It's easy for us to equate data (and by extension, decisions made with data) with objectivity, scientific fact, and truth. More often than we might realize, this uncritical acceptance is built into our daily lives. But, as I was (somewhat hilariously) reminded last summer, it's in our best interests to maintain a healthy skepticism about how data informs our actions.

One beautiful day last July, I had to travel from my apartment near St. Lawrence Market in Toronto to a medical appointment in North York. I decided to ride my bike. I'd never gone there by bike before, but I wasn't too worried about finding my way. I did the same thing I do when driving a car—I just turned on a navigation app on my phone.

I'm sure you've used these kinds of apps before—they combine network analysis (finding efficient paths, generating directions) with GPS (determining your location and speed in real time as you travel). These tools are really useful for all kinds of things, and my journey started out fine, with my app generating a Best Route optimized for cycling. On my map, it looked perfect, and the navigation app was genuinely helpful as I headed east past the Distillery District and north on the paved, multi-use Don Valley Trail. The ride was scenic and I felt confident.

When the main recreational trail began to branch off into a series of smaller trails, however, things began to unravel. The paved trail gave way to gravel. This wasn't ideal for me, but I trusted the app's algorithm about this route's overall efficiency. I figured I'd be back on pavement soon. Actually, though, the gravel became rougher. I had to dismount and push my bike up a steep hill. Once I made it to the top, I was instructed to turn right… onto a tiny footpath overgrown with Japanese knotweed. I wouldn't normally take a path like this on an urban bike commute, but according to the blue line on my app, it was the Best Route. So I gave it a shot.

The path grew narrower until it simply ended at an embankment overlooking a ravine. The view was very pretty, but this wasn't where I was supposed to be. I figured user error was to blame—I must have taken a wrong turn. I retraced my steps and returned to the gravel trail.

Thus began a series of data-driven misfortunes.

As my situation deteriorated, I found myself portaging across a creek, climbing up an embankment with my bike on one shoulder, and navigating a ledge alongside an active excavation site. At one point, I discovered a totally awesome forest mountain bike course (which might have been fun if I weren't on a road bike about as old as I am, and with other places to be). When I thought I'd finally returned to civilization, I realized I was standing on the grounds of an industrial facility surrounded by 'No Trespassing' signs. I question the legality of some of my movements.

Through it all, the software on my phone kept insisting that this was the Best Route.

I started to lose confidence I'd make it to my appointment on time. I called my doctor and left a voicemail: 'Hello. I'm somewhere in the woods south of your office. I'm trying to get there and I think I'm still moving north, but this is more of a wilderness expedition than I was planning for. If I make it out of here, I'm happy to pay the no-show fee.'

Finally, I decided to give up on algorithmic navigation. I followed the sound of cars out of the forest. I asked a person for directions. I did ultimately make it to my destination, but my navigation project had been a comedy of errors.

Phone screenshot of bike route

A generalized log of my route that my phone recorded. You can see that I took the subway on the way back—I was tired!

This happened months ago, but I was thinking about it again as I recently had a follow-up appointment with the same doctor. I still wanted to bike there, and still didn't know the best way, but this time I decided to be more thoughtful about it.

I went to the City of Toronto's open data portal and downloaded their Cycling Network dataset. Using this and the city's regular road network, I drew a path that made sense for me, favouring paved recreational trails and city streets that weren't too busy. I added in each street or trail's name to my curated route, as well as the distance in kilometres of each road segment I'd be traveling on. Then I added this layer to ArcGIS Field Maps (available with our ArcGIS Online subscription) and used it as my navigational tool. (I could have, alternatively, used QField for a free/open source option.) This time, I made it to my appointment in good time and with a minimum of adventure.

Smartphone with custom route map

My custom navigation map.

All of this was a reminder that the classifications and assumptions made by data-driven tools can really fall short when they're confronted with reality. I think it's interesting to note that the routing engine used by my app was not malfunctioning or malicious—it did its job and generated a Best Route with the data available to it. And it's not even that these data were wrong. The software was optimizing for the variables it knew about, and it had no way of knowing that its definition of 'best' and mine were wildly different. In my case, no one was harmed, but I'm sure we can all imagine contexts where uncritical data use has much more serious consequences.

'Data-driven' decision making is a buzzy term, but data do not—or at least, should not—ultimately make decisions. Every dataset is a simplified version of reality. To make the world computable, we reduce its complexity. Often this works remarkably well; other times it... doesn't. We need to use our discernment to look at a dataset in context, ask the appropriate questions of it, and consider how it might be supplemented with local knowledge and common sense.

I hope my slightly disastrous bike ride can be a reminder: genuinely helpful decision making doesn't come from surrendering our judgement to data; it comes from combining data and technology with care and consideration.

Data are not decisions