Research to Decision Memo
Turns a question into a one-page decision memo that separates evidence, assumptions, conflicting claims and unknowns, and cites the originals. The decision stays with you.
Who it is for
W20 Research to Decision Memo is an iSystematic workflow for every practice and firm we work with: clinics, law firms, accounting firms, property management, service businesses and independent professionals.
Organisations use it inside Vendor Intake and AI Diligence, the Research-to-Decision Office and Board and Committee Pack Preparation.
What it prepares
It prepares a one-page decision memo and an evidence ledger.
What always needs you
These always stay with you, at every level.
- The final decision
- Any consequential action that follows it
How it runs
It runs in five steps across three assistants, and you carry the work from one to the next.
Discover sources in Perplexity
Perplexity finds current sources for your question, scope and criteria.
Read the sources
You read the originals: a citation is a lead to inspect.
Compare in Claude
Claude compares what the sources say.
Write the memo in ChatGPT or Claude
The memo separates evidence, assumptions, conflicting claims and unknowns.
A person decides
The decision, and any action that follows, stays with you.
The free card
Prompt: “Separate evidence, assumptions, conflicting claims, and unknowns. Cite the originals. Explain what would change the recommendation.” Manual version: write down the question, its scope and your criteria. Search for sources in Perplexity and read the originals. Paste the prompt and the source documents into Claude to compare them, then into ChatGPT or Claude for the memo. You carry the work from one assistant to the next yourself, and you make the decision.
What you get
You get a memo you can check against its sources.
- A one-page decision memo that says what would change the recommendation
- An evidence ledger that ties each claim to its original source
When it stops
A claim that no source supports stays marked as unknown.
How we test it
We test the free card with fictional data, first against this workflow's own acceptance tests, then against the catalogue's test checklist items that apply to the manual version. The card is marked Ready only once it passes.
- Every citation actually supports the claim it is attached to.
- Two models agreeing is not treated as independent evidence.
- It works by copy and paste alone, with no connector.
- Every source it cites and every detail it pulls out is checked against the original.
- It is given missing and contradictory information, to check that it does not fill the gaps.
- It is asked for clinical, legal and financial judgements, and must refuse them.
- A hostile instruction is hidden in the content it reads, and it must not obey it.
- It is given duplicates, timezones and out-of-date information.
- Nothing is sent, spent, published or changed before you approve it.
Levels
It runs at two levels, and connects to research tools when they are available.
| Level | In this workflow |
|---|---|
| Manual (M) | Yes: research, read, compare and write across the assistants yourself |
| Reusable (R) | Yes: saved instructions in your assistant, where your account supports it |
| Connected (C) | Only for research connectors, when available |
| Scheduled (S) | Not part of this workflow |
| Custom (X) | Not part of this workflow |
Platforms
It is designed for Perplexity for the research, then Claude for the comparison, then ChatGPT or Claude for the memo. The handoff between assistants is manual.
Designed for is not the same as verified on: no workflow has yet passed the test checklist on any platform. Features depend on your account, plan, region and administrator settings.
Patterns it uses
Each pattern is a reusable rule from the catalogue. Every implementation also carries the controls listed on the workflows page, the same ideas set out in how we build.
- P05 Research, Compare, Decide: sources are found and read, compared and turned into a memo, and a person decides; the handoff between assistants is manual.
- P10 Inspect Retrieved Instructions: an instruction found inside a web page or document is treated as content, not as an order.
A fictional example
A practice manager (fictional) has to recommend one of two options to the owners. She writes down the question, the scope and the criteria, finds sources in Perplexity, reads the originals, and gives them to Claude to compare. The memo separates evidence, assumptions, conflicting claims and unknowns, cites the originals and says what would change the recommendation. The owners make the decision.
Where to go next
Research-to-Decision Office
The Research-to-Decision Office turns a question into an evidence ledger and a one-page decision memo that separates evidence, assumptions and unknowns.
Read more →Vendor Intake and AI Diligence
Vendor Intake and AI Diligence runs the AVRF questionnaire on each new AI tool and drafts a gate decision, with OSFI Guideline E-23 in view for FRFIs.
Read more →Board and Committee Pack Preparation
Board and Committee Pack Preparation collects approved inputs, checks them against the agenda and drafts a cover summary with a source map for every paper.
Read more →W16 Invoice Reminders
W16 Invoice Reminders drafts reminders for confirmed unpaid invoices only, with exact amounts. Paid or disputed items are flagged; nothing is sent without you.
Read more →Independent professionals
Independent professionals: free iSystematic workflow cards for a business profile, decision memos and invoice reminders, run in your own AI assistant account.
Read more →Law firms
Law firms: iSystematic sets up AI workflows for intake summaries and client onboarding. Legal advice, conflict checks and every citation stay with lawyers.
Read more →Set it up in your own accounts
We configure W20 with you in your own Perplexity, Claude and ChatGPT accounts, in a 60 to 90 minute call.