations, which could lead them to not be “compatible” with each other as mentor/mentee.

The lesson learned from these experiences is that automation can be utilized to limit the number of options but must also have some form of human element to verify the matching process. For a mentoring program, the best approach is to automate processes that help reduce the administrative burden associated with keeping track of potential matches, and to provide mentoring program administrators with the capability of identifying potential strong relationships and overseeing adjustments, but still have the ability to review and make any necessary adjustments. The answer to this issue lies in first identifying which parts of the process can be automated and determining which will require some level of critical thinking and evaluation in the process of matching mentor and mentee. The human element usually plays a greater role in the HR process than does a perfectly defined workflow.

An early attempt to automate employee training allocation based purely on role titles rather than verified skill levels did not deliver the expected outcomes. The system assumed that job designations accurately reflected capability, which led to mismatches between assigned learning paths and actual development needs. Engagement dropped, and training relevance weakened across several teams. Research from Deloitte has highlighted that skills-based workforce strategies outperform role-based models in adaptability and long-term performance, reinforcing the importance of accurate skills data in automation design. The key lesson was that automation in HR processes is only as effective as the quality and granularity of the underlying data. From the perspective of leading corporate training initiatives at Edstellar, sustainable automation requires continuous validation of skills intelligence rather than reliance on static organizational structures.

One automation I tried and failed at was using an AI meeting assistant to take notes during meetings, check-ins and coaching calls. Because it joined as a visible participant, it made people feel like they were being watched, and conversations lost a lot of trust and openness. It also made people feel like their information was being recorded and not kept private anymore. In theory, it sounds great. In practice, a dismal failure compared to previous methods. A tradeoff I made was switching to a bot-free system to take meeting notes. It stays invisible during the process, so I still get the AI summaries and action items, but without the awkward “bot joined the conversation” moment at the beginning of every meeting.

One HR process I attempted to automate was employee performance reviews. We created a system that evaluated employees using objective metrics such as task completion rate, meeting deadlines, work quality based on defect or revision rates, and attendance. I believed this would save managers time while making evaluations more consistent and objective.

The results, however, were not what I expected. Many employees received poorer evaluations than they deserved because the system could only interpret the numbers, not the circumstances behind them. An employee who missed deadlines, for example, might have been mentoring new team members. Those contributions were valuable but invisible to the automated review.

That experience taught me that performance reviews cannot rely solely on objective metrics. While data is useful for highlighting trends and identifying potential concerns, it cannot capture qualities such as leadership, collaboration, etc.

Since then, I have come to view automation as a tool to support managers rather than replace them, and that is what I would advise other organizations to do as well.

We once tried automating leave approvals through a simple app based system, expecting it to save time for both artisans and office staff. Within the first month, adoption stayed low at just 27.4%, since many artisans were more comfortable asking a supervisor directly rather than typing a request into an app. Leave delays actually increased slightly, since some requests sat unanswered in the system while supervisors still expected verbal confirmations separately. We eventually reverted artisans back to verbal or WhatsApp based requests, while keeping the automated system only for office staff, where adoption reached 91.6%. The lesson I learned was that automation only works when it matches how people already communicate, not when it forces a new habit onto a team that values personal interaction. Convenience for one group can quietly become friction for another if rolled out without checking comfort levels first.

Another automation that was not well-received by operations staff was assigning tasks with a lack of defined workflow or ownership rules. The system would send automated reminders and assignment prompts but, in many cases, was unable to provide clarification to staff on who was responsible for what, leading to inefficiencies and duplicated effort.

The point is: automation requires clearly defined workflow before any technology is introduced. Clarify who owns each task, when the task needs to be handed off, and which steps absolutely require a human decision to ensure staff (whether internal or virtual) remain aligned with your business’s objectives and that automation does not work against, but rather for, operational workflow.

An automation project around employee engagement surveys produced misleading results. Pulse surveys were sent automatically after milestones and at regular intervals, which increased response volume, but the insights became shallow. People started treating the questions like routine notifications, and response patterns reflected survey fatigue more than real sentiment. The dashboard showed trends, yet managers had less confidence in what those trends actually meant because the context behind each answer was missing.

The lesson we learned was to automate collection, not interpretation. Keep surveys shorter, use them less often, and follow notable shifts with live conversations or small group discussions. Reliable people data depends on trust and timing, and both weaken when every signal is reduced to another automated touchpoint.

The process we were trying to automate was the collection of performance reviews. The diagnosis of the problem may have taken longer than normal, as all the key metrics appeared to be just fine: response rate was rising, completion time was falling, and it looked like the automation was actually working based on all the figures shown by the dashboard.

What had happened in reality was that people became good at going through the process instead of using it. The friction we eliminated was not inefficiency; it was the friction that triggered thought before making an input. When the system made the process of getting feedback easy, it made it shallow at the same time, and we wasted two cycles on making apparently informed decisions.

The main point I want to make in this regard is that not every human process involves friction that should be eliminated. Sometimes friction is the aim of the process.

I have seen issues with HR process automation through onboarding task routing. The intent was for new employees to input into the system to generate a trigger for setting up their account, ordering equipment needed, sending training checklists to the proper individual, and prompting the manager with reminders. This all looks good on paper, but the actual process encountered many challenges due to exception handling. Contractors, part-time employees, remote employees, and all employees who changed roles were routed differently and placed on the incorrect checklist.

The lesson learned is that you must have all exceptions mapped before you automate. Automate the most common path before moving on to document areas where human intervention is required. Automation works best for repetitive type processes such as reminders, forms, and status update processes, but when all employees or positions are on the same workflow, it may not work as well. I prefer to initiate smaller trusted automations than larger automations that will cause the team to continually have to address issues.