Skip to main content

AI Solutions & Consulting

Proof, honestly framed

Case studies from our own operations


These are systems we built and run for our own businesses. They aren't client engagements, and we don't publish numbers we didn't measure.

Before the system

A lead can arrive from a paid ad, a website form, or a phone call within the same few minutes, and each one needs its own response. Whichever channel gets missed is the one that goes quiet, because nothing was chasing it.

With the system

It runs unattended in the background. The owner opens the console and replies to whoever wrote in.

An open notebook, a pen, and a cup of coffee on a desk.

Notes from the study

We built these for our own businesses first. Everything on this page is running now, in something I operate, which is the only reason I am willing to describe it at all.

Anthony

An open notebook, a pen, and a cup of coffee on a desk.

01

Lead intake that keeps its source

The problem

A lead can arrive from a paid ad, a website form, or a phone call within the same few minutes, and each one needs its own response. Whichever channel gets missed is the one that goes quiet, because nothing was chasing it.

What we built

  • Every ad lead form and website form posts straight into a phone console tied to the business's own number, so the reply goes out from the number the customer already saw.
  • A short alert reaches the owner the moment a real lead comes in, on a separate channel for a form submission versus an inbound call.
  • Each lead keeps the channel it arrived on attached to it, from first contact through to a closed job.

How it runs today

It runs unattended in the background. The owner opens the console and replies to whoever wrote in.

  1. Form / call
  2. Console
  3. Owner reply
  4. Job logged

Own operations, not a client result. We don't publish numbers we didn't measure.

People at a whiteboard, seen from behind, mapping a process.

02

A content pipeline that never posts blind

The problem

Posting to three platforms a day by hand is a job on its own, and skipping the review step to save time is exactly how the wrong post goes out. We wanted the drafting automated without ever losing the human check.

What we built

  • A drafting step that writes the post text and builds the image for Instagram, Facebook, and LinkedIn from the business's own material.
  • A shared board where every draft sits as pending approval until a person changes its status by hand.
  • A scheduler that checks that status at the moment a post is due, and holds anything that was never approved instead of guessing.

How it runs today

Nothing reaches a feed without a person clicking approve first. An unapproved draft just waits.

  1. Draft
  2. Pending approval
  3. Approved
  4. Published

Own operations, not a client result. We don't publish numbers we didn't measure.

A closed laptop on a tidy desk beside a plant.

03

An outreach engine that emails once

The problem

Reaching business owners who are actually hiring, and only them, takes a live signal matched against a clean list, and getting either one wrong burns the sender's reputation fast. We built it to run on its own without ever double-mailing anyone.

What we built

  • A daily scan for businesses showing a hiring or growth signal, checked against a standing suppression list before anything is drafted.
  • One personalized email per person, generated once and never followed up.
  • A daily send limit paced across the business day, plus a reply monitor that tells a genuine response apart from an automated one.

How it runs today

It sends on its own, at a pace set to protect the sending domain, and a real reply is the only thing that pulls a person in.

  1. Signal scan
  2. Suppression check
  3. One email
  4. Reply monitor

Own operations, not a client result. We don't publish numbers we didn't measure.

An office building at dusk with the interior lights on.

04

A billing and phone console for a plumbing business we operate

The problem

Coordinating licensed contractors for a plumbing business we operate still means every job needs an invoice, a payment that gets chased, and a customer who can reach a real person by text. We didn't want three separate tools for that.

What we built

  • A two-way texting and call console built on the business's own number, so a customer's reply always lands in the same conversation.
  • An invoicing step that refuses to generate a bill without a technician and the job attached to it.
  • Automatic aging and payment reminders that chase an unpaid invoice on a schedule and stop the moment it's marked paid.

How it runs today

The owner works from one console for conversations and one for billing, and a paid invoice quiets its own reminders.

  1. Job done
  2. Invoice
  3. Aging / reminders
  4. Paid

Own operations, not a client result. We don't publish numbers we didn't measure.

Hands at a keyboard on a working desk, lit from a window.

05

An article engine with a claims gate

The problem

Publishing useful articles several times a day is impossible by hand, and an AI writer left alone will invent statistics and clients. We wanted the volume without ever letting a made-up claim reach the site.

What we built

  • A generator that drafts third-person articles from the business's own approved material.
  • A hard gate that rejects any draft containing percentages, dollar figures, client names, testimonials, or banned phrasing before it can be saved.
  • An automatic publish-and-deploy step that pushes approved articles to the live site on a schedule.

How it runs today

It publishes on a schedule without a person in the loop, because the gate, not a reviewer, is what keeps the claims honest.

  1. Draft
  2. Claims gate
  3. Approved
  4. Published

Own operations, not a client result. We don't publish numbers we didn't measure.

A glass-walled office corridor with daylight along one side.

06

An email deliverability watchdog

The problem

Cold outreach dies quietly when a domain's authentication slips or bounces pile up, and nobody notices until replies stop. We wanted the sending to police itself instead.

What we built

  • A recurring audit of the sending domains' SPF, DKIM, DMARC and MTA-STS posture that scores each domain and alerts when it degrades.
  • A daily inbox-health check that sets the next day’s send allowance and pauses sending entirely when health turns red, with a text to the owner.
  • A bounce tracker that adds hard bounces to a suppression list enforced at send time.

How it runs today

The send volume follows the health score automatically, and the owner only hears about it when something turns red.

  1. Domain audit
  2. Health score
  3. Send allowance
  4. Suppression

Own operations, not a client result. We don't publish numbers we didn't measure.

People walking down an office stairwell, seen from above.

07

A Google Ads waste monitor

The problem

Automated bidding happily spends on competitor-brand searches, do-it-yourself queries, parts shoppers, and out-of-area clicks unless someone keeps pulling those out. Doing that by hand every week is exactly the job nobody gets to.

What we built

  • A recurring search-term audit grouped by waste theme, with negative keywords applied per theme.
  • A conversion-hygiene rule that keeps page views and tap-to-call events out of the optimized goal set, so bidding trains on real phone leads only.
  • A closed-job ledger that matches inbound calls to jobs actually won, so spend is judged on outcomes instead of clicks.

How it runs today

A monthly review plus continuous negative-keyword upkeep for a plumbing business we operate.

  1. Search terms
  2. Waste themes
  3. Negatives
  4. Closed jobs

Own operations, not a client result. We don't publish numbers we didn't measure.

An empty seating area in an office lobby, with daylight through tall windows.

08

Local presence on autopilot, with a person approving

The problem

A Google Business Profile that goes quiet loses ground, and asking for reviews by hand never happens consistently. We wanted the cadence handled without letting anything publish unread.

What we built

  • Scheduled profile posts for updates, offers, and events are drafted from templates with local details added, then held for approval before publishing.
  • Review-request links sent by text or email after a job is completed.
  • A Q&A routine that answers common customer questions directly on the profile.

How it runs today

Posts and review requests go out on a cadence for a plumbing business we operate, with a person approving the copy.

  1. Draft post
  2. Approval
  3. Published
  4. Review request

Own operations, not a client result. We don't publish numbers we didn't measure.

A person on site holding a tablet, seen from behind.

09

An interactive demo built into a prospect's page

The problem

A manufacturer deciding whether to talk to us wanted to see the analysis, not read about it. A slide would not have answered the question they were actually asking.

What we built

  • A dedicated page on our site for that prospect, with an interactive module that takes their input and returns an AI analysis in real time.
  • A server-side function that calls the model, with a fallback model if the primary is unavailable.
  • A watcher that checks the endpoint on a schedule and reports its status.

How it runs today

It is live and monitored, and it is the same shape we install inside client sites when a demo has to prove itself before a discovery call.

  1. Prospect input
  2. Model call
  3. Analysis back
  4. Watcher

Own operations, not a client result. We don't publish numbers we didn't measure.

Last updated

An empty boardroom in the morning, chairs pushed in around a long table.

Before the discovery call

Want to see what this looks like in your business?

It starts with a discovery call and a scoped pilot.

Book a discovery call
Book a discovery call