Skip to content

Prompt chains

6 chains · 23 prompts

AI prompts for the work around the RFP

An RFP is mostly judgement, and judgement is what an AI assistant is worst at. What it is genuinely good at is structure: turning notes into scope, claims into evidence, and a rubric into a defensible score. These chains do that, in sequence, with the thinking left to you.

Stage6 chains
Scope4 prompts

Stakeholder notes to a defensible scope

Turns messy interview notes into three artefacts a procurement team can actually use: a decision list, an in/out scope boundary, and the questions you must resolve before publishing.

When
After two to six stakeholder interviews, before anyone writes a requirements section.
Output
A one-page scope boundary plus a list of unresolved decisions with owners.
The chain
  1. 01Extract decisions, not wishes

    You are a programme manager preparing a Drupal procurement. Below are my raw notes from stakeholder interviews for {ORGANISATION}.
    
    Extract them into three lists: (1) decisions that have actually been made, with the person who made them; (2) stated requirements that are still preferences; (3) open questions that block scoping.
    
    Do not add anything that is not in the notes. Where the notes are ambiguous, quote the sentence and mark it ambiguous.
    
    NOTES:
    {PASTE_NOTES}
  2. 02Draw the scope boundary

    Using the decisions and requirements above, write a scope boundary for a Drupal request for proposals.
    
    Produce two lists: IN SCOPE as 6-10 work packages, each with a deliverable and an owner on the buyer side; OUT OF SCOPE as 5-8 items a vendor might otherwise assume, each with a one-line reason.
    
    Keep every item priceable by a vendor who has never met us. Flag anything that depends on an unresolved decision with [BLOCKED: decision].
  3. 03Predict the three failure modes

    Given this scope, identify the three most likely ways this project fails: the specific decisions that get deferred, the dependencies that slip, and the assumptions that turn into change requests.
    
    For each failure mode give: the warning signal we would see first, the clause in the RFP that would reduce the risk, and the question to ask vendors that exposes whether they have handled it before.
    
    Be concrete about Drupal: content migration volume, configuration management, editorial workflow change, integration ownership, and hosting ownership are common causes.
  4. 04Build the pre-RFP question list

    Write the list of questions we must answer internally before publishing this RFP, sorted by who can answer them (executive, finance, IT/security, content team, legal).
    
    For each question: why the RFP is weaker without the answer, a recommended default if we cannot get one, and the deadline relative to an RFP publish date of {DATE}.
    
    Keep it under 15 questions. Ruthlessly merge anything that would be answered by the same meeting.
Draft4 prompts

Ambitions into testable requirements

Converts goal statements into numbered clauses that an evaluator can score and a vendor cannot answer with 'yes, we do that'.

When
When you have objectives but the requirements section still reads like a brochure.
Output
Numbered requirements with acceptance criteria and a verification method for each.
The chain
  1. 01Rewrite goals as requirements

    Here are our project goals for {ORGANISATION}:
    {PASTE_GOALS}
    
    Rewrite each as 3-5 numbered requirements of the form "R{n}.{m} The platform must ..." such that each requirement could be verified from a written proposal or a demonstration.
    
    Rules: no adjectives that cannot be tested, no vendor names, one obligation per requirement, and every requirement must state the standard it is measured against (a number, a standard, or a named method).
  2. 02Pressure-test verifiability

    Review the requirements below as a hostile evaluator. For each one, state: how a vendor could claim compliance without actually complying, what evidence would prove compliance, and whether the requirement is testable as written.
    
    Then rewrite the five weakest requirements so that the evidence standard is explicit.
    
    REQUIREMENTS:
    {PASTE_REQUIREMENTS}
  3. 03Add acceptance criteria

    For each requirement below, add an acceptance criterion in the form "Accepted when ..." naming the artefact or test that proves it (a report, a demonstration step, a document, a measurement).
    
    Where an acceptance criterion depends on something the buyer must supply, say which.
    
    REQUIREMENTS:
    {PASTE_REQUIREMENTS}
  4. 04Find the missing sections

    Here is the table of contents of our Drupal RFP:
    {PASTE_TOC}
    
    Compare it against what a vendor needs in order to produce a fixed price and a credible plan. List the missing sections in priority order, say what each missing section prevents the vendor from pricing, and draft the first two clauses of each missing section.
    
    Assume the vendor has never worked with us and cannot ask questions until {QUESTION_DEADLINE}.
Specify4 prompts

A hosting annex that survives contract

Drafts the infrastructure section so that ownership, recovery and exit are testable, without writing a specification only one supplier could meet.

When
Before the hosting annex goes to legal, and again when you compare hosting answers.
Output
A grouped annex of clauses with an evidence request attached to each.
The chain
  1. 01Draft clauses from constraints

    We are procuring Drupal hosting and support. Our constraints are:
    - Hosting model: {OWN_CLOUD | VENDOR_MANAGED | ON_PREM | UNDECIDED}
    - Compliance obligations: {LIST}
    - Traffic profile: {PEAK_REQUESTS_AND_CONCURRENCY}
    - Content volume: {NODES_MEDIA_LANGUAGES}
    - Recovery expectations: {RPO_AND_RTO}
    
    Write hosting requirements grouped under: ownership, environments and deployment, performance and caching, backup and recovery, security, observability and support, exit and portability.
    
    Format each as "R7.{n} The supplier must ...". Every clause must be verifiable from a document or a demonstration. Do not name any vendor or product suite.
  2. 02Attach the evidence request

    For each hosting clause below, add a second sentence beginning "Evidence required:" naming the specific document, report, screenshot or demonstration that would prove compliance.
    
    Reject vague evidence: "confirmation from the supplier" is not evidence. Prefer artefacts that already exist for other clients.
    
    CLAUSES:
    {PASTE_CLAUSES}
  3. 03Check for lock-in

    Read these hosting clauses as a supplier whose business model depends on keeping the customer on its own platform. Identify every clause that threatens that model, and every gap where a supplier could satisfy the letter of the clause while keeping the customer locked in.
    
    For each gap, propose tighter wording. Pay particular attention to data export formats, credential ownership, infrastructure-as-code handover, and professional services that are only available from the incumbent.
    
    CLAUSES:
    {PASTE_CLAUSES}
  4. 04Prepare the comparison grid

    Build a comparison grid for evaluating hosting responses. Rows: our hosting clauses. Columns: {VENDOR_A}, {VENDOR_B}, {VENDOR_C}.
    
    For each cell, define what "complies", "partial" and "does not comply" look like in terms of evidence, so three evaluators score consistently.
    
    Add a final section listing the five questions that most reliably separate a genuinely portable offering from a rebranded managed platform.
Evaluate4 prompts

A scoring model that survives appeal

Builds weights, score definitions and gates so the decision is arithmetic rather than opinion.

When
Before publication, so the criteria in the RFP match the scorecard you will actually use.
Output
A weighted rubric, a 0-5 definition set, and pass/fail gates.
The chain
  1. 01Weight the criteria to our priorities

    We are buying Drupal development and hosting. Our priorities, in order, are: {LIST_PRIORITIES}.
    
    Propose 7-9 evaluation criteria with percentage weights summing to 100. For each: what it measures, what evidence in a proposal earns a high score, and what the risk is of weighting it too heavily.
    
    Constraints: price must be between 15 and 30 per cent. Accessibility must appear. Hosting must appear separately from the build. Explain any weight above 20 per cent.
  2. 02Define every point on the scale

    For each criterion below, write a scoring definition for 5, 4, 3, 2, 1 and 0.
    
    Each definition must describe observable evidence, not sentiment. A 5 must require something a reader can point at in the proposal. A 0 must describe a disqualifying absence.
    
    CRITERIA:
    {PASTE_CRITERIA_WITH_WEIGHTS}
  3. 03Write the gates

    Write the pass/fail gates for this procurement: the conditions under which a proposal is not scored at all, regardless of quality.
    
    Cover legal standing, insurance, named personnel, comparable delivery, security assurance, and ability to meet the required hosting model. For each gate: the exact evidence required, who checks it, and what happens if evidence arrives late.
    
    Keep gates to things that are genuinely non-negotiable; anything arguable belongs in the weighted score.
  4. 04Simulate the decision

    Using the rubric and weights below, simulate three plausible outcomes: a strong-but-expensive proposal, a cheap-but-thin proposal, and a balanced proposal.
    
    For each, show the arithmetic, the resulting rank, and the sentence an unsuccessful bidder would use in an appeal. Then identify which weight, if changed by five points, would most change the outcome, and whether that change is defensible.
    
    RUBRIC:
    {PASTE_RUBRIC}
Evaluate3 prompts

Proposal review with citations

Extracts claims and evidence from a vendor proposal, scores it against your rubric, and separates assertion from proof.

When
When proposals land, before the evaluation panel meets.
Output
A claim-by-claim evidence table and a defensible draft score.
The chain
  1. 01Extract claims and evidence

    Below is a vendor proposal for a Drupal project. Build a table of every substantive claim, with columns: claim, section reference, type (fact / assertion / commitment), and what evidence is offered (document, reference, metric, none).
    
    Then list the ten claims with the highest impact on delivery risk that have no evidence behind them.
    
    PROPOSAL:
    {PASTE_PROPOSAL_TEXT}
  2. 02Score against the rubric, with citations

    Score this proposal against the rubric below. For every criterion: give a score, quote the passage that justifies it (or state that no passage exists), and note the single question that would most reduce your uncertainty.
    
    Do not reward length or design. If the proposal does not address a criterion, score accordingly and say so explicitly.
    
    RUBRIC:
    {PASTE_RUBRIC}
    
    PROPOSAL:
    {PASTE_PROPOSAL_TEXT}
  3. 03Compare the finalists

    Here are my scored evaluations of {N} finalist proposals against the same rubric:
    {PASTE_SCORED_SUMMARIES}
    
    Produce a comparison that answers: where do they genuinely differ, where are they saying the same thing in different words, and which differences actually matter for delivery risk over three years?
    
    End with the three questions that would most change the ranking if answered badly.
Decide4 prompts

Clarification, references and best-and-final

Turns the gaps in a proposal into precise questions, a reference-call script, and a negotiation position.

When
After shortlisting, before award.
Output
A question set, a reference-call script, and the asks for a best-and-final round.
The chain
  1. 01Write the clarification questions

    Based on the gaps and unsupported claims below, write clarification questions for {VENDOR}.
    
    Rules: one question per issue, no compound questions, each answerable in under 200 words, and each phrased so that a vague answer is visibly vague. Group them by topic and mark which are genuinely disqualifying if unanswered.
    
    GAPS:
    {PASTE_GAPS}
  2. 02Design the reference call

    Write a 20-minute reference-call script for a client who bought a comparable Drupal project from this vendor.
    
    Ask questions that produce specific, checkable answers rather than testimonials: what was the original estimate versus final cost, what changed, how the vendor handled a missed deadline, who did the actual work versus who sold it, how the handover went, and what they would write differently in the contract.
    
    Include the three answers that should worry us most, and what to do if we hear them.
  3. 03Prepare the best-and-final asks

    We intend to award to {VENDOR} subject to a best-and-final round. Our leverage is {LEVERAGE}. Our concerns are {CONCERNS}.
    
    Draft the asks for this round, each tied to a specific concern: contract terms, evidence, staffing commitments, phasing, payment milestones, or price structure.
    
    For each ask, state what we will concede in exchange and what we will walk away over. Do not invent concessions we have not offered.
  4. 04Write the unsuccessful-bidder note

    Draft the note to unsuccessful bidders for this procurement.
    
    It must: thank them specifically, state the score against the published criteria without disclosing other bids, name the one area where their proposal was weakest, and keep the door open for future work.
    
    It must not: imply the decision was purely price, contradict the published criteria, or invite an appeal through ambiguity. Keep it under 150 words.

Rule one

Never let a model invent a fact about your organization. Ask it to mark unknowns as placeholders, then fill them yourself.

Rule two

Redact before you paste. Commercial terms, client names and personal data do not belong in a public model.

Rule three

Keep the output, not the conversation. Save the artefacts: scope boundary, clause list, score rationale.