Did shoppers get their cart?
A run counts as a success when the shopper presses “Looks good, I'm done”. From last 30 days.
- Runs that ended in an order
- 43%
- 33 of 77 runs
- Messages to order
- 1
- median; average 1.9
- Average rating
- ★ 4.5
- from 20 ratings
Runs
The 300 newest first. Names, stars and comments are public, with phone numbers, emails and links removed. Chat messages stay private.
- AnonymousDidn't order–2 turns8 model + 9 tool calls18.2 s$0.037
- AnonymousDidn't order–2 turns3 model + 3 tool calls9.0 s$0.017
- AnonymousDidn't order–2 turns3 model + 3 tool calls8.9 s$0.018
- AnonymousDidn't order–2 turns3 model + 3 tool calls9.7 s$0.019
- AnonymousGave up–1 turn3 model + 2 tool calls6.8 s$0.012
- AnonymousDidn't order–2 turns3 model + 3 tool calls9.4 s$0.017
- AnonymousDidn't order–2 turns1 model + 1 tool calls3.0 s$0.0037
- AnonymousDidn't order–2 turns3 model + 3 tool calls8.0 s$0.015
- AnonymousDidn't order–1 turn0 model + 0 tool calls0.0 s$0
- VedikaOrdered ₹1,908, 4 items★☆☆☆☆It doesn't understand "protein rich"6 turns9 model + 9 tool calls27.2 s$0.061
- AnonymousDidn't order–1 turn0 model + 0 tool calls0.0 s$0
- AnonymousOrdered ₹950, 6 items–2 turns6 model + 10 tool calls18.7 s$0.027
- DeepakOrdered ₹958, 8 items★★★★★I like that it asks follow up questions3 turns4 model + 5 tool calls15.5 s$0.019
- AnonymousDidn't order–2 turns4 model + 4 tool calls9.0 s$0.015
- AnonymousOrdered ₹1,092, 9 items–2 turns6 model + 6 tool calls20.5 s$0.057
More numbers
How much work each run took, from last 30 days.
- Runs
- 77
- Turns
- 151
- Helpful replies
- 90%
- 10 votes
- Turn, median
- 5.1 s
- p95 10.9 s
- Ended by code
- 76%
- no model call to finish
- Cost per order
- $0.053
- $0.0099 a turn
Running in production
The live site's last 30 days, from its request log, updated every minute.
Requests per day
Show as a table
| Day | Old 3-cart builds | Chat turns | Visitors |
|---|---|---|---|
| 17 Nov | 0 | 0 | 0 |
| 16 Nov | 0 | 0 | 0 |
| 15 Nov | 0 | 8 | 1 |
| 14 Nov | 0 | 65 | 9 |
| 13 Nov | 0 | 59 | 13 |
| 12 Nov | 6 | 6 | 2 |
| 11 Nov | 7 | 7 | 1 |
| 10 Nov | 2 | 0 | 1 |
| 9 Nov | 11 | 6 | 1 |
| 8 Nov | 0 | 0 | 0 |
| 7 Nov | 0 | 0 | 0 |
| 6 Nov | 0 | 0 | 0 |
| 5 Nov | 0 | 0 | 0 |
| 4 Nov | 0 | 0 | 0 |
| 3 Nov | 0 | 0 | 0 |
| 2 Nov | 0 | 0 | 0 |
| 1 Nov | 0 | 0 | 0 |
| 31 Oct | 0 | 0 | 0 |
| 30 Oct | 0 | 0 | 0 |
| 29 Oct | 0 | 0 | 0 |
| 28 Oct | 0 | 0 | 0 |
| 27 Oct | 0 | 0 | 0 |
| 26 Oct | 0 | 0 | 0 |
| 25 Oct | 0 | 0 | 0 |
| 24 Oct | 0 | 0 | 0 |
| 23 Oct | 0 | 0 | 0 |
| 22 Oct | 0 | 0 | 0 |
| 21 Oct | 0 | 0 | 0 |
| 20 Oct | 0 | 0 | 0 |
| 19 Oct | 0 | 0 | 0 |
What went wrong in real chats
- Turns whose cart failed a check
- 1%
- Turns that gave up (cart left as it was)
- 2%
- Requests that failed
- 3%
- Big changes held for the shopper's yes
- 8%
- First messages answered from the cache
- 18%
Where the time goes
| Model call | Calls | Average | p95 |
|---|---|---|---|
| plan (old 3-cart builder) | 16 | 4.3 s | 6.5 s |
| cart (old 3-cart builder) | 48 | 3.0 s | 4.6 s |
| One agent model call | 277 | 2.8 s | 6.1 s |