For callers, the two channels that actually work are OnlineJobs.ph for Filipino appointment setters and Upwork for native English when the niche needs it. Facebook groups for sales reps are a third option but the quality is wild. What matters more than the channel is how you filter: 1. Post the role with the pay and the KPI in the ad, not "commission only". Commission only gets you people who quit in week two. 2. Make every applicant send a 60 second voice recording reading your actual script. You will cut 80% on tone alone. 3. Paid trial day on real leads, not a mock call. You find out in an hour whether they can handle a "not interested". 4. Give them a tight script and one objective (book the appointment), not freedom to sell. Worth asking yourself first whether you need a caller at all. If the AI is booking and you just need someone to show up to calls, that is a different, cheaper hire.

It doesn't really matter. The Android doesn't care which model sits behind the webhook, it just needs a reply written back into the Chat-GPT custom field. Swap the model node in n8n/Make/Zapier and everything else stays as it is. Practical differences worth knowing: 1. Claude tends to follow a long system prompt more tightly and is less likely to drift out of character or over-explain. Good for SMS where you want short, human replies. 2. OpenAI has more examples floating around, so if you're copying a build from a video the nodes will line up with less fiddling. 3. Whichever you pick, watch message length. Models love writing paragraphs, and every 160 characters is another billed segment. Put a hard "keep replies under X characters" instruction in the prompt and test it. One thing you must add either way: an instruction that if the lead asks whether they're talking to a bot, it answers honestly and immediately that it's an AI. Most prompts tell the bot to stay in character no matter what, and that's the bit that breaks under the EU AI Act if any of your contacts sit in the EU (Ireland catches a lot of UK people out). Dan covers it here: https://www.youtube.com/watch?v=ZWqcUUbqDb4

Legal is one of the few niches where you genuinely can't do rev share, so price off case value instead. For criminal defence and PI, a signed case is worth thousands to tens of thousands, so a per-qualified-consultation or per-signed-case fee is easy to justify. How I'd structure it: 1. Setup fee, 1,000 to 2,500, covers the build, A2P registration, list clean, prompt and the first campaign. Never do the build free. 2. Then either a flat monthly retainer, or a per-booked-consultation fee. Retainer is simpler to get approved by their compliance and easier for you to forecast. 3. Make them pay the usage directly, or build it in with headroom. A 2,000 lead campaign runs about 130 in SMS, so price well above that. Two practical points. Old PI and criminal leads need consent checking before you text them, and the firm must confirm they have it, in writing. Also keep the AI off anything that reads as legal advice or a case outcome promise, that is a bar rule problem for them, not just a conversion one. Ask them for their average case value on the first call, every number above hangs off it.

Start from what the client pays you for, then build back. For an Android the numbers that actually matter are: 1. Leads entered (how many contacts started the flow) 2. Reply rate (contacts that responded at all) 3. Conversation rate (two or more replies, so the AI is actually engaging) 4. Appointments booked 5. Show rate 6. Closed/sold, if the client will feed it back The good news is the snapshot already gives you most of this for free via the AI Chat Pipeline. Each stage (Message Sent/No Response, Lead Responded, Multiple Responses, Lead Qualified, Send To Client) is a funnel step, so a pipeline-based dashboard widget gives you reply and qualification rates without any extra work. Add a widget for appointments from the calendar, and one for contacts added. Two things people skip and regret: - Tag or use utm_source so you can split results by lead source or batch. Otherwise month two blends into month one and you can't prove anything. - Send the client a short weekly summary rather than CRM access. Three numbers and one screenshot of a real conversation beats a dashboard login they never open. The conversation screenshot is what renews contracts. Keep the dashboard to six tiles max. More than that and nobody reads it, including you.

@John Paule You can get that from GHL if they use it to mark if each appointment showed or not, but usually you'd get that from them

Best way to get takers here is to post the actual numbers rather than the ask. Pay per meeting partners will only bite if they can see the maths works for them, so give: the niche and geography, what counts as a qualified meeting (job title, company size, intent), your show rate expectation, close rate and average contract value, and what you will pay per meeting. Without that, anyone decent assumes it is a fishing expedition and scrolls past. Two other things worth knowing before you sign anyone: insist on a short paid pilot (10 to 20 meetings) before any volume commitment, and agree up front who owns the domains and inboxes. Agencies that burn your main domain are common. Honestly though, pay per meeting is usually worse value than building the outreach yourself, because the margin they need forces them to spray. If you want the in-house route, we run B2B Leads Lab for exactly this, including a novel approach using bottles sent to decision makers (50 bottles, 7 paying clients). https://go.flexxable.com/b2b-leads-lab-g