Sales outreach · Production

Smart Recover increases AI message acceptance rate from 24% to 77% with PromptHub

The problem

Smart Recover wanted to use AI to generate human-like SMS messages for lead re-engagement, but their Airtable-based prompt testing quickly became unmanageable: it was cluttered, change tracking was nearly impossible, multi-model testing was out of reach, and there was no way to measure whether prompts were improving.

First attempt

An Airtable database used to test prompts quickly got out of hand: it grew cluttered after minimal use, couldn't track how changes affected outputs, made multi-model and multi-parameter testing nearly impossible, and offered no reliable measurement of prompt quality over time.

Workflow diagram · grounded in source
1
Lead abandons website
trigger
“re-engaging leads who abandon websites, using real-time, one-on-one SMS conversations managed by live agents”
2
AI generates outbound SMS
ai_action
“the Smart Recover team knew they could generate a percentage of their outbound messages with AI. The prompt used to generate support messages would be critical. The messages would need to sound human-like, have the proper context, and mi…”
3
Batch prompt-version testing
validation
“Easily testing 100s of prompt versions to generate better support messages for their clients. Batch testing was particularly helpful and enabled them to be more confident that changes in their prompt were performing at scale, rather than…”
4
Human agent reviews message
human_review
“24% of the time, the message was deemed of high enough quality to be sent by the human in the loop”
5
Approved message sent to lead
output
“using real-time, one-on-one SMS conversations managed by live agents”
Reported outcome

After adopting PromptHub, Smart Recover's AI message acceptance rate rose from 24% to 77% — an increase of more than 220% — with batch testing enabling confidence that prompt improvements held at scale rather than producing a few lucky outputs.

Reported metrics
AI message acceptance rate (baseline)24%
AI message acceptance rate (after PromptHub)77%
AI message acceptance rate increasemore than 220%
Prompt versions tested100s of prompt versions
Reported stack
PromptHubAirtableSlack
Source
https://www.prompthub.us/customers/smart-recover
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

After adopting PromptHub, Smart Recover's AI message acceptance rate rose from 24% to 77% — an increase of more than 220% — with batch testing enabling confidence that prompt improvements held at scale rather than pro…

What tools did this team use?

PromptHub, Airtable, Slack.

What results were reported?

AI message acceptance rate (baseline): 24%; AI message acceptance rate (after PromptHub): 77%; AI message acceptance rate increase: more than 220%; Prompt versions tested: 100s of prompt versions (source-reported, not independently verified).

What failed first in this deployment?

An Airtable database used to test prompts quickly got out of hand: it grew cluttered after minimal use, couldn't track how changes affected outputs, made multi-model and multi-parameter testing nearly impossible, and…

How is this sales outreach AI workflow structured?

Lead abandons website → AI generates outbound SMS → Batch prompt-version testing → Human agent reviews message → Approved message sent to lead.

Related sales outreach cases