128
people messaged the agent, all from paid Meta ads
122 on Messenger · 6 on InstagramHome › Case study
Case studySuccess story
35 Approved Dealer Applications From $407.55 in Meta Ads
How a dedicated AI agent handled dealer-program inquiries for a battery and powersports products distributor over ten weeks, with the numbers, the definitions and the limits.
35
dealer applications completed and approved
Client-reported · roughly 27% of contacts$407.55
total Meta spend across three campaigns
About $11.64 per completed application~14 sec
median AI reply time, including a set typing delay
81% first wrote outside business hoursThe situation
Most dealer inquiries arrived when nobody was available.
The business recruits dealers through Messenger-focused Meta ads. Before the agent, one team member answered every inquiry by hand, Monday to Friday, 9 a.m. to 5 p.m. Eastern. Weekend messages waited until the next business day, and the team always followed up.
104 of 128people (81%) sent their first message outside Monday-to-Friday business hours
$3.56–$6.53paid cost per messaging conversation, depending on targeting and image creative
85paid messaging conversations started, per Meta, across three campaigns
The objective
Reply fast, ask the right questions, and leave the decisions to people.
The agent was scoped to answer dealer-program questions from approved information, learn each prospect’s business and preferred model (stocking, dropshipping or both), share the application when it fit, and hand off anything needing a human decision.
BOUNDARIESThe agent never promised approval, pricing, margins, territory or credit terms. Those decisions stayed with the client’s team.
What the agent was set up to do
- Answer dealer-program questions from client-approved information
- Ask about business type, location, product interest and dealer model
- Tag each lead’s intent as early-stage, medium or high
- Share the dealer application when it fit, without pressuring anyone
- Flag conversations needing a person and close politely when someone was not a fit
Why 81% matters
A business-hours team can only answer a minority of these inquiries in real time. The rest waited. The agent gave the other four in five an immediate, accurate first reply, then passed qualified conversations to the team with context attached.
How we built it
A dedicated agent, tuned by reading real conversations.
We followed the same deployment sequence we use for client work: discover, design, connect, test, launch and optimize.
Setup and optimization
Quick approvals, four to five rounds of tuning, and a team that stayed in the loop.
Client approvals were quick. After launch, the agent went through four to five rounds of optimization as new questions appeared. The agent also flagged six questions it could not answer from approved information, such as franchise fees, private label, consignment and payment terms. The team answered those manually to keep answers accurate instead of letting the agent guess.
TEAM VISIBILITYEvery conversation was duplicated into a separate tool, so the team could read it directly in Messenger as well as in ConvoRev, and stepped in whenever the agent could not answer.
The deployment, step by step
- Discover: mapped the inquiry source, common questions, the application as the conversion event and the governing policies
- Design: wrote the agent’s goal, qualification questions, boundaries and escalation triggers
- Connect: configured Facebook and Instagram messaging, contact tagging, team notifications and message duplication
- Test: ran test conversations before launch, including unclear and out-of-scope questions
- Launch and optimize: monitored early conversations and tuned the agent four to five times
Where people took over
18 people were tagged for human handoff, 21 tasks were created for the team, and the team sent 38 manual messages to 23 people. The client reports every conversation was followed up. Existing-customer matters, such as a product return, went to a person rather than the agent.
The results
From first message to approved application.
We report the path from attention to outcome, because messages sent and AI replies are operating metrics, not proof of value.
64 engaged
People who sent two or more messages, half of all contacts. 43 sent three or more, and 21 sent five or more.
ENGAGEMENT · 50%39 links sent
People who were sent the dealer application link in conversation. The agent tagged 15 as high intent, 7 as medium and 7 as early-stage.
NEXT STEP · AI-ASSIGNED TAGS35 approved
Completed dealer applications, all approved by the client, with about $5,000 in client-reported revenue from this channel.
CONVERSION · CLIENT-REPORTED| Stage | Metric | Result |
|---|---|---|
| Spend | Total Meta ad spend, three campaigns | $407.55 |
| Attention | Paid messaging conversations (Meta) / people who messaged (platform) | 85 at $3.56, $3.95 and $6.53 each (about $4.79 average) / 128 |
| Responsiveness | AI reply time across 217 measured replies | Median about 14 sec; 90th percentile about 41 sec |
| Responsiveness | First message outside Mon–Fri business hours | 104 of 128 (81%) |
| Conversion | Completed applications / approved | 35 / 35 (client-reported) |
| Cost | Ad spend per completed application | About $11.64 (excludes software, management and staff time) |
| Handoff | People flagged / team tasks / manual replies | 18 / 21 / 38 messages to 23 people |
| Quality | Accuracy of agent answers | Client reported no inaccurate answers; no formal audit of every transcript |
| Commercial impact | Revenue from dealers acquired through the channel | About $5,000, client-reported (revenue, not profit) |
What we learned
Speed started the conversation. Depth produced the better leads.
The client’s team observed that prospects who stayed in the conversation longer were higher quality and generated the revenue. This came out of a strategy pivot: about 50 contacts were tagged as having been given dealer portal access, to test whether it led to faster first orders. It was an observation, not a controlled test.
ALSO LEARNEDHigh-intent leads often moved to email with the team on their own, and conversation review caught a website bug that stopped an application form from submitting. The client fixed it.
Definitions behind the numbers
- Engaged means two or more inbound messages from the same person
- Intent levels were assigned by the AI agent; the client reports high-intent leads proved strong, but the tags were not formally audited
- Application link sent means the link appeared in an agent message, not that an application was completed
- 85 vs. 128: Meta counts conversations within its attribution window, while the platform counts distinct people who messaged
- The 35 applications are not matched one-to-one to the 39 link recipients, and the 27% ratio assumes each application came from a different contact
Read this before you compare
What this case study does not show.
A single deployment has real limits. We would rather state them than let you assume more.
The limits, plainly
- One deployment, one use case. A dealer program is not an appointment, estimate or ecommerce business.
- Client-reported outcomes. Applications, approvals and revenue come from the client and were not independently audited.
- Approval rate is not a benchmark. All 35 applicants were approved under the client’s own standards, so it says little about how another business would judge applicants.
- Revenue is not profit, and the ad-spend figure excludes software, management and staff costs.
- Attribution is imperfect. Other activity may have influenced some applicants, and we cannot trace each application to a specific conversation.
- AI-assigned tags. Intent levels reflect the agent’s judgment.
- Small sample. 128 people supports directional conclusions only.
- Reply speed includes a deliberate delay set in the agent, so it shows configured behavior, not the fastest possible response.
- No guarantee. These results do not promise leads, applications or revenue for any other business.
Common questions
Questions about this deployment.
Was this a real client deployment?
Yes. ConvoRev built and managed it for an independent, paying business with no ownership, referral-fee or shared-investor relationship with ConvoRev. The client approved publishing the aggregate results anonymously.
Can I expect the same results?
No. Results depend on traffic, offer, targeting, creative, information quality and follow-up. This documents one deployment and does not predict outcomes elsewhere.
Did the AI approve or reject dealers?
No. The agent answered questions, gathered information, tagged intent and shared the application. The client’s team made every approval decision and answered questions the agent could not.
Is the revenue figure verified?
It is client-reported. It is revenue, not profit, and ConvoRev did not independently audit the client’s sales records.
How much work did setup take?
Setup covered discovery, approved knowledge, testing and four to five rounds of optimization after launch. Client approvals were quick. Ongoing review and the team’s manual answers were also part of the work.
See the conversation for yourself
Curious how a dedicated agent would handle your inquiries?
Open the ConvoRev chat on this page and ask how an AI agent would qualify prospects for a business like yours. Then talk with us about a controlled first deployment.
If the chat does not load, email hello@convorev.com. Channel availability, plan features and integration requirements vary. Managed services may be scoped separately.
Best solution for automating client communications for agencies
How agencies can automate lead response for clients
Managed AI agents vs. DIY automation
Add AI conversations without replacing your CRM
Sources: platform message export and Meta Ads Manager reporting, July–September 2026, used with the client’s permission; application and revenue figures reported by the client.
Related: AI lead qualification · AI lead response · After-hours lead response