Skip to content

Resources · Templates

AI project templates and checklists

Six working documents for the parts of an AI project that usually go unwritten: choosing the first use case, preparing a chatbot’s content, running an automation, questioning vendors, scripting a voice agent and recording data risks. Plain Markdown and CSV, free to download with no sign-up, and generic: none of them is a client document.

Book a Strategy CallBrowse all resources

Project templates

Templates you can download

Each preview is read from the real file when this page is built, so what you see is what you download.
  • ai-use-case-scoring-sheet.csvSample data

    AI use-case scoring sheet

    Template v1 ·

    Columns A, B, L, M, P, Q, R, T, U of 23, spreadsheet rows 6, 7, 8
    IDUse case (one sentence in the words of the people who do the work)Data classWho sees a mistake firstValue 0-3Feasibility 0-3Risk 0-3 (3 = safest)Priority (P+Q+R)Decision
    EX-1EXAMPLE (sample data) - delete this row. Draft replies to order-status emails for an agent to check and sendPersonalA customer, after a person checks2327Build first
    EX-2EXAMPLE (sample data) - delete this row. Approve or reject refund requests automaticallyPersonalMoney, health, legal outcomes or a regulator2103Not now
    Preview of the register in ai-use-case-scoring-sheet.csv: 9 of its 23 columns. The example rows are Sample data and say so in the row.
    CSV4 KBv1 ·

    AI use-case scoring sheet

    A register for every AI idea your teams suggest, one row each, with the facts that decide it (volume, time per item, systems touched, who sees a mistake first) recorded before anyone scores value, feasibility and risk.

    What the file contains (5)
    1. How to use it, and the two rules above the total
    2. The register: 23 columns, 10 rows to fill in and 2 example rows
    3. Scoring guide
    4. Disqualifiers (any one means Not now)
    5. Allowed values
    Download AI use-case scoring sheet,CSV · 4 KBHow to use it: AI use-case scoring sheet
  • chatbot-knowledge-base-readiness-checklist.mdMarkdown

    Chatbot knowledge-base readiness checklist

    Template v1 ·

    1. Scope: what the assistant may and may not answer

    • We have listed the questions customers ask most, taken from real tickets, chats, emails or call notes, not from memory.
    • Each of those questions has one approved answer somewhere in our content.
    • We have written a never-answer list: topics the assistant must hand to a person (for example complaints, refunds above a limit, account security, legal or medical questions).
    • We know who the assistant serves (prospects, customers, staff or a mix) and whether they should see different content.
    Preview of section 1 of 8 in chatbot-knowledge-base-readiness-checklist.md. The download is the whole file.
    Markdown4 KBv1 ·

    Chatbot knowledge-base readiness checklist

    Eight sections that decide whether an assistant can answer from your content: scope and a never-answer list, a source inventory, content quality, access, freshness, starter test questions, hand-off and sign-off.

    What the file contains (8)
    1. 1. Scope: what the assistant may and may not answer
    2. 2. Source inventory
    3. 3. Content quality
    4. 4. Access and sensitive content
    5. 5. Freshness
    6. 6. Test questions (the start of your evaluation set)
    7. 7. Hand-off to a person
    8. 8. Sign-off
    Download Chatbot knowledge-base readiness checklist,Markdown · 4 KBHow to use it: Chatbot knowledge-base readiness checklist
  • workflow-automation-runbook.mdMarkdown

    Workflow automation run-book

    Template v1 ·

    7. Replaying a failed run

    1. Find the failed run and read the error.
    2. Fix the cause (bad data, an expired credential, an outside system that was down) before replaying.
    3. Check which steps already completed for this record (see "Safe to repeat?").
    4. Replay with the same idempotency key, so a step that already succeeded is not repeated.
    Preview of section 7 of 12 in workflow-automation-runbook.md. The download is the whole file.
    Markdown4 KBv1 ·

    Workflow automation run-book

    One per workflow: owners, the trigger and its duplicate key, credentials recorded by location and never by value, each step with its retry and failure behaviour, replay steps, a manual fallback, limits and a monthly check.

    What the file contains (12)
    1. 1. Identity
    2. 2. Trigger
    3. 3. Systems and credentials
    4. 4. Steps
    5. 5. AI steps (skip if there are none)
    6. 6. When it fails
    7. 7. Replaying a failed run
    8. 8. Manual fallback
    9. 9. Switching it off
    10. 10. Known limits
    11. 11. Monthly health check
    12. 12. Change and incident log
    Download Workflow automation run-book,Markdown · 4 KBHow to use it: Workflow automation run-book
  • ai-vendor-rfp-questions.mdMarkdown

    AI vendor and agency RFP questions

    Template v1 ·

    A. Understanding of the problem

    1. Describe our problem back to us in your own words, including what you would leave out of scope.A complete answer restates the task, the users and the measure of success, and names at least one thing it would not do.
    2. What do you need from us before you can give a fixed price?A complete answer lists specific inputs (sample documents, system access, volumes), not only "a discovery phase".
    Preview of section 1 of 10 in ai-vendor-rfp-questions.md. The download is the whole file.
    Markdown5 KBv1 ·

    AI vendor and agency RFP questions

    Twenty-nine questions in nine groups, from data handling and evaluation to ownership and exit, with notes on what a complete answer contains, a scoring scale and red flags.

    What the file contains (10)
    1. A. Understanding of the problem
    2. B. Approach and architecture
    3. C. Data handling
    4. D. Quality and evaluation
    5. E. Safety and misuse
    6. F. Security and access
    7. G. Ownership and exit
    8. H. Team and continuity
    9. I. Price and change
    10. Red flags in answers
    Download AI vendor and agency RFP questions,Markdown · 5 KBHow to use it: AI vendor and agency RFP questions
  • voice-agent-call-script-and-escalation-map.mdMarkdown

    Voice agent call script and escalation map (inbound calls)

    Template v1 ·

    5. Escalation map

    Table preview; the file has 2 more columns and 5 more rows
    TriggerHow it is detectedActionGoes to
    Caller asks for a personPhrase or keypressWarm transfer[Front desk]
    Emergency or safety riskWords you list: [____]Read the emergency instruction you approved, then transfer[____]
    Complaint or legal threatIntentTransfer[Manager on duty]
    Preview of section 5 of 8 in voice-agent-call-script-and-escalation-map.md. The download is the whole file.
    Markdown5 KBv1 ·

    Voice agent call script and escalation map

    An inbound call script with the disclosure line, intents and read-backs, what the agent does on silence or confusion, an escalation map of triggers, destinations and after-hours behaviour, and the things it must never do.

    What the file contains (8)
    1. 1. Line details
    2. 2. Opening
    3. 3. What the agent handles
    4. 4. When the agent does not understand
    5. 5. Escalation map
    6. 6. The agent must never
    7. 7. Closing and call summary
    8. 8. Test calls before launch
    Download Voice agent call script and escalation map,Markdown · 5 KBHow to use it: Voice agent call script and escalation map
  • ai-project-risk-and-data-handling-checklist.mdMarkdown

    AI project risk and data-handling checklist

    Template v1 ·

    6. Risks specific to AI systems

    Based on the OWASP Top 10 for LLM Applications 2025 (https://genai.owasp.org/llm-top-10/, checked 2026-10-18). Fill in a control and an owner for every row that applies.

    Table preview; the file has 3 more rows
    RiskAsk yourselfYour controlOwner
    Prompt injection (LLM01)Could text the system reads, such as a web page, an email or a document, contain instructions it might follow?blankblank
    Sensitive information disclosure (LLM02)Could an answer reveal data this user should not see?blankblank
    Improper output handling (LLM05)Is model output checked before another system uses it?blankblank
    Excessive agency (LLM06)Does the system have more permissions or actions than the task needs?blankblank
    Preview of section 6 of 9 in ai-project-risk-and-data-handling-checklist.md. The download is the whole file.
    Markdown5 KBv1 ·

    AI project risk and data-handling checklist

    A data inventory, the rules that apply, provider terms with the date checked, access, logs, the AI-specific risks from the OWASP list for LLM applications, people in the loop, a pre-launch gate and a one-page system card.

    What the file contains (9)
    1. 1. Data inventory
    2. 2. Rules that apply
    3. 3. Model and service providers
    4. 4. Access and secrets
    5. 5. Logs and records
    6. 6. Risks specific to AI systems
    7. 7. People in the loop
    8. 8. Before launch
    9. 9. System card (one page, kept with the system)
    Download AI project risk and data-handling checklist,Markdown · 5 KBHow to use it: AI project risk and data-handling checklist

Guidance

How to use each template

When to reach for each one, who should fill it in, and what to do with it once it is filled in.

  1. 01 · CSV

    AI use-case scoring sheet

    Download AI use-case scoring sheet,CSV · 4 KB
    When to use it
    When ideas arrive from several teams, or you have more of them than budget, and need to agree an order before talking to any vendor.
    Who fills it in
    Whoever proposed each idea fills in its facts. A small group that includes the budget holder then scores every row together, in one sitting.
    Afterwards
    Run the top two or three rows through the first-project rubric in the scoping guide, and attach the winning row to your scope brief.

    The method behind it:Scoping your first AI project: readiness, the variables that set price and time, and how to choose

  2. 02 · Markdown

    Chatbot knowledge-base readiness checklist

    Download Chatbot knowledge-base readiness checklist,Markdown · 4 KB
    When to use it
    Before you ask anyone to build a chatbot or help-centre assistant, and once more in the week before launch.
    Who fills it in
    The people who answer customer questions today, sitting with the owner of each content source.
    Afterwards
    Hand the source inventory and test questions to the builder as their starting point. When the first four sections are fully ticked, the content is ready to build on.

    The method behind it:How to build a RAG chatbot, step by step: an assistant that cites its sourcesHow to evaluate an AI system: build your first 100-question evaluation set

  3. 03 · Markdown

    Workflow automation run-book

    Download Workflow automation run-book,Markdown · 4 KB
    When to use it
    Start it while the workflow is being designed, finish it before go-live, and change it in the same commit or edit that changes the workflow.
    Who fills it in
    The builder writes it. The business owner checks the failure and fallback sections, because those decide what happens to the work when something breaks.
    Afterwards
    Store it beside the workflow definition, and let the monthly health check at the end tell you when the two have drifted apart.

    The method behind it:Reliable workflow automation: retries, idempotency and dead lettersEvent-driven automation: a reference architecture

  4. 04 · Markdown

    AI vendor and agency RFP questions

    Download AI vendor and agency RFP questions,Markdown · 5 KB
    When to use it
    When you are about to ask vendors to bid on an AI build, or need to compare a shortlist on the same terms. We expect to be asked them too.
    Who fills it in
    Procurement or the project lead assembles the request. Your technical and data owners score the answers, each on the questions they can judge.
    Afterwards
    Ask for the evidence behind every top score before you shortlist. If you would like a proposal from us as well, send the request through our RFP page.

    The method behind it:Send us your RFP or security questionnaireAI agency, in-house team or freelancer: who should build your AI system?

  5. 05 · Markdown

    Voice agent call script and escalation map

    Download Voice agent call script and escalation map,Markdown · 5 KB
    When to use it
    Before a voice agent is configured, so the build follows your rules instead of inventing them, and again before every change to the script.
    Who fills it in
    The people who answer your phones today, whoever takes transferred calls, and the person who approves customer-facing wording.
    Afterwards
    Run the test calls in the last section with colleagues who did not write the script, then keep the file as the approved script of record.

    The method behind it:AI voice agent buyer’s guide: what a minute costs, where the latency goes, what consent you needInbound voice agent: a reference architecture

  6. 06 · Markdown

    AI project risk and data-handling checklist

    Download AI project risk and data-handling checklist,Markdown · 5 KB
    When to use it
    Before any of your data leaves your systems for a build or a pilot, and again before launch.
    Who fills it in
    The project owner together with your privacy or legal advisor. The builder supplies the facts about providers, regions and architecture.
    Afterwards
    Open items become build tasks or risks accepted in writing. The system card stays with the system and is updated whenever it changes.

    The method behind it:Designing AI agents with guardrails: patterns, permissions and when not to use oneSecurity practices for the AI systems we build

Method

Five habits that make templates work

These are working documents, meant to be filled in and argued over rather than filed.

  • Fill them in with the people who do the work today. Managers know the goal; the people doing the work know the exceptions.
  • Leave a blank rather than guess. A blank is the next question to answer, and it is cheaper to answer before a build than during one.
  • Keep one copy per project next to its other documents, and date every revision.
  • Share them with whoever builds the system, whether that is your own team, another vendor or us. None of them assumes a particular builder.
  • Treat nothing here as legal advice. The risk checklist and the call script mark the places where your own advisor has to confirm the rules.

The checklists end in a sign-off, a log or a set of test calls on purpose. A template nobody revisits goes out of date as quickly as the system it describes.

Outside references in the templates

  1. OWASP Top 10 for LLM Applications 2025 (opens in a new tab) · OWASP Gen AI Security Project · accessed
  2. AI RMF Core: Govern, Map, Measure, Manage (opens in a new tab) · NIST Trustworthy and Responsible AI Resource Center · accessed

FAQ

Frequently asked questions

Which file formats are the templates in, and what opens them?

Five are Markdown files and one is a CSV spreadsheet. Markdown is plain text with light formatting marks: any text editor opens it, and most documentation tools show it with headings, tables and tick boxes. The CSV opens in Excel, Google Sheets, Numbers or LibreOffice; save it in your spreadsheet’s own format once you start.

Why are the templates not Word or Excel files?

Plain text opens on any device without a licence, carries no macros and shows every change line by line in version control, which matters for a run-book or risk register that changes over the life of a system. If your team works in Word or Google Docs, paste the text in and carry on there.

Do we have to sign up or give an email address to download a template?

No. The links are plain files. Our own analytics count how often each file is downloaded, with no names, email addresses or cookies, so we can see which templates are worth improving.

Can we send these templates to other vendors bidding for our project?

Yes. The RFP question list is written for exactly that, and a filled-in knowledge-base checklist or scoring sheet makes any vendor’s estimate more accurate. We expect to answer the same questions ourselves when we bid.

Are any of these templates taken from a client project?

No. They are generic documents written from how we run our own projects. The two example rows in the scoring sheet are marked as Sample data in the row itself; delete them before you start.

Which template should we fill in first?

The scoring sheet, to agree which idea goes first. Before any data leaves your systems for a pilot, complete the risk and data-handling checklist. Then use the template for the kind of system you chose: the knowledge-base checklist for a chatbot, the call script for a voice agent, the run-book for an automation. The RFP questions come in when you ask vendors to bid.

Will you go through a filled-in template with us?

Yes. Bring the scoring sheet or the risk checklist to a strategy call and we will work through it with you. If we then scope a project together, your filled-in documents become part of its brief, so nothing is written twice.

How will we know when a template has been updated?

Each file starts with a version number and a date, and its card on this page shows the same ones. Both change only when the content does. Outside lists the templates refer to, such as the OWASP risks in the data-handling checklist, carry the date we last checked them.

Next step

Have a workflow in mind? Let’s scope it together.

Book a free 30-minute strategy call. Bring one process you want to automate and we will talk through scope, integrations and a sensible first step.

Book a Strategy CallGet an AI Plan

Free 30-minute call · no commitment

Send an RFP or security questionnaire