top of page

The Picnic Basket Was Still Missing

Writer: Suzanne Hansen
Suzanne Hansen
7 days ago
2 min read
A steel water bottle nestled in packing paper inside an open cardboard box.

A steel water bottle is a perfectly good thing to find in a package. Unless you ordered a wicker picnic basket and need it before Friday.


That’s the fictional conversation I bundled with Reply Checkup. The customer reports the wrong item and gives the deadline. The bot replies, “Thanks for writing. Your order is in progress.”


The customer repeats both details. This time the bot says, “I can pass your note to the support team.”


Okay. Did it?


The conversation doesn’t tell me whether a note was sent. It also doesn’t tell me whether anyone can get the basket there before Friday. If I’m the customer, I have to decide whether to wait, try another channel, or buy a second picnic basket. I’ve given the bot the information it needs twice, and I’m still the one figuring out what happens next.


I kept running into some version of this in Beyond First-Response Accuracy. An assistant could explain what had gone wrong in its earlier answer, and I’d still have to go through the revised advice to see which recommendations depended on the mistake. With the picnic basket, the same question gets wonderfully unglamorous: am I any closer to getting the thing I ordered?


While building Reply Checkup, I started with a few wording patterns I could check consistently. The broken-return-label walkthrough on my portfolio shows one of them: the customer says the link in their order email is broken, and the bot sends them right back to it. I can put the flag beside the exact reply and show why it fired. That’s a separate example from the picnic basket you’ll find in the tool.


The basket shows why I also wanted somewhere to record my own reading. I’d ask the team to acknowledge the wrong item and the deadline, then give the customer a concrete next step. If the bot can check replacement options, I’d want it to do that. If it can only offer to pass along a note, I’d want to know what the customer should expect and what the bot can actually confirm. I’d run the conversation again after the change and see whether the next reply leaves me with a decision I can make.


I tried automating some of that judgment, too. In one small experiment, all 14 draft assessments passed the format checks. Two still made substantive mistakes when I read them against the conversations. One pointed to the wrong earlier claim. So yes, I built a checker and then had to check the checker. I kept that part experimental.


For the first version, I want to make it easier to see the conversation, check what the tool picked up, and write down what I still think needs attention. The Reply Checkup portfolio preview shows that difference with the return-label example. The picnic basket is the one I’ve set aside for you to review yourself.

 
 
 

Comments


bottom of page