We’re Collecting More Data Than Ever - So Why Are We Still Making Bad Decisions?

We have more data than at any point in human history. Sensors report temperatures every few seconds, websites record every click, finance teams can summon a dashboard before the coffee has gone cold, and somewhere a smart toaster is probably preparing a quarterly report on browning performance.

Yet organisations still make spectacularly poor decisions. Not occasionally, either. They make them with confidence, colour-coded charts, and a steering group called something like the Data Excellence Council.

The problem is not that data is useless. Far from it. I have spent much of my working life building temperature monitoring and IoT systems, and I am firmly in favour of knowing whether a freezer containing expensive stock is quietly warming itself into a very costly apology. Good data prevents waste, protects patients, and gives engineers a fighting chance of finding a fault before it becomes an incident report.

But collecting data and using it well are entirely different skills. One is often a procurement exercise. The other requires judgement, curiosity, context, and occasionally the uncomfortable admission that the spreadsheet is answering the wrong question.

We measure what is easy, not what matters

Most systems begin with an understandable question: what can we measure? Temperature, humidity, battery level, device connectivity, dwell time, clicks, throughput. These are useful because sensors and software can capture them neatly.

The harder question is: what decision will this measurement improve?

In temperature monitoring, a fridge temperature graph can look wonderfully reassuring. A smooth line, a tidy daily pattern, perhaps a tasteful green band. But if the probe is in the wrong place, poorly calibrated, or measuring air while the product temperature is what matters, the graph is merely decorative. It is a very expensive screensaver.

I have seen teams become fixated on sensor uptime while missing the practical question of whether staff could respond to an alarm at 2am. Ninety-nine point nine percent connectivity sounds impressive until the missing 0.1 percent contains the one alarm that mattered. Metrics are not reality. They are a deliberately incomplete representation of reality, which is less catchy but rather important.

The same mistake appears everywhere. Customer support measures call handling time, then wonders why customers feel rushed. A delivery operation measures stops per hour, then acts surprised when drivers take shortcuts. A hospital measures bed occupancy, then discovers that a full hospital is not necessarily an efficient one. It is often simply a hospital with nowhere useful left to put anyone.

Context is the bit dashboards tend to misplace

Data tells us what happened. It rarely tells us why without help.

A temperature excursion may be a failing compressor, a door left open during a busy delivery, a defrost cycle, a cleaner unplugging the unit to reach a socket, or a sensor placed beside an enthusiastic fan. The same line on a chart can mean five entirely different things.

This is why I am wary of the modern habit of treating dashboards as decision engines. Dashboards are good at making information visible. They are not good at understanding a site, a process, or the human beings trying to keep both functioning on a wet Tuesday afternoon.

If you want better decisions, put the data beside operational knowledge. Ask the person who works there. Look at the equipment. Read the alarm history alongside maintenance records. Find out whether the apparent trend began after a software update, a layout change, or Barry from facilities relocating a sensor because it was "in the way". Barry may be right, of course. He is often closer to reality than the boardroom.

For disabled people, missing context can be particularly damaging. Organisations may collect accessibility feedback, complaint categories, website completion rates, and call-centre statistics, then declare a service usable because the average journey time has improved. Meanwhile, someone using a screen reader cannot complete the form at all, or a customer cannot reach the only available support channel without making a phone call.

An average can be a remarkably efficient way to hide an exclusion. If ten people stroll through a process in two minutes and one person is blocked completely, the dashboard may still look splendid. The person excluded is not reassured by the mean.

More data can make us less decisive

There is a comforting belief that if we collect enough information, uncertainty will politely leave the building. It does not. It simply gets a larger storage allocation.

When every team has dozens of metrics, decisions can become a negotiation between competing charts. Nobody wants to be wrong, so the organisation asks for another report, then another segmentation, then perhaps an AI summary of the reports nobody has had time to read. This feels rigorous. It is often procrastination wearing a lanyard.

I am not anti-AI here. Used properly, it can find patterns, reduce manual work, and help teams sift large volumes of information. But it cannot rescue vague objectives or poor-quality inputs. An AI model trained on confused measures will produce confusion at scale, which is efficient in the same way as a faster photocopier is efficient when you are copying the wrong document.

Better decision-making needs a willingness to define thresholds in advance. What temperature demands immediate action? What level of customer friction is unacceptable? What evidence would make us change course? If nobody can answer those questions before the graph arrives, the data is likely to become ammunition rather than insight.

Build a smaller, more useful truth

The best monitoring systems I have worked on were not necessarily the ones with the most sensors or the flashiest dashboards. They were the systems where everyone understood three things: what was being measured, why it mattered, and what they were expected to do next.

That means choosing a small number of measures tied to real decisions. It means checking data quality, including calibration, placement, gaps, timestamps, and boring-but-essential details such as whether devices are actually reporting in the correct time zone. It means combining quantitative evidence with observation and lived experience.

Most importantly, it means treating data as a prompt for conversation, not a substitute for it. Ask what the numbers cannot tell you. Ask who is missing from the dataset. Ask whether the target is changing behaviour in a useful direction or merely encouraging people to become creative with definitions.

We do not need less data because data is the problem. We need less reverence for it. The useful organisation is not the one with the largest data lake, where facts go to float about looking important. It is the one that can connect a reliable signal to a sensible action, listen when reality disagrees, and adjust before a minor warning becomes a major mess.

That is not glamorous, but then neither is replacing spoiled stock at dawn. I know which one I would rather spend my morning doing.