Mindstack
Kestrel Advisory Group
Workspace
Workflows 6
Process Library LATER
Facilitator View LATER
Architect Handoff LATER
DK
Diane Karras
Senior consultant
Workflows / {{ crumb }}
Record flagged — proceeded with 2 readiness gaps
AUTOSAVED 14:22
Help
{{ s.num }} {{ s.label }}

Workflows

Client engagements in deconstruction. Each workflow moves through readiness, deconstruction, reframing, and rating before it becomes a build specification.

New workflow
Workflow
Status
Vertical
Last activity
Tasks
{{ w.name }}
{{ w.client }}
{{ w.status }}
{{ w.vertical }}
{{ w.activity }}
{{ w.tasks }}
Cohort activity
COMING LATER
Peer progress, facilitator review queue, and shared reframe patterns from the Q3 healthcare cohort will appear here. Reserved so the layout does not shift when it ships.
This quarter
Workflows deconstructed4
Tasks rated118
Specifications handed off2
Step 01 · Readiness gate

Before you take a workflow apart

Deconstruction work is only as good as the mandate behind it. Confirm what is in place. Gaps do not stop you — they are recorded on the engagement so the gap is visible later.

{{ g.num }} {{ g.title }}
{{ g.detail }}
{{ g.advisory }}
Confirmed
Missing
{{ gateBtnLabel }}
{{ gateSummary }}
Step 02 · Workflow setup

Define the workflow

Workflow name
Name it the way the client refers to it, not the way you would file it.
Vertical
Determines which standard process library we compare against.
{{ v.label }}
Business outcome
One or two sentences. What does the client actually get when this workflow runs?
{{ outcomeHint }} {{ outcomeCount }}
Research standard process
Back
Step 03 · Process suggestions

Standard tasks for healthcare compliance

Drawn from 34 vendor-oversight workflows in this vertical. Treat it as a starting inventory, not a description of your client. Accept, edit, or dismiss each line — and add what we missed.

{{ acceptedCount }}
accepted of {{ totalCount }}
AI proposed from vertical research
YOU added by you
{{ dismissedCount }} dismissed
{{ t.num }}
{{ t.src }}
{{ t.name }}
{{ t.aLabel }}
{{ t.editLabel }}
{{ t.dLabel }}
Add task
Build the canvas
Back
Step 04 · Deconstruct

Before you map anything

You are here to answer one question about your client's workflow: should this work exist in the shape it exists in now?

Most AI work skips that question. You take the workflow as written, make each step faster, and hand the client a quicker version of what they already had.

The four stages ahead force the harder question. Deconstruct the workflow as it runs today. Reframe and rate it against what the work is actually for. Architect a solution an engineer can build without calling your client. Win the room with a case your client's CFO will sign.

You bring the domain knowledge. The platform holds the structure and pushes back when your answer looks thin.

Start with one workflow you know cold.

Start mapping
{{ l.text }}
{{ n.code }}
{{ n.name }}
{{ n.inGlyph }} {{ n.inLabel }} {{ n.time }}
Step 04 · Deconstruct
Vendor BAA review and risk tiering
12
tasks
5h 45m
cycle time
{{ detailedCount }}/12
detailed
Input type
DOCDocuments SYSSystem data INTInterview JDGHuman judgment
−
{{ zoomLabel }}
+
Fit
Add node
Drag to pan · click a node to open its detail
Continue to reframe
{{ sel.code }} · Task detail
{{ sel.name }}
×
{{ sel.time }}
current time cost
{{ sel.freq }}
frequency
{{ sel.owner }}
performed by
Task name
Decision made or output produced
the thing that exists after this task
Inputs required
{{ i.label }}
Likely technology
what gets applied to tasks of this shape
{{ techName }} What’s this?
Handoff recipient
Handoff format
Current time cost
Where errors typically arise
Requirements interview
{{ turnCount }} exchanges · reviewing this task
{{ m.av }}
{{ m.text }}
{{ m.quote }}
Your reply
Send reply
Use suggested answer
Buildable. This task now reads as a rule with a named exception path — recorded for rating.
Save task
Cancel
{{ nextHint }}
{{ concept.kicker }}
{{ concept.term }}
×
Applied to {{ concept.appliedTo }}
In plain language

{{ concept.summary }}

How it works
Question from the reviewer Search the document library Relevant excerpts retrieved AI reads excerpts + question Answer with citation BAA Checklist Prior rulings The library it reads from The answer is only as good as what is in here — but it can no longer invent a clause that is not. Open book, not closed book: it looks things up before it answers. Raw document any layout Layout & field detection Structured output VENDOR SERVICES EFFECTIVE Northwind Labs Lab results 2026-03-01 Halden Systems Claims intake 2025-11-14 OCR scanned or photographed pages, turned into text Scanned documents join the same pipeline once OCR has read them. Not —read this whole file— but —give me these six answers—. Trigger new BAA received Gather context checklist + prior BAAs Take action compare clauses Clear enough? yes no Output: pass logged, no human needed Escalate to a person flagged for human review loop continues on the next BAA the human-in-the-loop point The agent part is not the intelligence — it is the loop, and knowing when to stop and ask instead of guessing.
Related concepts
RAG look up, then answer Embedding text turned into searchable numbers Chunking long documents cut into searchable pieces Retrieval the search step itself Vector database where the library lives Generation writing the answer from what it found Document processing OCR scanned images turned into text Layout parsing understanding where fields sit on the page Schema-driven extraction knowing which fields to look for Table extraction pulling structured tables out of unstructured pages Agent a loop, not an answer Trigger what starts it Context what it gathers before acting Action what it actually does Escalation threshold when it stops and asks a human Single-shot prompt vs. agentic loop the difference from a chatbot
Close
Reference only — not a recommendation for this task
Step 05 · Reframe

What is this workflow actually for?

The deconstruction describes how the work is done today. Reframing asks what would have to be true for the outcome to arrive without most of it. Pick the framing you will defend to the client.

Stated today
{{ wfOutcome }}
{{ f.title }}
{{ f.body }}
Implication · {{ f.implication }}
{{ f.retained }}
of 12 tasks retained
Your reframe statement
Compare both paths
Back to canvas
Step 06 · Dual path

Augment the workflow, or redesign it

Both are legitimate. They differ in what they require from the client, not in ambition. Choose one to carry into rating — you can record the other as the fallback.

{{ p.kicker }}
{{ p.tag }}
{{ p.title }}
{{ p.summary }}
{{ p.saving }}
cycle time
{{ p.effort }}
build effort
{{ p.change }}
change load
What changes
· {{ pt.t }}
Requires · {{ p.requires }}
Redesign needs an executive sponsor who can change how the work is done. You marked sponsorship as missing at the readiness gate — recording this path will carry that gap onto the engagement record.
Rate tasks on the {{ chosenPathShort }} path
Back
Step 07 · Rate

Score each task for automation candidacy

Five criteria, 1 to 5. Click a cell to set the score. Composite is weighted — data availability and repeatability count double, because they decide whether a build survives contact with the client's systems.

{{ ratedCount }}/12
tasks scored
Task
Repeatability
Data avail.
Rule clarity
Error cost
Volume
Composite
{{ r.code }} {{ r.name }}
Eliminated by the reframe — not rated
{{ r.composite }}
Click any segment to score 1–5. scored unscored
See candidacy ranking
Back
{{ rateHint }}
Step 08 · Candidacy ranking

What to build first

Ranked by composite score against hours returned per month. The first band is the build scope you take to the architect; everything below it is documented, not scheduled.

{{ b.label }}
{{ b.meta }}
{{ r.rank }}
{{ r.name }}
{{ r.note }}
{{ r.hours }}
{{ r.score }}
If you build band one
Hours returned / month{{ bandHours }}
Cycle time change−41%
Tasks in scope{{ bandCount }}
Watch items
T08 scores well but its threshold is still “roughly 5,000 records.” Fix the number before the architect sees it.
T10 remediation memos rank low on repeatability. Recommend augmentation, not automation.
Build the engagement record
Back to rating
Step 09 · Engagement record

Vendor BAA review and risk tiering

Everything decided in this workflow, in the order a reviewer will ask for it. This is the artifact the client keeps.

Prepare handoff
Readiness gaps recorded at the gate
{{ g.t }}
{{ r.kicker }}
{{ r.title }}
{{ r.body }}
Back to ranking
Export as client PDF
Step 10 · Architect handoff

Build specification

What gets built. Your domain knowledge goes into the spec completely enough that the build team never has to interrupt your client to ask.

{{ specToggleLabel }}
{{ sendLabel }}
One unresolved definition blocks handoff
T08 routes on “roughly 5,000 patient records.” An engineer cannot implement “roughly.” Set the exact threshold, or mark the routing as a human decision and drop T08 from the build scope.
Set threshold
BLOCKING
Threshold set to {{ thresholdSet }} records. T08 is buildable and back in scope.
{{ sp.code }} {{ sp.name }}
{{ sp.score }}
Rule
{{ sp.rule }}
Exception
{{ sp.exception }}
Systems
{{ sp.systems }}
Acceptance
{{ sp.acceptance }}
Package contents
{{ p.t }}
{{ p.sub }}
Recipient: Ana Villaseñor, solutions architect. She receives the spec, the canvas, and the readiness gaps — not the interview transcripts.