The Denuto intake methodology · v1

How we run the twice-weekly ideation intake.

Two 60-minute sessions a week. Four pitches a session. Every pitch leaves the room with one of four outcomes — never as an open question. This page is the pitcher-facing playbook: what to bring, how the hour runs, and what you'll walk out with.

Submit a canvas →See example ideas →
Step 1 · Before you book a slot

The six-point sniff test.

Answer these before you submit. All six yes — book a slot. Five yes and one “not yet” — talk to the platform team about the missing piece. Fewer than five — this is a pre-intake conversation, not an intake slot.

  1. 01Can I state the workflow as "read X → reason over Y → write Z"?
  2. 02Can I hand over three real, redacted documents today?
  3. 03Is the output shape stable enough that I can show a good example?
  4. 04Is the volume more than a handful per quarter, OR is the single case regulatory-weighted?
  5. 05Is there a playbook, a template, or at least a worked example of “good”?
  6. 06Does a human review the output before it leaves NAB?
Step 2 · What you send 48h ahead

The Pipeline Canvas — one page, no deck.

Email denuto@nab.com.au with subject Intake canvas — <workflow name>. Missing sample docs or the “Today” section — bounced, not queued. We hold this rule.

Read
What document(s) come in?
  • Format(s): PDF, DOCX, email, scan, portal export
  • Typical length and structure (templated vs. free text)
  • Volume today — per week or per month
  • Three sample documents, attached and redacted
Reason
What question(s) get answered?
  • Primary question in one sentence
  • Rules / playbook source: does one exist? who owns it?
  • Judgment vs. lookup: rule-checkable, or expert call?
Write
What comes out?
  • Output artifact — memo, redline DOCX, checklist, email
  • Template attached if one exists
  • Audience and what they do with it next
  • The “good” signal — how a reviewer knows the output is usable
Today
How does this get done now?
  • Who does it (role, team)
  • How long per case, per week
  • What recurringly breaks
  • What won't change — regulator, policy, format constraints

Plus an “Ask” line — what you want leaving the room: prototype, feature in an existing pipeline, just new playbook rules, or advice.

Step 3 · The hour itself

60 minutes. One visible timer.

Up to four pitches, 12 minutes each. The host cuts on time. Over-time defaults to Park — that's the point.

  1. 0:00
    Open (5 min)
    Roll call, read the queue, check in on last session's prototypes-in-flight.
  2. 0:05
    Pitch 1 (12 min)
    5 min pitch · 5 min interrogate · 2 min decide + scribe.
  3. 0:17
    Pitch 2 (12 min)
    Same structure, same clock.
  4. 0:29
    Pitch 3 (12 min)
    Host cuts on time. Over-time defaults to Park.
  5. 0:41
    Pitch 4 (12 min)
    Only if the queue has four. Otherwise this time reclaims.
  6. 0:53
    Close (7 min)
    Scribe reads each Prototype Plan back; owners confirm; next date + bounced canvases.

Inside each 12-minute pitch block

01
Read
Can we parse these documents today?
Platform voice
On these three samples — not in theory. If no, the pipeline doesn't exist yet, however good the idea.
02
Reason
Is there a playbook, or is this judgment-only?
Platform + Legal
Judgment-only work can be an assist surface (extractions, citations). It cannot be auto-draft. Asking for auto-draft on judgment is a red flag.
03
Write
Is the output shape stable across cases?
Platform voice
If every case has a bespoke output, it's not a pipeline — it's a prompt. Re-scope to extraction-only, or park.
04
Risk surface
Regulator, privacy, counterparty, reputational?
Legal / Risk voice
Does the output leave NAB? Does it make a commitment? Does it need a human sign-off gate?

Five minutes pitch · five minutes interrogate (roughly 75s per question, in order) · two minutes to decide and scribe the Prototype Plan live in the room.

Step 4 · What leaves the room

Four outcomes. No fifth bucket.

Every pitch exits as exactly one of these. No “let's think about it.” No open questions.

01
Prototype lane
Platform-shaped, feasible on real documents, named owner.
  • 2-week prototype slot on a platform engineer
  • 2-week check-in in the next session's open block
  • Your three sample docs become the first fixtures
02
Shared-pipeline feature
Fits inside an existing pipeline — Contract Review, Tenderiser, or Intake itself.
  • Named pipeline and feature owner on that pipeline's team
  • Linked into that pipeline's backlog
  • No new surface spun up — the work routes through existing pipelines
03
Playbook-only
Doesn't need code. Needs rules, examples, or fixtures in an existing pipeline's playbook.
  • Handed to the playbook owner on the relevant pipeline
  • Followed up in that pipeline's cadence, not back in intake
04
Park
Not ready. The specific missing piece is named so you know what to come back with.
  • Common reasons: no real documents yet, shifting output shape, no playbook or examples, one-off volume, regulated artifact with no human gate
  • A park is not a no — it's a specific ask. Come back with the missing piece.
Step 5 · The artifact

The Prototype Plan — written in-room.

For every pitch in the Prototype lane, the scribe fills this on a shared doc, visible to the room, before the block closes. If it isn't written in-room, the decision wasn't actually made.

# Prototype Plan — <workflow name>
Session: <YYYY-MM-DD>   Owner: <engineer>   Pitcher: <name>

## One-line problem
<the read-reason-write loop, one sentence>

## Read-reason-write sketch
- Read:    <what documents, what parser path>
- Reason:  <what agent(s), what playbook or rules>
- Write:   <what artifact, what template>

## First three fixtures
1. <sample doc>
2. <sample doc>
3. <sample doc>

## Success bar
What evidence would prove this is pipeline-worthy at the check-in?
- [ ] <concrete, measurable>
- [ ] <concrete, measurable>

## Risks we flagged
- <risk, owner, mitigation>

## Check-in
Date: <+2 weeks>   Format: 10-min slot in the next-next intake

Plans live at docs/intake/YYYY-MM-DD-<slug>.md on main. The check-in outcome appends to the same file so the whole life of the prototype reads as one document.

Anti-patterns

The usual rejects — named so we save 10 minutes each.

“AI for <thing>” with no specific document in hand
Park. Come back with documents. The platform is pipeline-shaped; without docs there's no pipeline.
Auto-draft on judgment work
Downgrade to assist — extractions plus citations. Let the human draft.
A bespoke upload surface for each new idea
Route through /intake. Duplicate uploaders are the architectural sin the hub exists to prevent.
“Bolt an LLM onto our existing tool”
That's a prompt, not a pipeline. If it doesn't read, reason, and write, it isn't Denuto-shaped.
Novelty demo with no repeatable workflow
Park until volume exists.
Mid-pitch scope expansion (“can it also do X?”)
One canvas per workflow. Pitch X separately in a future slot.
You've read the playbook

Write your canvas. Book a slot.

One page. Three documents. The read-reason-write loop. We'll do the rest in the hour.

Submit a canvas →Back to platform overview