How we work
The Δ Method:
how we deliver.
Diagnosis, discovery, architecture, development with testing, security, observability and measurement. Six phases, each with a gate to move on.
01The Δ Method
Pick a phase to see what happens in it.
Diagnosis and discovery
We measure the starting point and understand the problem inside the operation, before any solution.
What we do
- Measured diagnosis of site, systems, data and funnel
- Interviews with the people who use it and the people who decide
- Map of the current process: where time and money are lost
- AI opportunities with estimated return
- Result letter: what changes for the customer, written before any code
You get
- Diagnosis report with baseline
- Result letter
- Opportunities ranked by return
Gate to move on
Baseline measured and problem approved by the decision-maker.
- Reference
- Double Diamond (Design Council) · Working Backwards (Amazon)
- Reference timeline
- 1 to 2 weeks
Human support throughout: one person you know, not a ticket.
02PDCA cycle
Plan, do, check, act. Every cycle.
The Δ Method runs in PDCA cycles, the continuous improvement loop from quality engineering. Each turn closes with the measured number against the target: what worked becomes the standard, what did not goes back into the plan.
Δ Method dashboard
The KPIs we track.
On every project.
Every project ships with a dashboard: eight dimensions, each indicator with a target and a way to measure it. Technical targets follow public standards (Google’s Core Web Vitals, OWASP, WCAG, DORA). Business targets come out of the diagnosis and go into the proposal.
dashboard · Δ method / diagnosis
measuring
Diagnosis
Before writing any code.
Target
100%
Baseline measured
How we measure: Dated measurement, median of 3
Target
before the roadmap
Risks mapped
How we measure: Threat model and assumptions
Target
in the proposal
Target per indicator
How we measure: After minus before
Default targets of the Δ Method. Each project adjusts the business targets to its own baseline.
03AI in production
Why our AI makes it past the pilot.
Most AI pilots never reach production. The ones that do follow the same rules, and they apply to every agent we ship.
01
One workflow, one metric, one owner
No "AI for everything". One process, one number that matters and one person accountable for it.
02
Real cases before code
Without a set of cases with the right answer, the agent does not start. That is how accuracy is measured.
03
A person at the decision point
The agent prepares and suggests; where there is risk, a person approves.
04
Cost per task, measured
The right model for each step, with an allowance. You know what each conversation costs.
05
Regression on every change
Changed the model or the prompt? The cases run again before release.
04The Δ
Every delivery ends with this equation.
It is the name of the company and it is the commitment. Without a baseline, the work does not start.
Δ = after − before
Δ −13.0s
05Principles
What we do not negotiate.
01
Measure before claiming
No number without a source and a date. Where nothing is measured, we say so.
02
Nothing hard-coded
What changes from one client to the next becomes configuration. Your system is yours, not a workaround.
03
Your code, your data
A dedicated database for each client. If the contract ends, you take everything.
04
Privacy by default
Personal data protected from the design up, not patched on later.
05
A guaranteed way back
Every release can roll back to the previous version in minutes.
Tell us what you want to build.
A person reads it and replies. No bots, no queue.