We keep hearing questions about where automations should stop and human work should start. Our recent conversations gave us a view into that question from two different altitudes:

“If you get like 60% off of their plate and they can just then come in and use the human sort of judgment for what’s important, that’s a win… As soon as [managers] try and solve every edge case, the tool’s just not gonna work.”

- CFO, PE-Backed Landscaping Business

“We were told to automate the design review process because it was taking 15 minutes per order. We first got it down to 2 minutes but then one of us would still, you know, need to check. Now that we’ve kept adding context the AI takes about 15 minutes to run.”

- Quality Control Manager, PE-Backed Specialty Design Manufacturer

The CFO knows where to stop in principle. The manager found out where to stop by overshooting it. Having seen that calibration play out in multiple companies, we tend to segment each workflow into three parts.

In most of our builds, the middle of a sequence gets automated first. For instance, we built an intermediate step to convert raw data into standard reports for a client’s sales team. Having human input from the sales team on either side ensured data integrity and output quality.

We believe the start of a process should be the next automation, especially when inputs are constant. Because of this we often help clients wire up their data to existing automations, which is exactly what we did for that sales team.

The last mile is where we draw the line. Just like the CFO’s quote above, it’s the human judgment that’s valuable. And just like the manager’s quote, the finishing touches are often the hardest to get right.

Here are some last-mile areas where we’ve seen teams struggle:

1. Insights: automating idea generation isn’t the same thing as picking the best one.

2. Communication: getting rid of the emdash helps but all good writing is rewriting.

3. Edge Cases: chasing every possible outcome is a never-ending waste of time.

If you read last week’s blog (Recycle the Garbage), then you know it’s only fitting we’re telling you what not to do. However, we always try to leave you with a game plan.

Here’s the immediate action for operators to take:

• For your current or next automation project, split the sequence into start / middle / end. Write down what portion of each step should be automated versus human and have that ready to share with the team who is going to build it.

Only add new segments to automate once the first one is complete. The best automations we’ve built have always consisted of multiple small wins stacked together over time.