Smart CSAT Model Upgraded live

In case you didn’t already know, we launched the Smart CSAT Model upgrade into GA on 16 September ! Thank you so much to those who participated in the BETA and shared feedback to make it a success. Summarising the impact and what we learnt: 

Three improvements:

• Better identification of which participant is the customer, so scoring is inferred from the right person
• More reliable handling of open-ended conversations
• A more discerning scale — ratings are better distributed across 1–5, so a 5 means more than it used to

Impact: Because scores are calibrated differently, expect metrics to shift slightly - this will differ by account. Here's what we saw in our BETA testing with 10 customers:

• Mean star rating: 4.03 → 3.80. 5 stars is harder to earn.
• SmartCSAT% (share of 4–5 star): 64.9% → 60.9%. Much of the movement was 5s shifting to 4s, which stays inside the 4–5 band. This is a useful metric to continue to use when making historical comparisons 
• Scoring coverage is lower. The model now skips ambiguous conversations rather than guessing, and flags low confidence more reliably — so some conversations will show no score where they previously had one.

Beta customer quote: "The smart CSAT scores definitely took a bit of a dip, but it seems to be accurate for what I would rate those convos too, so all is well there. I am excited to get more realistic scores, and this seems to be doing that."