Methodology
The readiness gate is built from two components. An architecture that institutions configure — templates, criteria, injected context, and calibration — and an AI assistance layer that runs it, grading every answer and guiding proposers before a note ever reaches a human reviewer.
Two components, one readiness gate
The application separates how an evaluation is built from how it is run. One side is configured by evaluators; the other is experienced by proposers — and the two are deliberately decoupled.
The architecture
How an evaluation is built. Evaluators author templates and criteria as data, add steering guidance, and calibrate against a trusted baseline — composable pieces that adapt per country without code changes.
The AI assistance layer
How that evaluation runs. Proposers complete the form and every answer is graded per field against its criteria in context — returning a verdict, reasoning, and an iteration loop before human review.
Catching gaps before a reviewer has to
A large share of proposals arrive incomplete or weakly evidenced. That triggers repeated cycles of comments and resubmissions before a note is even in a state to be graded — slow, inconsistent, and spending skilled reviewer time on basic completeness and formatting checks.
Proposers get immediate, structured feedback and strengthen the note first time around. Submissions reach evaluators more complete and better evidenced, so reviewers can apply judgement and sign-off instead of chasing the basics.
The architecture
The configuration layer. How evaluators design what gets evaluated — and how those pieces compose into a reusable readiness system.
What an evaluation is assembled from
Each block is authored by evaluators and stored independently. Together they define how a submission is graded — before a single proposer touches the form.
Templates
The structured form, authored as data — sections and fields in YAML, compiled to a validated template ahead of time. Each template has its own id, so a country variant is a new template, not a code change.
Supporting criteria
A separate, swappable layer that defines the standard each field is graded against. Criteria are decoupled from the template, so they can be tuned per institution and country without touching the engine.
Context injection
Template-level guidance injected into the prompt. It steers how feedback is written — tone, emphasis, what to foreground — without ever overriding the rubric or the verdict.
Calibration
Known submissions with expected verdicts run against a trusted baseline, confirming the model still lands where a human reviewer would. Any change to criteria or guidance is a calibration event — measured, not assumed.
Composability
Templates, criteria, and guidance are modular and copyable. A calibrated template becomes a starter kit — so lessons learned on one port forward to the next country or sector without rebuilding the evaluation from scratch.
What a Concept Note template covers
A single, structured template covering the breadth of global PIM practice. It is deliberately opinionated and generalised to start, then fine-tuned for country-specific contexts as the tool rolls out.
Strategic rationale
Why the investment is needed and how it fits wider priorities.
Investment identification
What the project is — its objectives, scope, and boundaries.
Description & impact
What will be delivered and the outcomes it is expected to produce.
Costs & financing
Component costs, financing, and operating / revenue scenarios.
Management & risks
Delivery arrangements and a risk register with mitigating measures.
Documentation & land status
Permits, approvals, and land-status checklists.
Inputs range from short narratives to cost worksheets, with-and-without-project scenario comparisons, and compliance checklists — and every field is graded through the same engine.
The AI assistance layer
The runtime layer. How a configured evaluation runs against a live submission — grading each answer and guiding the proposer toward a stronger note.
From draft to human review
Every Concept Note moves through the same path. The first five steps are where the assistance layer does its work; the last is where people take over.
- 01
Draft
The proposer completes the structured Concept Note, section by section. Answers autosave as they go, so a draft survives a reload.
- 02
Map to criteria
Each field is linked to the assessment criteria and guidance that define a complete, high-quality answer for that specific question.
- 03
Evaluate
Evaluation runs per field, against its criteria and the surrounding context of the note. Results arrive progressively as each field is graded.
- 04
Verdict & reasoning
Each answer resolves to a Pass / Not Clear / Fail verdict with plain-language reasoning and concrete, actionable suggestions for improvement.
- 05
Strengthen
The proposer revises weak answers and re-evaluates. Editing an answer marks its earlier feedback as stale, and re-running refreshes the verdict.
- 06
Human review
Once strengthened, the note reaches evaluators at a higher baseline quality — so their time goes to judgement and sign-off, not basic screening.
What every answer is checked for
A field can be strong on one of these and fall short on another. Guidelines are configurable, so institutions can tune what each one demands.
Completeness
Are all the elements the question asks for actually present and evidenced — or are there gaps a reviewer would have to chase?
Quality
Is the answer specific, clear, and well-evidenced to the expected standard, rather than vague or asserted without support?
Conformity
Does it satisfy the template and policy requirements for this section, and stay consistent with figures and statements made elsewhere in the note?
What each field verdict means
Evaluation is per field, not per document — so issues are easy to locate and fix. Every verdict carries reasoning and suggestions, so it's always clear why, and what to do next.
The answer meets its assessment criteria.
All the elements the question asks for are present, clearly stated, and consistent with the rest of the note. Nothing more is needed before human review.
Partially meets the criteria — needs strengthening.
Some required elements are missing, ambiguous, or weakly evidenced. The feedback names exactly what to add or clarify so the answer can be strengthened.
Does not yet meet the criteria as written.
Key requirements are absent or contradict other parts of the note. The answer should be reworked, using the reasoning and suggestions provided, before it progresses.
What the whole tool is built on
Human-in-the-loop
The module accelerates and standardises early screening — it never replaces the reviewer. Evaluators keep control of the grading criteria and the final judgement.
Criteria-driven & configurable
Every answer is assessed against explicit quality, completeness, and conformity guidelines — not opinion. Those guidelines are a configurable layer, so the gate adapts to each institution and country context.
Consistent & transparent
The same yardstick is applied to every submission, and each verdict comes with its reasoning. Feedback is explainable and unbiased rather than a black-box score.