The Smoothie Problem
What happens in the blender?
A key consideration when selecting process improvement solutions
Suppose you and a few friends go to a local restaurant, and one of your friends orders a strawberry-banana smoothie. He watches it being prepared and sees that precisely 5 strawberries are placed in the blender. Once the mixing is complete, he tastes it and says:
“Wow, that 3rd strawberry is really delicious!”
Is that realistic? Of course not… at least when the subject is smoothies. It’s possible to evaluate the taste of the 3rd strawberry before it hits the blender. But once the strawberries are mixed with the other ingredients, it’s a completely different story. The interaction makes it impossible to accurately assess their individual taste.
The same is true when it comes to process improvement. I worked in a children’s hospital once that was trying to improve the utilization of their MRI machines. They pinpointed several problems, many of which had simple fixes:
- Different shift starting times for nursing, anesthesia, and radiology techs consistently resulted in the first case starting late, which ate away at utilization. The fix was to coordinate functional schedules- a minor adjustment.
- Radiologists were giving the techs that ran the machines instructions that were too vague, resulting in delays while the techs chased them down for clarification. The fix was to mandate that the radiologists enumerate specific studies for each patient for the following day before leaving the night before.
- Parents hadn’t been informed that their kids would need anesthesia prior to the MRI, resulting in many procedures getting postponed at the last minute because kids had eaten too close to procedure time. This resulted in empty scanner time. The fix was to explain the fasting rule to the parents on the mandatory pre-procedure phone call.
- Speaking of the pre-procedure call, the scripting was changed to solve another problem. The current situation was that the nurse would call and say “Your appointment is at 9:30, so be there by 8:30.” Families routinely ignored the 60-minute guideline, resulting in delays and hurting utilization. Scripting was changed to “Be there at 8:30 and expect to be there four hours.” That way visit length expectations were set and parents would know that 8:30 was the necessary arrival time.
There were a few more, but you get the idea. The hospital implemented a wave of priority one recommendations at the same time, and utilization went up 12%. Good news? Certainly… but there could be smoothie implications.
How successful was each solution? Some analysts might break it down as if each solution was independent- shift time adjustments were responsible for 2.4%, improved specificity of instructions contributed 5.2%, etc. But that can be like evaluating the taste of each strawberry. If you implement six solutions at once, they all go in the blender and interact. There’s a critical piece of information here that serves as a great example of the dangers of assuming independence: many parents didn’t realize their kids needed to be put to sleep for an MRI. This implies time to get them to sleep and for recovery, which is why the procedure took four hours.
So what? Well, the pre-procedure call was conducted very close to the day of appointment. When the scripting was changed it could definitely increase the on-time arrival percentage for the families that showed up. Unfortunately, it could also increase the last-minute reschedule rate when families were told they would need to be at the hospital for four hours. That “solution” could actually be damaging to utilization, because the hospital couldn’t substitute a patient in with so little notice. Without this “improvement” they might have gone up 16% instead of 12%.
Another response to the question of how successful each solution was could be “who cares- we went up 12%?” The above illustration shows that reaction also misses the point.
So what’s the best approach? Is it the glacially slow and terminally boring method of making one change, measure, make one more, measure, and so on? Could be. But the best-case scenario is having enough process knowledge to determine which subset of potential solutions truly are independent from each other.
That’s when things will improve the fastest and the smoothiest.