Activity
Mon
Wed
Fri
Sun
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
What is this?
Less
More

Owned by Jason

This community is your unfair advantage! AI tools, prompts & blueprints to start or grow a business faster than everyone else!

AI Automation Club

1.3k members • Free

DevsUnlocked

11 members • Free

ZeroOne Systems

14.8k members • Free

AI Automations by Jack

3.6k members • $87/month

AI Automation Vault

36.8k members • Free

AI Automation Circle

13k members • Free

Skoolers

160.3k members • Free

AI Automation Society

455k members • Free

AI Automation Agency Hub

334.1k members • Free

1247 contributions to AI Automation Society
If you had a business automation startup, how would you promote its affiliate program?
Got curious about this and figured I'd ask people who've actually done it. The obvious first step is finding your early customers and offering them the affiliate deal directly. That part makes sense to me. What I'm stuck on is the next step: if that doesn't get enough traction on its own, where do you actually find people who'd be interested in promoting a SaaS affiliate program in the first place? Curious how others have approached this, especially early on with a small user base.
0 likes • 36m
I would think about affiliates less as "promoters" at first and more as workflow partners. Look for adjacent consultants, implementers, agency owners, and niche community operators who already have trust with the exact people feeling the pain your SaaS solves. Then make it easy for them: a clear referral path, a simple one-page explanation, a few proof assets, and a way to introduce the product without feeling like they are selling something random. Incentives matter, but early on trust and fit matter more. Once you see which partner type actually sends qualified leads, then scale the incentive program around that pattern.
Most "automation projects" fail before automation even starts
Ten years selling and implementing B2B systems taught me something automation content usually skips: most "this didn't work" automation projects were dead before a single workflow got built. The pattern's always the same. Someone maps the process from memory, the way they think it runs, then automates that version. The automation runs perfectly. It just automates the wrong bottleneck, because the real one sits upstream of where everyone assumed, often further back than anyone expects. Pull the last 20–30 real cases for that process, whatever the record is: tickets, emails, support logs. Timestamp when each one entered and left every step. Don't ask anyone what they think happens; the timestamps don't lie. The step with the longest average gap between "entered" and "left" is your real bottleneck. It's almost never the one people point to in the planning meeting. That's usually the step someone remembers being annoying once, not the one quietly eating the most time. Automate that step first. Everything else can wait. What's the process you automated that turned out to be solving the wrong problem?
1 like • 37m
This is exactly the discipline aerospace forces into you: you don't automate the story people remember, you validate the system that actually exists. The part I would add is failure modes: pull the timestamps, then mark where each case reworked, waited on a human, waited on missing data, or got bounced because the input was wrong. Throughput usually isn't lost in the obvious task, it's lost in handoffs, queues, unclear ownership, and bad inputs upstream. Once you can prove the constraint with evidence, then automation becomes a scalpel instead of a guess. Automate the step that removes the real constraint, then measure the next bottleneck after the system shifts.
#7dayAISChallenge - Day 1
What I Built: I built a newsletter automation in Claude Code that researches a topic, generates AI infographics, creates a styled HTML email with Jinja2, and sends it through the Gmail API. My first end-to-end newsletter focused on Epigenetics and longevity. One Thing That Clicked: Debugging Gmail. I had a bunch of problems setting up the API. This taught me to isolate one variable at a time rather than making random changes. After a few tries, I discovered I was authenticated with an account that didn't actually have a Gmail mailbox.😅 One Thing I’d Improve Create a schedule that sends me the newsletter, maybe every Monday, and include a random factor to pick topics I am interested in (AI, Cybersecurity, Health, Manga, Baseball). TIP: Every time I am using AI to generate things, I wish I could understand more, so I can talk deeply about them. I always tell Claude to generate a Learn folder and to add 10 things I should learn from this project. Sometimes, I even ask to look for a good YouTube video <10 min and add the link for me to view.
#7dayAISChallenge - Day 1
1 like • 16h
That Gmail mailbox discovery is exactly the kind of debugging lesson that sticks: isolate one variable, prove it, then move. For the Monday version, I would keep it simple: scheduler picks from your topic backlog, drafts the issue, then waits for your human yes before Gmail sends it.
Built duplicate detection into my lead system today — and hit a bug that made "empty" data lie to me
Problem: same buyer messages twice, system logs them as two different leads. Messy CRM, agent calling the same person twice off two different records. Fixed it: 1. Clean the phone number first — strip spaces and country code, so "+91 98451 23670" and "9845123670" register as the same person 2. Search the database BEFORE saving anything new 3. Match found → update their existing record with fresh details 4. No match → create a new one The gotcha: a search with zero real results doesn't always come back truly empty — sometimes it hands you a placeholder that LOOKS like a result but has no actual data in it. My logic was reading that placeholder as "found a match" and routing every single new lead down the wrong path. Took way longer to catch that than to build the actual logic. 👇 Anyone else gotten burned by "empty" data that technically wasn't empty? Curious what tripped you up.
Built duplicate detection into my lead system today — and hit a bug that made "empty" data lie to me
1 like • 16h
That placeholder trap is exactly where I like to separate "no data" into distinct states before anything is allowed to write: empty response, null response, placeholder response, malformed response, and valid match. A valid match should have to prove itself with the actual required fields, not just exist as an object. In systems work, I usually treat the write path like a guarded gate: normalize first, search second, validate the search result shape third, then update or create. The bug you found is sneaky because the automation technically did what you told it to do, it just accepted a weak signal as truth. Great catch, and this is the kind of edge case that makes the whole lead system more trustworthy.
A three-question check for your second brain
After Nate's second-brain walkthrough, I tried a simple check: could my setup find information I'd already saved? For three questions, the correct files didn't appear in the first three search results. The information existed, but those files weren't included in the search. After adding the missing file locations and adjusting the ranking, the correct sources appeared in the top three for all three questions. Small test, but it gave me something specific to fix. Here's a version you can try: 1. Pick three questions your saved notes should answer. 2. Ask them without supplying the answers or file names. 3. Ask for the supporting file and passage, then check them yourself. 4. If something is missing, check whether the search includes that file before adding more information. Prompt: “Answer from my saved files. For each answer, show the supporting file and passage. Flag missing or conflicting evidence.” Worth trying before expanding your library. Having a note saved doesn't guarantee it will be found.
1 like • 16h
This is the right instinct. I would treat retrieval like a tested capability, not a storage problem: define the question, expected source, acceptable rank, and what counts as a failure. Top three is a great acceptance criterion because it forces the system to prove it can surface the right evidence under normal use. The part most people skip is your last step: checking coverage before adding more notes. Otherwise the library gets bigger while the retrieval problem stays hidden.
1-10 of 1,247
Jason Elam
7
2,843 points to level up
@jason-elam-2821
AI Strategist & Consultant. I help SMB owners build custom AI Powered Solutions to help them empower their employees, NOT replace them.

Online now
Joined Feb 16, 2026
Powered by