This is a machine-assisted translation of the original Dutch version.
Bad Decisions Made from Data
Imagine buying a new machine or starting a big project because you trusted a set of numbers, only to find out those numbers were wrong. Here are three examples I saw this past year:
- A "faster" process with two extra steps. A process with two extra steps was measured 5 seconds faster than the same process without them.
- Ignoring hidden costs. A theoretical saving on how you load containers forgot that storing and reloading containers costs much more in Europe than in Asia.
- Over-investing after COVID-19. Companies forecast by extrapolating the previous year's figures, so they kept buying capacity they no longer needed.
When this happens, you waste money, trust, and human brainpower.
So, how do you stop numbers from lying to you? How do you make sure your analysts make smart choices? Here is how to keep things honest and sensible.
What Managers Should Do
1. Measure what the customer actually sees
Your internal delivery scores often lie. If you only measure "shipped on time", you miss all the orders that were supposed to leave that day but didn't.
Also, measuring against a "promised date" instead of the "date the customer asked for" gives you an incomplete picture.
Ask your customers how they measure your performance. Compare their numbers with yours. Don't argue with them. Figure out why the numbers are different and how you can do better.
2. Demand simple processes
A simple process is one that anyone can understand just by looking at it or drawing it on paper. You can instantly see who the customer is, who the supplier is, and where the real work happens.
Simple processes are easy to learn, easy to follow, and easy to measure. You can go straight to the factory floor or warehouse, observe what is happening, and figure out the problem yourself. You don't need complicated computer systems or fancy software tools.
Keep your processes simple. Do not buy expensive data analytics software if you aren't sure your raw data is reliable.
3. Think about the whole picture (End-to-End)
An analyst might try to save money on trucks coming into the warehouse because that is their specific job. But saving money in one spot might break something else further down the line.
Teach your team that an improvement is only good if it helps the whole company. Encourage them to talk to teams upstream (suppliers) and downstream (customers) to see the full impact, both good and bad. Make sure you know who is affected by your decisions. Spend time learning this when you join a team, and build a strong network across departments.
4. Situational Leadership: Guide your team based on what they need
When you hire a new analyst, don't just hand them a computer. In their first few weeks, have them actually do the physical job they will be analysing. Let them see how a work shift runs, where the bottlenecks are, and what goes wrong daily.
After that, adjust your leadership. Some people need help using spreadsheet tools; others are already experts. However, almost everyone needs help answering: "Are these numbers telling the truth?" or "Which part of this data actually matters?" Walk through their work together. Don't just tick off approvals blindly on "trust": build trust as they show their learning.
What Analysts Should Do
1. Start with the right data
If someone asks you how customers place orders, go find the dataset that shows the original customer orders. Look at how online shopping baskets are filled to understand what buyers really want.
Do not use historical shipping data to answer this question. Shipping data is distorted by warehouse stock levels, delivery schedules, and mistakes made along the way. Answer the exact question that was asked.
2. Check and clean your data
Getting a pile of data is easier than ever. Getting good data is just as hard as it has always been.
Look at your data with a fresh eye. You don't need a 40-page presentation showing how you cleaned out weird mistakes (outliers). Just clean them up and show one clear summary slide: (1) 3 big opportunities found in the exceptions (2) 20 data points removed because of a Monday team meeting.
Context matters: Do not copy-paste numbers without understanding them. Account for seasonal trends, strikes, supply delays, or major events like COVID-19. Check your findings with experts before sharing your final report.
3. Talk to experts (and non-experts)
Talk to people who know the job well:
Before you start: Ask them where their gut tells them the problem is, and how they would measure it.
While you work: Share early findings with an open mind. Do they see the same problems? Why or why not?
When you finish: Share your final summary and thank them for their help, both in person and on your slides.
Test with non-experts: Try explaining your analysis to someone outside your department (or even a family member). If they don't get it, your explanation is too complicated.
4. Trust yourself, but stay humble
You were hired because you are clever and good at what you do. Use your skills. You have a fresh pair of eyes, and it is your job to speak up if you spot something wrong.
At the same time, remember that you are still learning the business. Connect with people who have been doing the job for twenty years. Be humble, admit your mistakes quickly, and learn from them, but stay confident in your technical skills.
No More Spreading Lies
By following these eight rules, you take the lies out of your data and bring real life back into your decisions.
Here is what you win:
- You stop throwing money away on useless tech systems.
- You avoid projects that don't help people on the shop floor.
- Your analysts become trusted partners.
- Your business makes better profits, and everyone feels less stressed.
Remember: The biggest wins rarely come from heavy statistics. The main daily routine is usually already running okay. The real magic happens when you fix the weird exceptions and extra tasks that pop up. Measure how often those happen and shorten the time they take. That simple fix will make a huge difference that everyone can see.