Client work

Recent work.

Four engagements: what we found, what we built, and what changed.

Ad account cleanup

Furniture retailer
What we found

The ad account counted shopping carts as if they were sales. One month showed $627K in reported results against $22.8K of real purchases, and the newest campaign was learning from those phantom numbers.

What we built

A full teardown of the ads, the product feed, and the analytics. Then a product-level margin model built from true landed costs, feeding a weekly dashboard.

What changed

Pricing and ad decisions now run on real numbers. The project grew into a monthly role steering the whole ecommerce operation.

Order automation

Print-on-demand shop
What we found

Sales lived in one system and production costs in another, and nothing matched. No one could say which products made money.

What we built

A product-by-product profit map. All 285 production orders matched to their exact cost, twelve months of fees reconciled, every claim tied to a real order number.

What changed

The owner knows her margin on every product and which ads pay for themselves. Next build on the list: orders that flow to the printer without anyone touching them.

Client intake AI

Law office
What we found

Five phone lines and two people answering them. After hours and during busy stretches, callers hit voicemail, and people who need a lawyer call the next number on the list.

What we built

An AI intake line that answers in English or Spanish, collects the case details, flags urgent matters, and emails the office a structured summary about 30 seconds after the call ends. It introduces itself as AI and never gives legal advice.

What changed

Test calls end with intake details in the inbox before anyone has picked up a phone. The firm is piloting it for overflow and after-hours calls.

CEO dashboard

Training company
What we found

The CEO was checking the business by hand twice a day, across four systems that did not talk to each other. An earlier dashboard had died inside two months when its data connections broke quietly and nobody noticed.

What we built

One screen pulling every system on its own schedule, with a health panel showing the state of each connection and an email to the CEO the moment a feed stops. Each user sees only the pages they should.

What changed

The twice-daily manual check is gone. Every figure on the screen is live, and the whole system runs on the company's own accounts.

Bring us your version of these problems.