A Custom AI Toolkit I Built to Run the Business
A suite of agentic skills that automate the research, validation, and grunt work an agency repeats every week.
The problem
Why this needed to exist
Running a lean agency means doing the same heavy, repetitive thinking over and over: is this niche worth pursuing, where's the unmet demand, is this lead list even worth dialing? Off-the-shelf tools each do one slice, none of them fit my exact workflow, and stitching them together by hand every time is where the hours disappear.
So instead of buying ten half-fits, I built my own AI tools — each one a focused, multi-step agent that does a job end to end.
The approach
How I built it
Each skill is an agentic workflow on the PAI framework, not a single prompt — it chains web research, SEO and data APIs, scraping, and a scoring rubric into one repeatable command. They share a common spine (gather signal → score → rank → output evidence) so they're reliable enough to actually depend on.
The toolkit covers the full front-of-business: SignalDig (problem-first opportunity mining from real demand), ArbitrageMiner (single-feature API-arbitrage discovery), IdeaPipeline (daily idea generation with a running scoreboard), NicheResearch (scoring one vertical against a specific offer), and LeadWrangler (turning a raw scraped CSV into a deduped, phone-classified, outreach-ready call list). My client-assessment pipeline is part of the same family.
- PAI agent framework
- Web research + SEO/data APIs
- Web scraping
- Multi-dimensional scoring
- TCPA-aware phone classification
The outcome
What it actually does
These aren't slideware — they run. SignalDig has scored well over a hundred niches into a ranked board; NicheResearch has graded verticals against my core offer; LeadWrangler turns messy Maps/Apollo exports into dial-ready lists with mobile-vs-landline and VoIP classification baked in. Each is a single command I can fire any day.
Together they form an internal automation stack that makes a one-person operation punch well above its headcount.
- Multiple live, custom-built skills in daily use
- SignalDig: 100+ niches mined and ranked with evidence
- LeadWrangler: raw CSV → deduped, phone-classified, outreach-ready
- Shared scoring spine so results are consistent and trustworthy
What I learned
The curve I already climbed
The real skill isn't writing a clever prompt — it's composing multi-tool agent workflows that are dependable enough to bet decisions on, and blending soft signals (what people complain about) with hard ones (search volume, competitor strength) so the output is actually trustworthy. Most AI automations fall over the moment they hit a messy real-world input; making them robust is the work.
Because I've built a whole suite of these, I know how to scope an automation that augments judgment instead of replacing it. That's directly transferable: I can build the same kind of custom tool around a client's specific workflow, not just hand them another generic app.
I don't just plug in off-the-shelf AI — I build custom automations around your exact workflow, because I've already built a working toolkit around mine.
Want this kind of thing built for your business?
Book a quick call and I'll show you what would actually move the needle for you.