Skip to content
PlotUProperty intelligence · AI · evidence

Site research · 8 min read

One brief, every site: replacing ad-hoc research with a repeatable triage workflow

A small developer was researching every site differently, in whoever's head was free that week. The fix was a format, not a platform.

Published

About this case study

This is an anonymised composite drawn from recurring engagements. No client is identified, no figures are invented, and no outcome is guaranteed. The workflow is the substance.

The situation

A developer running a small pipeline of prospective sites had no consistent way to compare them. Research happened in bursts: a planning portal session here, a spreadsheet there, a phone call to an agent whose notes lived in an email thread. Two sites could not be compared on the same terms because they had never been researched on the same terms.

The details that follow are a composite of recurring engagements of this kind. No client is identified, and nothing here is a claim about a specific company's results.

How the workflow runs

  1. Define what 'good' looks like before any tooling

    The first deliverable was a one-page brief format: site particulars, planning history chronology, constraints and designations, access and title flags, and the two assumptions the appraisal rests on. Every subsequent automation was built to produce that shape and nothing else.

  2. Gather the source set deliberately

    For each site, the same source set is assembled: planning history from the local authority portal, the relevant local plan policies, constraint mapping, and any appeal decisions in the immediate area. What is missing is noted explicitly rather than discovered later.

  3. Let AI do the reading, with citations

    The model reads the assembled documents and drafts the brief — a chronology of applications, the policies each officer report relied on, and a constraint list. Every policy statement carries a pointer to the source paragraph. Anything it cannot source is marked for a human to check, not presented as a finding.

  4. Human review against the checklist

    The developer reviews the draft against the fifteen triage questions, verifies every policy claim against its cited paragraph, and makes the go/no-go call. That judgement is never delegated.

  5. Record the decision, not just the outcome

    Each site file closes with the brief, its sources, the date, and the decision with its reason. Parked sites become re-openable instead of forgotten.

What changed

The qualitative shift was from research as an event to research as a system. Sites stopped being evaluated on whoever's memory was available, and the pipeline became something that could actually be ranked.

  • Every site now produces the same shaped document, so comparisons are honest
  • Policy claims travel with their sources, so nothing unsourced reaches an appraisal
  • The go/no-go decision is written down with its reasoning
  • A parked site can be revisited months later without re-reading everything

Where the professional line sits

  • The brief is research, not planning advice — acceptability judgements still go to a planning consultant, now better briefed
  • Every policy statement must cite a source paragraph before it is relied on
  • The go/no-go decision is always made and recorded by the developer

FAQ

Common questions

Is this a real client case study?
It is a composite of recurring engagements, anonymised so no client can be identified. We do not publish client names, logos or fabricated metrics — the workflow described is the real substance.
How long does the triage take per site once the workflow exists?
The reading and drafting compresses from days of fragmented effort to a single focused session. The review and decision still take the time they deserve — that part is not meant to be fast.
Do I need special software to run this?
No. The workflow is deliberately tool-agnostic: a defined brief format, a source set, a capable language model, and a review checklist. Our free 30-minute AI Site Triage is the manual starting version.

Keep reading

More case studies

Discuss your workflowAll case studies