Our customer service history is a mess of outdated policies, angry rants, and inconsistent advice. How do we clean up this legacy text data so we do not train a customer support AI on our worst mistakes?

Question

Tyler Smith · Accepted Answer

You do not need to spend weeks manually cleaning thousands of old email threads and customer tickets before you build your AI. That is a waste of your team's valuable capacity and will stall your momentum.

Instead, isolate your best data from the start. Identify your top three customer service reps, the ones who consistently live your core values and get great customer ratings. Pull a sample of their last fifty completed tickets. This small, clean set of high quality transcripts represents your gold standard.

Next, use your documented standard operating procedures as the baseline. Your core processes, developed during your EOS® journey, contain the correct answers and policies.

When you build your AI assistant, upload your current standard operating procedures and your gold standard transcripts as the reference library. Explicitly instruct the AI in its system prompt to ignore any historical database records and only use the provided library to draft responses.

To keep your team in the loop, set up your workflow so that the AI only drafts the response. A human who GWC™'s the role must review and click send. This eliminates the risk of bad legacy data poisoning your output while keeping your operational momentum high.