Selected work
What we shipped, and what changed
Three projects with the numbers attached. Every figure below is measured against a stated baseline. If we cannot show the baseline, we do not claim the result.
Replacing a fleet spreadsheet with a live ops portal
- Industry
- Logistics
- Service
- Web Application
- Timeline
- 6 weeks
- Scope
- 3 depots, 40 drivers
The problem
Dispatch ran on one shared spreadsheet. Two coordinators overwrote each other most mornings, proof-of-delivery photos lived in WhatsApp, and nobody could answer where a load was without three phone calls.
What we built
We built a dispatch board with per-user accounts and an audit trail, a driver view that works on a phone with poor signal, and photo capture attached to the job rather than a chat thread. Existing depot codes were kept so nobody had to relearn the workflow.
The outcome
Manual data entry dropped by 62% against the four weeks before rollout. Coordinators stopped reconciling conflicting rows, and delivery disputes could be settled from the record instead of a search through chat history.
−62%
manual entry vs. 4 weeks pre-rollout
3
depots live in 6 weeks
0
overwrite conflicts since launch
“The morning scramble just stopped. That was the whole problem and it went away in the first week.”
A storefront rebuild that paid for itself in a quarter
- Industry
- Retail
- Service
- Web Application
- Timeline
- 9 weeks
- Scope
- Storefront + checkout
The problem
The old storefront took 4.2 seconds to load on mobile and lost most carts at the payment step. Nobody could say where buyers dropped off, because nothing was instrumented.
What we built
We rebuilt the storefront and checkout, cut the payment flow from five steps to two, and added event tracking at every stage so the funnel became visible. Product data was migrated from the existing catalogue without a freeze.
The outcome
Online orders reached 3.1× the prior quarter, and the build paid for itself inside the first quarter. The team can now see which step is leaking before it costs a month of sales.
3.1×
online orders vs. prior quarter
0.9s
mobile load, down from 4.2s
1 quarter
to pay back the build
“We finally know which step is losing us money. Before, we were guessing.”
An AI assistant that reads the contracts nobody wants to
- Industry
- Professional services
- Service
- Custom AI Solution
- Timeline
- 2 weeks discovery + 7 weeks
- Scope
- ~1,200 contracts
The problem
Two associates spent most of Monday reading supplier contracts for the same six clauses. The work was necessary, unbillable, and the first thing to slip when the week got busy.
What we built
A discovery week confirmed the documents were consistent enough to be worth automating. We built clause extraction with a confidence score, routed anything uncertain to a human review queue, and shipped an evaluation harness so accuracy is measured rather than assumed.
The outcome
The team saves about 14 hours a week against a four-week baseline. Nothing is auto-approved — the assistant drafts and a person signs off, which is why the team actually uses it.
14 hrs
saved weekly vs. 4-week baseline
~1,200
contracts processed
100%
flagged clauses human-reviewed
“It drafts, we approve. That split is what made the team trust it.”
How we report
Why these numbers are boring on purpose
Anyone can put a big percentage on a page. These are the rules we hold ourselves to when we write one down.
Every metric has a baseline
A percentage with nothing to compare against is decoration. We state the period we measured before, and the period we measured after.
We round honestly
Approximate numbers are marked approximate. Suspicious precision is a sign someone is working backwards from the conclusion.
We do not publish what we cannot show
No invented testimonials and no logos we have not earned. Where a client cannot be named, the numbers still have to hold up.
Your project could be the next one here
Tell us what is not working today. We will tell you what it would take to fix it, and what it would cost.
Free consultation · No commitment · Replies within a day