Some nights (ok, most nights) when I get home from work, I like to turn on Netflix. One of the best things about Netflix is that in the brief, unsure moment when I’ve finished a series and don’t know what to watch next, Netflix matches me with a new show based on my tastes. Often, the Netflix algorithm makes great recommendations (*Narcos*), but sometimes it is way off the mark (*Friends*).

Netflix is by no means alone in using data to make predictions. Facebook uses behavioral data to predict what advertisements people will click. Google Maps uses traffic data to predict the quickest driving route. None of these algorithms are right all the time, but they’re right often enough to turn a handy profit.

This type of Silicon Valley prediction doesn’t work for everyone though. Say you work at a chemicals manufacturing plant that produces several tons of plastics per day. Just like Netflix, you may decide to use historical data to predict what process set-points you can change to decrease your costs. Maybe the algorithm suggests something like lowering the temperature during a particular reaction.

But what happens if the algorithm is wrong? For Netflix, the extreme worst case scenario is that I am so offended by their *Friends *recommendation, I cancel my account, losing Netflix $11 per month. But for your chemical company, one wrong prediction can be catastrophic. If lowering the temperature makes the plastic too unpliable and your factory’s yield plummets, you may lose hundreds of thousands of dollars and put client relationships at risk.

At Fero, we understand that your company deserves better than the Netflix treatment. That’s why with every prediction we make, we use some of the most cutting-edge developments in mathematics to provide a *confidence interval*. A confidence interval is a range of expected values based on how confident we feel about our estimate. We can see confidence intervals in action when we compare how Netflix and Fero-style algorithms might make suggestions about saving money on your plastic manufacturing process.

Based solely on this information, the Netflix-style algorithm seems to suggest that setting a temperature to 85 °C will save you more money than increasing the styrene by 2%. But now, let’s see how this data looks with confidence intervals.

Fero’s confidence intervals (shown here as dotted bars) indicate that the algorithm is 90% sure the predicted profit will fall somewhere in that range. Notice how much larger the first interval is than the second. Perhaps there is not enough temperature data or too much variability to make a confident prediction about the effect of decreasing temperature.

The Fero algorithm reveals that decreasing temperature will most likely increase profit. However, because the confidence interval extends below zero, there is a slight chance it could actually *lower* profit. Meanwhile, an increase in styrene will at worst, do nothing. With this information, you may decide to go full steam ahead increasing the styrene in a production batch and experiment with decreasing the temperature on a smaller batch.

At Fero, we built our platform around showing confidence intervals because they are critical for decision making. We want to give your employees all the information we can, including how confident we are in our own predictions, so they can make informed decisions about improving your processes.

If you’re interested in switching to a data-driven approach to manufacturing, request a demo today.