Ahmadreza Samadi / HomeAn interactive tool by Ahmadreza Samadi
Before you
build the AI.
Make one decision worth helping with.
Work backward from a real action to the evidence it needs. Leave with a focused brief and a first experiment you can run before committing to a full product.
Your decision brief
01—03A smaller promise.
A clearer test.
Illustrative example loaded. Change any detail, then rebuild your brief.
01 / The decision
A property researcher Which properties deserve a closer investigation before an investor commits more time? How ready is that evidence? Some is available; gaps remain If the answer is wrong… It creates meaningful rework or cost Who reviews before action? A person reviews every result
The next action
Shortlist candidates and request missing information or a manual inspection.
02 / The evidence
Dated property imagery, relevant property attributes and source provenance. Record what the imagery cannot establish.
Resolve first
Make the gaps visible. Identify missing sources and define when the tool must ask for information or hand the decision back to a person. Keep the reviewer in the test. Record corrections and the time needed to review each result, alongside the model output.
03 / The smallest useful experiment
- Choose a small, explicitly scoped set of cases for this decision. Include typical cases, missing evidence and plausible failures. Do not treat a small pilot as proof of broad reliability.
- Establish the baseline: A researcher checks each listing and builds a shortlist manually. Record time, outcomes and correction effort.
- Prepare evidence-backed suggestions for the same cases, manually or with a simple prototype. Keep the reviewer in the test. Record corrections and the time needed to review each result, alongside the model output.
- Compare against your criterion: Reduce research time while retaining useful candidates, with missing evidence visible and review effort included. Set measurable pass and stop conditions before testing; include delivery and review costs.
Proceed only when
The evidence supports the intended action, failures have a clear fallback, and the combined workflow improves on the baseline. If it does not, narrow the decision or revisit the inputs.
A structured thinking aid, not an AI assessment or a readiness score. Guidance follows your selections; it does not validate the evidence.
Keep a working copy
A one-page experiment card.
Prefer a fixed worksheet? Download the blank English PDF with 10 editable fields for your decision, evidence, baseline, acceptance criterion, review rules, test and observed result. It complements the full interactive brief above.
Download the fillable experiment card (PDF, English) →
See a completed fictional planning example (PDF, English) and reading notes (plain text, English). The proposed pilot has not run; no results or savings are claimed. Start real work from the blank card.
A thinking aid by Ahmadreza Samadi, not a validated assessment. Complete the form in a PDF reader that supports editable fields and save a local copy; the download is not connected to this page's entries.
The thinking behind the tool
The output is only part of the product.
For property workflows, read where AI should stop in real estate.
The supporting evidence, the next action and the cost of review determine whether an answer is useful. This worksheet puts those choices on one page.
For the next step, record costs and acceptance decisions with the AI workflow cost ledger or experiment workbook (Excel, English). Three worked examples (ZIP, English) show retries, no accepted results and explicitly known zero costs. The included teaching data are fictional. You supply the acceptance rule; these resources do not verify answer quality.
Read “Start with the decision, then build the AI” →