Observe the Task Before Automating It

The job

A cheese plant’s packaging line ends at a conveyor. Cases of shredded cheese, each about 20 kg (44 lb), come off a case sealer at 6 a minute. One person picks each case off the conveyor, turns, and stacks it on a wooden pallet, 5 cases to a layer, 8 layers high. Then they wrap the pallet, call the forklift, and slide in an empty one. Management asks: can a robot do this?

MIT’s Work of the Future task force studied questions like this one across industries and found that automation changes tasks, not whole jobs, takes years to spread, and goes along with new kinds of work ¹. So the first step isn’t picking a robot. It’s finding out what the job actually is.

Watch the job

Sit with the operators for 30 cycles on each shift, including the night shift, a product changeover, and a day the line ran short-staffed. Time each step. A summary might look like this:

Ask the operators what makes a bad day. They’ll tell you things the timing sheet won’t: cases from one supplier’s cardboard crush on the bottom layer; the sealer jams when the tape roll is nearly empty; the last two layers are the ones that hurt.

The lifting

The stacking is the hazard. NIOSH’s lifting equation gives a recommended weight limit: start with a load constant of 23 kg (51 lb), the most nearly anyone can lift all day under ideal conditions, then multiply by factors that shrink it for how far the load is from the body, how low or high the hands start, how far it travels up or down, how much the person twists, how often they lift, and how good the grip is ². The lifting index is the actual weight divided by that limit; above 1 means increased risk of back injury ².

A 44 lb case lifted 6 times a minute, from a belt to a pallet on the floor, with a twist, starts close to the 51 lb ideal limit and loses ground on every factor. Measure the real distances and run the numbers: the bottom layers and the top layer will come out worst. That’s useful, because it tells you which lifts any fix has to remove first.

What the robot would have to do

Split what you watched into kinds of work:

• Repetitive, predictable handling: picking identical cases and placing them in a fixed pattern. A good automation candidate.
• Support work: wrapping, labeling, pallet change. Can be automated separately (stretch wrappers and pallet dispensers exist) or stay manual.
• Judgment and recovery: deciding whether a crushed case is reworked or scrapped, clearing jams. Hard to automate and easy to forget.

Recovery deserves its own line in the study. If a robot picks a case in 4 seconds but a mis-pick takes a skilled person 8 minutes to sort out, the jam rate matters more than the robot’s speed.

Five ways to fix a job

1. Eliminate the task: fix why the sealer jams (tape-roll sensor, a better cardboard spec).
2. Simplify it: a pallet on a scissor lift or turntable keeps every lift near waist height and removes the twist.
3. Help the person: a vacuum lift assist takes the weight.
4. Automate part of it: a robot stacks, a person handles pallets, labels, and rejects.
5. Automate the whole bounded task, and create the new work it needs: cell tending, maintenance, changeover programming.

Usually the answer is a mix, and the cheap fixes (1 and 2) come first, because automating a process that jams four times a shift just builds the jams into steel.

Try it

Pick a job you know (a workshop, a kitchen, a loading dock). Watch it for an hour, time the steps, and fill in the table above. Mark which tasks are handling, support, and judgment, and write one fix from each of the five categories.