How we work
No demos. A method.
Every engagement follows the same governed path from discovery to operation. Here is the entire thing.
The method
Nine steps. The same nine, every time.
- 01
Discovery
A listening call, not a pitch: where the time actually goes, what breaks under deadline, and what you'd never let software do alone. If we don't see a fit, we say so and stop here.
- 02
Workflow Map
We observe the real workflow and time the manual baseline — how inputs arrive, who touches them, and where the minutes live. The map is built from watching the work, not from a questionnaire.
- 03
Data Inventory
Every document and system the workflow touches gets catalogued, with a representative sample de-identified for the work ahead. Retention and deletion terms go in writing before any real file moves.
- 04
Risk Classification
Each kind of data is classified, and that class decides where it may be processed — what stays inside your boundary and what never leaves. The hard lines are drawn here, before anything is built.
- 05
Pilot SOW
One bounded slice of the workflow, in writing: what the system will do, what it will never do, and success metrics measured against your own baseline — not our claims. Nothing expands past this scope without a new agreement.
- 06
Sandbox
The system is built and exercised on the de-identified sample — never your live operation. It has to work here before it earns anything more.
- 07
Shadow Mode
The system runs alongside your team on real work with zero external effect: it drafts and flags in parallel, and we compare its output to what your people actually did. Misses cost nothing and teach us everything.
- 08
Human-Reviewed Pilot
Live work, with a person approving every output — the system drafts, your team decides, and nothing external ever sends itself. Every assist is logged, so the pilot ends with evidence instead of anecdotes.
- 09
ROI Review
Measured results against the baseline from the workflow map, plus the full assist log, reviewed together. We expand deliberately or we stop — go/no-go is a shared decision, not a renewal default.
Who runs this
Eric Yun
FOUNDER & CTO
A full-stack machine-learning engineer, Eric built and operates Pleadly — litigation support that runs under attorney-client privilege constraints — and designs Miko's inference, routing, and approval-gate architecture personally. Every engagement in this method is run by the person who built the system, not handed to a delivery team.
Private briefings under NDA.
We'll walk through the architecture, our current certification status, and a sample deliverable — candidly.
Book a briefing →