Recipe 4 of 5 · report production
Most of the time spent on a disposal report isn't analysis — it's reading, summarising, and formatting documents that already contain the information. Here's how to hand that part to AI without losing control of accuracy.
Asset reports sit at the intersection of two expensive activities: assembling information from multiple sources, and translating it into a format decision-makers can act on quickly. A standard disposal report might draw on an appraisal, a technical inspection, a legal title summary, cadastral data, market comparables, and a broker's pricing view. AI doesn't replace the analyst's judgment — it removes the reading and formatting work, so the analyst's time goes to the part that actually requires their expertise.
Before involving any AI tool, fix the structure of your output. A standard disposal report might have: Executive Summary, Property Description & Physical Condition, Legal & Cadastral Status, Valuation & Market Positioning, Recommended Disposal Strategy, Outstanding Actions. Once this template is stable, the AI's job is to populate it.
Collect the source documents for the asset into your AI project folder. The quality of the AI output is directly proportional to the quality and completeness of the inputs.
Work section by section, not in a single prompt.
The instruction to flag missing data is important. An AI that fills gaps silently produces a report that looks complete but contains assumptions. The [MISSING] flag makes the review task explicit.
Before assembly, review every flag in the drafted sections. For each gap, either supply the missing information from source documents or note it explicitly as unavailable. This ensures the final report is accurate and complete, with no silent assumptions carried through.
Once sections are reviewed, prompt the AI tool to insert them into the template and produce a full draft for your review.
Yes, section by section, provided you fix the report structure first and instruct the model to flag missing data explicitly rather than fill gaps silently. Working in one large prompt across all sections at once produces noticeably weaker results than a section-by-section approach.
Constrain it to "using only the attached document," require third-person factual language, and require an explicit [MISSING: description] flag for anything not present in the source — then review every flag before the report goes out. This makes omissions visible instead of silently smoothed over.
Because each person's individual prompt — and how rigorously they review [MISSING] flags — differs. Without a shared, enforced template and policy layer, consistency depends entirely on each analyst's discipline.