Activity
Mon
Wed
Fri
Sun
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
What is this?
Less
More
8 contributions to AI Automation Society
3 silent failures that break lead follow-up workflows (and the daily test that catches them)
A lead follow-up workflow can look healthy for days while nothing reaches anyone. No errors in the logs, the API hands back a message id, and the client finds out when a prospect complains. These are the three I run into most: 1. Trusting the send step A success code or message id from the send API only means the provider took the request. Delivery can still fail afterwards because of trial account limits, a dead webhook or a status callback nobody reads. If you only watch the send step you can miss weeks of follow-ups that never landed. 2. Silence that looks like a slow week A dead trigger or broken webhook doesn't throw anything. Volume just drops to zero. I treat "zero sends in a window where we normally send N" as a failure, same severity as a red error. 3. Retries that double-send, and skips nobody can see A timeout after the message already went out can make the retry send a second copy. A skip with no log line looks broken to the client, so I log skips too, with the reason. What catches most of this for me: every morning one test lead goes through the exact same path real leads use, ending at my own phone or inbox. If it doesn't arrive, I know before the client does. On top of that I alert when no delivery status comes back within 10 minutes, and on failed or undelivered callbacks. Watching for thrown errors alone misses most of it. Which of these has burned you on a live workflow, or is there a fourth one I'm missing?
0 likes • 11h
@Muhammad Sajid That's a good fourth one, and probably the most expensive when it goes wrong since it hits the client's number or domain. An opt-out test lead plus a unique key per lead and step covers both problems nicely.
0 likes • 4h
@Sam Jak Timing the first reply is the useful part, since a slow answer is often the first sign something upstream is off. Logging that time each day also lets you spot a gradual slowdown before it turns into a full failure.
Pricing automation work from 6 numbers instead of hours (worked example)
"What should I charge?" shows up here almost every day, and the usual answer is hours × rate. That's how a build worth $10k gets quoted at $2k. The method I use starts from the client's numbers instead. You need 6 of them from the discovery call: 1. Hours per week the staff spend on the workflow, and what that hour costs them 2. Deals or appointments lost per month because of it 3. Revenue per deal and the client's margin 4. Your build estimate in hours 5. Your minimum hourly rate 6. What you were about to quote The math: - Cost of doing nothing = (hours × cost × 48 weeks) + (lost deals × revenue × margin × 12) - Conservative value = half of that, because you're estimating - Price = 20% of the conservative value, and never below build hours × your minimum rate Made-up example, a small law firm: a paralegal spends 15 h/week on intake at $40/h, and they lose about 2.4 matters a month to slow replies, at $5,000 per matter and 50% margin. - Cost of doing nothing: $28,800 + $72,000 = $100,800 a year - Conservative: $50,400 - Price: about $10,000 (the floor at 45 h × $150 is $6,750) - A $2,000 quote would have been under the builder's own floor Two habits go with it. The proposal opens with the cost of doing nothing in the client's own numbers, so the price gets compared to that and not to some freelancer's hourly rate. And if one of the 6 numbers is a guess you can't defend, a small paid diagnostic to measure it comes before the build quote. What percentage do you price at? I've been using 20% of the conservative first-year value, and I'm curious whether people go higher on retainers.
1 like • 1d
@Dr. Stephen Buscema Same day is the part that matters. A week later nobody remembers why the number moved, and that's when the arguing starts.
1 like • 11h
@Dr. Stephen Buscema A change log in the proposal itself is a nice touch. The client can see what moved and why without having to ask.
30 days of daily AI-edited reels: what actually moved saves (real dashboard)
Since Oct 30 we've been putting out 2 to 4 short-form reels a day on TikTok, Instagram and YouTube. The screenshot is our real dashboard for the last 30 days: 95 posts, 39.2K views, 2.1K saves. Most reels did okay. One clearly beat the rest: 3,879 views and 209 saves on Instagram, 2,302 views and 138 saves on TikTok, and 70 people commented a keyword on it. What it did differently: 1. The hook put a free option right next to a paid tool. People stopped because they thought they might be overpaying. 2. It was useful enough to come back to. Saves pushed reach more than likes did. 3. It ended with a one-word comment ask instead of "link in bio." Every comment showed us which hook landed. What we changed after it: we now track saves first, write every script around one concrete comparison, and review the numbers every morning before writing the next one. What hook format has gotten you the most saves, as opposed to views?
30 days of daily AI-edited reels: what actually moved saves (real dashboard)
0 likes • 6d
@Ali Abdullah Steal away. It only takes a few minutes and it keeps you from writing the next one blind.
0 likes • 4d
@Jeff Randle Will do. I'll give it a few weeks first so the numbers actually mean something.
The 5-minute morning check I run on every client automation
I've had automations that looked fine for weeks and then quietly stopped doing the one job that mattered. No error email. No Slack ping. Just silence, and a client who assumes it's still working. So every weekday morning I run the same short check on each live workflow. Takes about 5 minutes per client: 1. Did the last expected run actually fire? Look at the last successful timestamp, not the "active" toggle. 2. Did the destination still accept the output? One rejected row in the inbox, CRM, sheet, or calendar is enough. 3. Did any login or API token expire since yesterday? 4. Did the client's offer, form, or FAQ change in a way that breaks the replies? 5. Is there a queued failure nobody opened? If something failed, I fix it before the client notices. I also keep a one-line log: date, check result, what I changed. That log is the difference between real monitoring and a vague promise. The free Claude Code starter I use for client work is here (MIT, no email wall): https://github.com/CalebAnderson-Coder/ai-agency-os-starter It has the skills I run around quoting, scope, and client folders. The morning check above is still a manual habit; I haven't packaged it as a skill yet. What do you check daily that I should add to this list?
0 likes • 9d
@Jose Sangiuliano That’s a useful addition. A per-client cap plus an alert when spend jumps above its normal range can catch a runaway retry loop before the invoice does.
1 like • 9d
@Jose Sangiuliano Where I can, I'd put the keys on the client's account. They see the spend directly and the cap lives with them. If it has to run on mine, the per-client cap and alert become non-negotiable.
The 3 lines I add to every automation scope doc so "small changes" don't eat the margin
Most scope creep on automation projects starts with a vague scope doc. "Lead follow-up automation" sounds clear when you sign it. Three weeks later the client asks for WhatsApp, a second form, and invoice emails, and all of it fits inside that sentence if you squint. These days every scope doc I send has three extra lines: 1. Inputs. Exactly which forms, inboxes or sheets the workflow reads from. "The contact form on the homepage" is in scope. "Any new lead source" is not. 2. Outputs. What it sends, where it goes, and how many message templates are included. Example: 1 SMS + 2 follow-up emails, written once, edited up to 2 times. 3. Change budget. A small number of free tweaks after launch (say 2 per month), and anything beyond that gets quoted as an add-on before I start. When a "can you just quickly add..." email lands, I split it into separate asks and check each one against those three lines. Whatever is in scope I do right away and mention in the reply. Whatever isn't gets a price and a date, or a swap for something still unbuilt. That conversation is much easier when the line was written down before anyone got annoyed. I turned the checking part into a free Claude Code skill called scope-guard. You paste the client email and your scope, and it splits the asks, flags what's out of scope, prices the extras and drafts a reply. There's a worked example with a fictional plumbing company. It's in this MIT repo, no email needed: https://github.com/CalebAnderson-Coder/ai-agency-os-starter Do you keep a free change budget in your projects, or quote every extra from day one?
0 likes • 9d
@Madonna Lapid That’s a useful finish line. A test lead plus the client confirming the expected destination makes handoff much less subjective.
0 likes • 9d
@Zubair Aqeel That's a great fourth line. Naming one approver stops side requests from other people turning into work nobody actually signed off on.
1-8 of 8
Brian Anderson
5
328 points to level up
@brian-anderson-1806
Founder of Agencia IA. I build AI agents and systems that turn messy sales ops into booked calls, follow-up and revenue.

Active 4h ago
Joined Feb 14, 2025
Powered by