Diagnose the system before you judge the person.

Watch any leadership team in the first ten minutes after something goes wrong and you will learn everything you need to know about their operating culture.

A campus rollout misses its go-live date by two weeks. A wrong number makes it onto slide four of a board pack. A production release takes down a payment page for forty minutes on a Tuesday afternoon.

In a weak organisation, the first question in the room is almost always the same: "Who dropped the ball?"

Someone's name is produced. A stern message is sent. A manager promises to "have a hard conversation" with the engineer or analyst whose fingers were last on the keyboard. Perhaps an extra sign-off signature is added to the checklist so everybody in the room can feel that decisive leadership has occurred.

Then everyone goes back to their desks, leaving the exact same broken process, the exact same ambiguous policy, and the exact same noisy data sitting quietly in the dark, waiting for the next person who sits in that chair.

Finding a culprit takes five minutes. It is theatrical, emotionally satisfying, and almost always lazy.

Across twenty years of engineering, product building, and enterprise transformation, one operating principle has saved me from more bad decisions than any other: diagnose the system before you judge the person.

Separating diagnosis from accountability is not about being soft, and it is not about making excuses for poor work. It is about respecting reality. Until you have examined the process, the policy, the tools, the data, and the incentives a person was working inside, you have no idea whether you are looking at an underperformer or a normal human being trapped in a bad design.

The 94 / 6 law: what Deming and aviation safety figured out decades ago

The idea that most workplace errors are built into the environment is not a piece of modern HR sentiment. It comes from the hardest-nosed industrial statisticians and safety engineers of the last century.

W. Edwards Deming, the statistician and quality pioneer whose methods helped rebuild post-war Japanese manufacturing, spent decades measuring defect rates across factories, supply chains, and administrative offices. Early on, he estimated that 85% of operational problems were caused by the system and 15% by the worker. Near the end of his life, after another twenty years of empirical data, he corrected himself. His first estimate had been too generous to management.

The real split, Deming wrote, was 94% system, 6% special individual cause.

When a factory produced defective parts, or a service organisation missed its cycle-time targets, more than nine times out of ten the root cause lived in management-owned variables: unclear specifications, poorly maintained tools, conflicting targets that rewarded volume over quality, or supplier inputs that varied wildly from batch to batch. Scolding or replacing the worker standing at the end of the line was statistically pointless, because the new worker would walk into the exact same 94% headwind on Monday morning.

Decades later, cognitive systems scientist Sidney Dekker extended this insight in his landmark work on aviation and clinical safety, "The Field Guide to Understanding 'Human Error.'"

Dekker contrasted two ways leaders look at failure:

• The "Bad Apple" View (The Old View). This view assumes that the organisation's processes and software are fundamentally sound, and that things only break because unreliable humans fail to pay attention, cut corners, or forget the rules. The remedy is always aimed at the person: name, shame, retrain, write a stricter circular, or fire.
• The Systems View (The New View). This view recognises that human error is never the cause of an incident; it is the symptom of deeper trouble inside the system. Complex organisations are full of competing pressures: move fast, keep costs down, keep customers happy, never break a compliance rule. Frontline staff spend their entire week bridging gaps between broken tools and contradictory policies. When a capable person makes a mistake, their action almost always made sense to them in that moment given the screen they were looking at, the clock ticking behind them, and the incentives management set.

Safety psychologist James Reason pictured this as slices of Swiss cheese. Every defence layer in an enterprise (written policy, intake validation, software alerts, supervisory review, and frontline execution) has holes in it. A major failure only happens when the holes in four upstream layers line up at the exact same moment, allowing a bad input to pass straight through to the last person in the chain.

If you fire the person at Slice 05 and leave Slices 01 through 04 untouched, you have not fixed the incident. You have just reset the timer.

Modern software engineering research proves this out on the balance sheet. Years of data from Google's DevOps Research and Assessment (DORA) programme across thousands of technology organisations show that elite performers consistently practice blameless, system-first post-mortems. Why? Because when you punish the engineer who triggered an outage, the rest of the team learns a dangerous lesson: hide near-misses, delay escalating risks, and never touch fragile legacy code. Psychological safety is not a perk; it is an early-warning telemetry system.

Three rooms where the "person" was not the problem

Over the years, I have walked into dozens of situations where leadership was ready to replace a team or reprimand an individual, only to find the real culprit hiding in the wiring.

1. The 2 a.m. refinery alarm that cried wolf

Early in my career as a process engineer working on refinery automation and optimisation, a night-shift operator missed a temperature and pressure excursion during a unit transition. By the time it was caught, off-spec product had run for nearly an hour.

The morning write-up was predictable: operator inattention during night shift.

Before accepting that conclusion, I sat at the operator's console during the next transition and watched what actually happened on the screens. Within twenty minutes, the real problem was obvious. A faulty level sensor on an adjacent vessel had been triggering a non-critical nuisance alarm every three to four minutes for over a fortnight. Across an eight-hour shift, the console chimed more than a hundred times for nothing.

To silence the buzzer, operators had developed a reflex: reach over, hit acknowledge, and keep working. Worse, the critical pressure trend that actually mattered during a grade switch was buried two sub-screens deep behind a menu that cleared every time the alarm banner popped up.

The operator was not lazy. We had trained his brain through two weeks of bad instrumentation to ignore the console, and we had hidden the one number he needed behind three clicks. We replaced the faulty sensor, cleaned up the alarm hierarchy so a chime meant genuine danger, and pinned the transition trend to the main display. The same operator ran that unit for the next year without a single miss.

2. The "slow" institutional reviewers

Years later, across a group of schools and higher-education institutions, we heard a constant complaint from the field: central reviewers were taking days, sometimes over a week, to clear routine fee and audit approvals. The casual verdict in the hallways was that the review desk had become a bureaucratic bottleneck staffed by people who were not moving fast enough.

We pulled a sample of delayed files and walked the timeline hour by hour.

The people sitting at those desks were working flat out. They were slow for two structural reasons that none of them had the authority to fix on their own:

• First, the written policy had a wide grey zone where it was unclear whether a local principal could approve a concession or whether it required central sign-off. Because nobody wanted to be accused of overstepping their authority in an audit six months later, everyone forwarded files upward for cover.
• Second, the submission form allowed applicants to send requests without attaching mandatory supporting documents. Reviewers were spending half their week acting as document chasers over email and phone calls.

Blaming the reviewers for five-day cycle times would have been like blaming a toll-booth operator for a traffic jam on a highway where three lanes merge into one without signs. Once we rewrote the policy thresholds, enforced complete evidence packets at submission, and added an AI verification layer to pre-check the numbers against the rulebook, those exact same reviewers cleared files in minutes.

3. When the physical centres closed overnight

The sharpest test of this principle came when I was Co-founder and CEO of a blended-learning business. We had grown from our first physical coaching centre to a network of centres joined to a digital learning platform.

Then the 2020 pandemic lockdown hit, and every physical centre shut its doors overnight.

In that first week, cash flow projections looked terrifying. Across the education sector, the standard playbook was immediate: furlough or lay off the physical centre staff, assume classroom teachers and centre coordinators cannot run a purely online operation, and hire a different digital crew later.

Had we judged our people by the collapse of the physical channel, we would have hollowed out the company.

Instead, we diagnosed what actually created value in our system. Parents and students did not pay us for classroom chairs and fluorescent lights; they paid for the teaching relationship, the academic rigour, and the trust our faculty and coordinators had built with families. The bottleneck was not the people; it was the delivery channel.

Within days, we shifted the entire operating rhythm onto our digital platform. We retrained and supported centre staff to run live online cohorts and student mentorship from home, offered fee relief to families whose livelihoods had been disrupted, raised debt capital on the credibility of our operating discipline, and retained every single employee without layoffs. The team responded with extraordinary ownership, kept operations near break-even through the worst of the crisis, accelerated new online revenue lines, and carried the business all the way through to a successful acquisition.

Does this mean nobody is ever accountable?

Whenever I explain this philosophy to a new leadership team, someone inevitably asks: "If we always look at the system first, does that mean nobody is ever held accountable for poor performance?"

The exact opposite is true.

In a sloppy organisation where processes are undocumented, targets are unrealistic, and data is unreliable, true accountability is impossible. Every time a project misses a deadline, the owner can reasonably point to three broken upstream dependencies, an ambiguous brief, or a last-minute priority change from above. Good performers get demoralised because their heroic effort is swallowed by bad plumbing, while genuine underperformers hide comfortably inside the noise.

When you clean up the system first, all the camouflage disappears.

If the goal is clear, the decision rights are written down, the upstream inputs are verified, and the coaching has been provided, then whether someone delivers or not becomes unmistakable. Diagnosing the system does not weaken accountability; it makes accountability fair, fast, and undeniable.

The five-layer diagnostic before you judge the person

Whenever a commitment is missed or an individual seems to be struggling, I require leaders to walk through five layers in strict order. You do not move to Layer 5 until Layers 1 through 4 have passed inspection.

Layer 01: Clarity and realism of the commitment

Start with the contract between leader and owner. Was there a single named owner, or was responsibility split across a committee where "everyone" owned it? Was the deadline set by working backward from an engineering and operating reality, or was a date announced in a meeting without checking the calendar? And did both people agree in writing on what "done" actually looked like? Half of all "missed deliveries" are actually unexamined wishful thinking from Day 1.

Layer 02: Policy contradictions and perverse incentives

Look at what the system actually rewards on a Tuesday afternoon. Kerr's classic management paper called this "The Folly of Rewarding A, While Hoping for B." Are you asking a sales team for clean CRM data while paying commission on unverified bookings? Are you asking an engineering lead for zero production bugs while judging them solely on how many feature tickets they close by Friday? And when a risk appeared two weeks ago, did your culture make it safe to raise a red flag early, or do you shoot the messenger who brings bad news to a status review?

Layer 03: Data, tools, and upstream dependencies

Walk the person's actual daily workflow. Are they spending four hours a day reconciling five spreadsheets because two internal systems do not share a common ID? Did their deliverable stall because another department sat on an API key or a legal review for ten days with no escalation SLA? If you put a Formula 1 driver into a car with three flat tyres and a clogged fuel line, changing the driver will not improve your lap time.

Layer 04: Skill, confidence, and coaching support

Ask yourself an uncomfortable question as their manager: have I actually taught this person how to solve this class of problem? Moving from senior individual contributor to manager, or from running a local project to leading a cross-business transformation, requires a completely different mental toolkit. Have you sat with them on the first two drafts, shown them how to frame options and trade-offs, and given them air cover with senior stakeholders? If there is a skill or confidence gap that you have never coached, the failure is on your calendar before it is on theirs.

Layer 05: Individual ownership, behaviour, and will

Sometimes you walk through Layers 01 to 04 and find that the brief was crystal clear, the timeline was fair, the tools worked, and you coached the gap closely, and the work is still careless, late, or defensive.

Now, and only now, you are looking at the 6%.

Because you did the diagnostic homework first, you do not need to raise your voice or guess at motives. You can sit down with documented evidence and run a clean, respectful sequence: set a time-bound improvement plan with unambiguous milestones; if the role is simply the wrong shape for their strengths, redesign the scope or reassign them to where they can succeed; and if ownership and behaviour still do not move, make a clean exit decision without dragging it out for a year.

The mark of a mature leader

It takes very little skill to walk into a room after a failure, point at the person closest to the wreckage, and demand why they let it happen. Anyone with a title can do that.

It takes discipline to pause the room, walk upstream past the blinking alarm, and ask what part of the policy, the intake, the tooling, or your own leadership set that person up to fail.

Do that consistently and two things happen across your organisation. First, people stop hiding problems from you; you hear about the hairline crack on Tuesday morning instead of the collapsed bridge on Friday night. Second, every mistake makes the machine permanently stronger, because you fix the pipe instead of just mopping the floor.

Separate diagnosis from accountability. Fix the system first. The people inside it will surprise you with what they can carry when the road underneath them is finally paved.

Blaming a person takes five minutes and changes nothing. Fixing the system takes a week and prevents the next hundred mistakes.