Metis Method

Your team hates babysitting AI that thinks like an intern.Supercharge AI with their expertise!

Interviewed by phone. Codified as rules. Inside the AI they already use.

The difference, on one question

Your team can now query AI and get genuinely insightful answers, leveraging the expertise of your best people.

Week two of a $48M pursuit. The buyer’s CFO emails: “We like the team. We need 10% off to get this through the board. Can you do it?”

AI, out of the box
“Thank you for the feedback. We value the partnership and are committed to finding a pricing structure that works for Northwind. We would welcome a discussion on a revised proposal that reflects a 10% adjustment while preserving the scope and quality of service.”

Polite, generic, and it just gave away 10% of a $48M contract.

With your best person’s judgment loaded
We do not reduce price without a change in scope, and your board should not approve a thinner version of what your team chose. Two paths to your number: move the transition fee into year 2, which cuts year-1 cost by 13%, or a gain-share on the app rationalization. Either way the board sees 3-year total cost.”
  • Rule 03 · price moves only against scope · K. Rosner
  • Rule 11 · offer a structure, never a discount · K. Rosner
  • Rule 14 · CFO in the loop: lead with 3-year total cost · D. Okafor
3 rules. Every deal. Not this one.

The facts were in the files both times. The judgment was not. It came from one 30-minute call with the person who has answered this email 40 times, and it now applies to every pursuit.

How it works

Interviewed by phone. Codified as rules. Live in the AI they already use. Kept current every week.

01

30 minutes on the phone

Anchored on a real decision they made. We show them what the model would have said and capture the reaction.

02

Their judgment, as rules

Checkable conditions with triggers and exceptions. Each carries a name, a date, and a freshness score.

03

Switched on, per person

A connector each person enables in Copilot or whatever you run. Hosted or in your tenant.

04

Refreshed every week

A standing 30-minute call. What worked, what did not, what changed. Approved diffs land the same day.

No access to your documents needed. The interviews happen with your people; the rules are served back into the stack you already run.

Kept current

30 minutes a week, per person. Their AI never goes stale.

  1. 1

    The call

    What worked, what did not, what surprised you, what you learned. Human-led at first, then a voice agent that already holds their record.

  2. 2

    The diff

    Each call drafts rule changes. Nothing is written for them.

  3. 3

    The approval

    They approve or decline. Nothing reaches their manager without their yes.

  4. 4

    Live the same day

    Approved diffs reach their AI before the next draft.

When a rule passes its decay date unconfirmed, the AI says so in its answer.

See the engine running
Approval queueTuesday call, 27 min
Finalist Q&AAdd ruleApprove

When the buyer's CFO sits on the committee, lead every pricing answer with 3-year total cost, not the year-1 rate.

Owner: K. Rosner · Source: Northwind finalist, Sep 2
Finalist Q&AReviseApprove

Reference-call offer: name the reference in the answer. “Available on request” reads as a dodge.

Owner: K. Rosner · Replaces rule dated May 12
Internal commsFlag staleApprove

Hourly-employee tone rule last confirmed 9 weeks ago. Confirm or retire.

Decay: 6 weeks · Owner: J. Alvarez
The pilot

Let’s test if we can improve your AI.

Bring the 3 questions your best person answers better than anyone. We show you what your AI says today, and what it says with their judgment loaded.

If the second version is the one you would send, we run four weeks with one function: their next three deliverables, measured against your own baseline, in writing, whether or not it flatters us.

Book a call
One number, agreed up front

How much of the AI draft your people still rewrite

71%Baseline58%Week 141%Week 230%Week 324%Week 4

Measured on every draft, automatically, on the sections the file covers. Baseline from their last five deliverables.