Your outputs are only as good as your inputs
Tips for developing automated rosters that your managers can trust
Forecasting workforce need is a useful exercise for budget-setting and workforce planning, but only if you get it right. Many businesses are struggling to achieve automated forecasts that recommend accurate numbers, so managers are forced to change rosters last minute and lose trust in the process. It’s easy to get overwhelmed by all the potential considerations with workforce forecasting, and to forget basic inputs. A machine is only as good as its orders, and in data science this concept is called GIGO: Garbage in, garbage out.
Classic GIGO mishaps occur when information is incomplete or inaccurate. Kids are great at demonstrating this:
When instructions are taken too literally…
…or when there’s room for interpretation…
Our topic today is more concise: what ‘instructions’ do you need to consider when automating roster shifts? I regularly work with clients to help them:
- Accurately forecast the metrics that underpin labour demand in a given business (sales, footfall traffic, bookings, etc), and
- Use automation to translate those forecasts into employee headcount recommendations for future rosters
At Ento we use leading AI forecasting techniques for the first point, and consistently generate highly accurate forecasts. The tricky bit is the second bit, where we translate demand into headcount need. Based on experience, I have three tips below:
Tip #1: Start top-down
Don’t get stressed about a detailed study on how long it takes to create every type of cocktail, including the custom half-sweet espresso martini with a shot of chocolate milk and a cherry on top for the over-enthused patron. There are so many variables, and while there is a time and place for time and motion studies, getting started is not it. A top-down approach that I find helpful includes:
- Doing an analysis on current efficiency, such as historic sales $ per employee per hour
- Breaking this analysis up in a way that makes sense for your industry. It may be most helpful to see different efficiency levels based on time of the week, based on number of people rostered, or based on role type
- If known, discuss whether average efficiency should be increased (you know staff have too much down time) or decreased (you know staff are overworked), or whether the location is already operating well
Using this information, you can create automated headcount suggestions that resemble how you’re already performing today. This can be a good starting point for building trust in automation outputs.
Tip #2: Interview rostering managers
Managers make many intuitive decisions based on their experience, and the logic seems so obvious to them that it’s not recorded. For example, a restaurant may be extra busy every time the local school sports team has a home game, or a retailer may have an extra large stock-in process every three months.
It’s like the activity that many of us did in school, where you try to write detailed instructions on brushing your teeth. People forget that they need to describe what the toothbrush looks like, that you grab the end without the bristles, that only the tip of the toothbrush goes in your mouth, etc. Robots are horribly literal (remember GIGO) so you need to list all of the obvious things.To capture this sort of knowledge, I like to include two interviewing steps:
Step 1 – have a manager build a roster and explain to me why they are choosing to include each shift, and
Step 2 – show them an auto-generated roster with all of the rules we discussed, and have them explain why it would or wouldn’t work (we almost always need to create more rules)
This approach has helped to identify important details that change the automated roster drastically, such as accounting for clean-up time or the impact of customers paying large bar tabs all at once.
Tip #3: Aspire to be predictable
I’m often asked if our automation can account for stock-in time at retail locations, and the answer is yes, but only if you already know when the stock-in times will be. Many businesses today will have delivery at random times or even on random days, depending on the logistics company they’re working with. When delivery is random, you need a bit of cushioning in the roster at all times, just in case. But when delivery is reliable, you can automatically account for stock-in without taking a hit on customer service.
The general rule with automation is that we can build a rule around anything, but only if you can articulate exactly how and when it occurs. Machines don’t do ‘sometimes’. Sure, we can run a comparison on your historic demand to figure out how variables such as the weather impact your business. But that analysis will only be helpful if you can forecast the weather accurately when you’re building your roster two weeks in advance, and you probably can’t.
My closing note isn’t so much of a tip, but a message: You’re going to find flaws as you go, and that’s okay. Robots are not perfect (yet). Just like how an astronaut shouldn’t do a surgeon’s job, robots and humans each have strengths and weaknesses. The robot can get you a 98% complete version in seconds instead of hours, and the human can tell you what’s wrong with the other 2%, because humans love to find things that are wrong. To illustrate, I give you the Mona Lisa.
If someone asked you to paint the Mona Lisa from scratch, you might be stressed about your ability to create something even remotely close to the Mona Lisa, and you’d want time to practice and improve. However, if you were given the Mona Lisa with a moustache, and just had to identify what was wrong, you’d end up with a higher quality and more efficient result. Automation can probably get you the Mona Lisa with a moustache, and your managers just need to learn how to identify the moustache instead of learning how to create the painting from scratch.
Did you find these tips helpful? Comment with any tips, GIGO stories and moustaches you’ve found with forecasting, or reach out to chat more!