AI helps project delivery when it drafts the documents you already owe people: a weekly status report from your raw notes, a risk register that is reviewed every week, and a handover checklist built from the project folder. You fix the template and the status rules, the AI fills the gaps, and you sign off every page before it leaves.
This is for the delivery lead in a small firm who runs several projects at once and writes the same three documents over and over. It is not about project-management software, though the templates paste into whatever you use. The pattern holds throughout: raw notes in, structured draft out, human check, human send.
Three documents, three readers
Start by being clear who reads each document and what they do with it, because the template follows the reader, not the other way round.
| Document | Reader | Cadence | What the reader does with it |
|---|---|---|---|
| Status report | Client or firm owner | Weekly, same day | Decides whether to worry, and whether you need anything |
| Risk register | Delivery lead and the owner of each risk | Reviewed weekly, shared monthly | Acts on the next step for each live risk |
| Handover note | The person taking over or running aftercare | Once, at a stage change | Runs the project without calling you |
If the reader would not miss a section, cut it. Status reports in small firms usually fail by being long, not short.
The illustrative project: a shop-front fit-out
For the examples, take a signage company that installs shop-front signs and window graphics. Its delivery lead is running a fit-out for a client with three shops: survey, artwork approval, fabrication, installation over two weekends, and handover to an aftercare person who will deal with faults. Five people touch it. The delivery lead keeps a weekly scratch note in a plain document, which becomes the raw material for everything below. All names, dates and figures in the examples are invented.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowKeep one scratch note through the week
The quality of an AI draft depends on the quality of the notes. The habit that matters is writing a line when something happens, not remembering on Friday. Use five headings and put a line under them as the week goes:
DONE: [what finished, with date]
SLIPPED: [what moved, by how much, why]
DECISIONS: [who decided what, and when]
ISSUES: [open problems, who owns each]
SPEND: [costs committed or changed]A filled-in Thursday for the fit-out: "DONE: artwork approved for shops 1 and 2 (Tue). SLIPPED: shop 3 artwork waiting on client's landlord, 4 days late. DECISIONS: client agreed to move installation of shop 3 to the second weekend (Wed). ISSUES: fabricator quote for the shop 2 projecting sign is $380 above budget; owner: delivery lead. SPEND: $1,150 deposit paid to fabricator."
The weekly status report, drafted in two minutes
The status report needs fixed sections so the reader learns where to look. The set that works for most small projects is: overall status with the reason in one sentence, progress since last week, next week's plan, risks and issues needing attention, decisions or help needed from the reader, and budget and dates against plan. Put the template and the colour rules into the prompt.
Write a weekly status report for the client from my notes. Use exactly these sections: Overall status (one word and one sentence); Done this week; Next week; Risks and issues; What I need from you; Budget and dates.
Status rules: GREEN = no milestone more than 2 days late and spend within budget. AMBER = a milestone 3 to 7 days late or spend up to 10% over. RED = a milestone more than 7 days late, or spend more than 10% over. Apply the rules to the facts; if the facts fit two colours, choose the worse and say why.
Use only facts in my notes. If a date or figure is missing, write [MISSING]. Plain sentences, under 250 words.
Plan: survey 3 Mar; artwork approval 10 Mar; fabrication complete 24 Mar; install weekends of 28 Mar and 4 Apr; handover 11 Apr. Budget $14,600.
Notes: [paste scratch note]Illustrative output:
Overall status: AMBER. Artwork approval for shop 3 is four days late, waiting on the landlord.
Done this week: Artwork approved for shops 1 and 2 on Tuesday. Deposit of $1,150 paid to the fabricator.
Next week: Fabrication starts on shops 1 and 2. Chase landlord for shop 3 artwork.
Risks and issues: Shop 3 approval is the critical path to the second install weekend. Quote for the shop 2 projecting sign is $380 over budget.
What I need from you: Please ask the landlord's agent to confirm artwork by Wednesday.
Budget and dates: Committed spend [MISSING] of $14,600. Handover 11 Apr unchanged.Three checks on that draft. First, the colour: shop 3 is four days late (amber by the rule) and the extra $380 is 2.6% of the budget, so amber is right. Second, the "[MISSING]" is useful: the model did not invent a committed-spend figure, so you add it from your records. Third, the phrase "critical path" may be jargon for your client; swap it for "the date the second weekend depends on". Then send it yourself.
If you want Claude-specific prompts that do the same job, Claude prompts for project management: weekly status and risk log covers them in their own tutorial.
Why the colour rules belong to you
A status colour is a promise. If the AI picks it from a mood in your notes, it will drift towards green whenever your language is cheerful, and a report that is always green stops being read. Writing the rules down once also removes the weekly argument with yourself. When you change a rule, tell the client.
A risk register that people actually update
A risk is something that has not happened yet and could hurt the project. Anything that has already happened is an issue and goes in the status report. A small project needs five to ten live risks, each with an owner and a next action. Use these columns:
| Column | What goes in it |
|---|---|
| ID and date raised | R1, 3 Mar |
| Risk statement | "If [cause], then [event], leading to [effect]" |
| Likelihood (1-3) | 1 unlikely, 2 possible, 3 likely |
| Impact (1-3) | 1 minor, 2 noticeable, 3 serious |
| Score | Likelihood times impact |
| Owner | One named person |
| Next action and date | One concrete step |
| Trigger | The sign it is happening |
The AI's best use is turning scattered concerns into risk statements, because most people write risks as vague worries. Paste a week of your scratch notes and emails (client names removed) and ask:
From these notes, identify up to 6 project risks (things that might happen, not things that have happened). Write each as: "If [cause], then [event], leading to [effect]". Suggest likelihood and impact on a 1 to 3 scale with a one-line reason. Suggest an owner by role and one next action. Do not add risks that are not supported by the notes.Illustrative output (extract): "R2. If the landlord's agent does not confirm shop 3 artwork by Wednesday, then fabrication for shop 3 misses the slot booked with the fabricator, leading to installation moving past the second weekend. Likelihood 2 (already four days late); impact 3 (client's trading date). Score 6. Owner: delivery lead. Next action: ask the client to chase by Wednesday. Trigger: no reply by Wednesday noon."
What to fix: the score. The model gave likelihood 2 but its own reason ("already four days late") argues for 3. Because it is your register, you change it to 9, and R2 moves to the top. Read every score against its reason line, since mismatches like this are common. Also delete anything the notes do not support; a model sometimes adds a stock risk such as "scope creep" to look thorough. If scope creep is not in your week, it is not on your register.
The five-minute weekly review prompt
Here is last week's risk register and this week's notes. For each risk, say: no change, score up, score down, or closed, with the evidence from the notes. List any new risk not on the register. Do not change scores yourself; recommend them.The final sentence keeps the decision with you. In use, the output is a list of five or six proposed changes that take two minutes to accept or reject. The register stays current because updating it no longer feels like a chore. For risks of using AI itself in the business, which is a different register, a simple AI risk register for small businesses has a template.
The handover checklist, built from the project folder
Handovers fail on what the giver forgot to mention, not on what they wrote. A handover is complete when the receiver can answer five questions without calling you: what is live, what is owed, who the people are, where the files are, and what could go wrong next. Build the checklist around those five.
Create a handover checklist for a project moving from delivery to aftercare. Use the headings: What is live; What is owed (to us and by us); People and contact roles; Where things are; What could go wrong next; First week actions for the receiver. For each item use a checkbox and a one-line description. Use only the information below; mark anything unknown as [ASK DELIVERY LEAD].
Project facts: [paste your summary, decisions log and a list of folder names]An illustrative extract of the result for the fit-out:
What is live
[ ] Shops 1 and 2 signs installed 28 Mar; shop 3 installed 4 Apr
[ ] Window graphics on shop 2 awaiting a final inspection by the client [ASK DELIVERY LEAD: date]
What is owed
[ ] Final invoice to client after sign-off of all three shops
[ ] $380 variance on shop 2 projecting sign: agreed by client in writing? [ASK DELIVERY LEAD]
What could go wrong next
[ ] Shop 3 illuminated letters: supplier warranty 12 months; contact in folder 07Sit down with the receiver and walk through each item. Every [ASK DELIVERY LEAD] gets an answer in the meeting, and the answers go in the document. A handover that leaves question marks in it is a handover that rings you at 7pm. For examples of the same idea in other settings, hotel shift handover notes with ChatGPT and auto repair shop SOPs for intake and handover show how a checklist is built and checked in practice.
A shorter report for a busy owner
Not every reader wants six sections. A firm owner who glances at ten projects wants a version that fits on a phone screen. Keep the full report as the record and ask the assistant for a short version of the same facts, rather than writing a second report:
Rewrite this status report in five lines for a busy owner: line 1 status colour and project name; line 2 the one thing that changed; line 3 the one thing that could go wrong; line 4 what I need from the owner, or "nothing"; line 5 next milestone and date. Use only what is in the report.Illustrative result: "AMBER, shop-front fit-out. Shop 3 artwork is four days late. Risk: the second install weekend may move. Need: nothing yet; I will ask on Wednesday if the landlord has not replied. Next: fabrication starts Monday, 24 Mar completion." Five lines means the owner reads ten of them in the time it takes to read one full report. Check the short version against the long one, because compression is where small changes of meaning creep in: "nothing yet" and "nothing" are different promises.
Change requests and decisions: the two logs behind the report
Two small logs prevent most arguments at the end of a project. A change log records every request that alters scope, cost or date, who asked, what it cost and who agreed. A decision log records what was decided, by whom, and why. Both are tables with a line per entry, and both are easy for the assistant to maintain from your scratch note.
From this week's notes, extract (1) any change to scope, cost or dates, as: date, requested by, what changed, cost impact, agreed by whom (or NOT AGREED); (2) any decisions: date, decision, decided by, reason. Do not infer agreement. If a request was made but not confirmed in the notes, mark NOT AGREED.In the fit-out, the output flags one line as "projecting sign quote $380 over budget, requested by fabricator, NOT AGREED". Without that flag, the extra cost would appear in the final invoice with no record that the client had said yes. The rule "do not infer agreement" is the most important line in the prompt. Models are inclined to treat silence as consent, and in a dispute that habit is expensive.
A kick-off check and a close-out note
The start and end of a project are good places for a short AI-drafted note, because both follow a pattern. For the kick-off, paste the signed scope and ask for a one-page list of what is in, what is out, key dates, who decides what and the three questions the team must answer in week one. Put the "what is out" list in front of the client early. Most scope arguments are about things nobody said were excluded.
At close-out, paste the final notes, the change log and the budget outcome, and ask for a lessons note: what ran to plan, what slipped and why, what you would change, and one rule for next time. Illustrative result for the fit-out: "Artwork approvals from a third party (the landlord) took longer than the plan allowed in two of three shops. Rule for next time: add a landlord approval step with its own date to every multi-site quote." That rule goes into your next project's risk register as a standing item, which is how a project report improves the next project rather than sitting in a folder.
A second setting: a six-week office refurbishment
To check the templates are not specific to one trade, try them on a different project: a six-week office refurbishment run by a three-person interiors firm. The status colour rules change (a milestone is the day the joiner starts, not artwork approval), the risk register gains a supplier lead-time risk for furniture that the firm does not control, and the handover gains a "snagging list" section for small defects the client reports in the first month. The structure stays identical. Only the plan dates, the risk statements and the "what is live" section change, which is why the prompts are built to hold the template steady and take the facts as input.
That is the test of a good template: a different project needs different facts but no new document. If you find yourself redesigning the template for every job, it is too specific, and the AI will be producing a different document each week that nobody learns to read.
The sign-off rules that keep you honest
The rule that holds the routine together is that a person with accountability checks and signs each document before it leaves. Write that into a table and stick to it:
| Document | Who signs | What they check before sending |
|---|---|---|
| Status report | Delivery lead | Colour against rules; every date and figure against the plan and invoices; no sentence not in the notes |
| Risk register | Delivery lead, with each owner agreeing their risks | Scores against reasons; no invented risks; every live risk has an owner and next action |
| Handover | Delivery lead, then the receiver | All [ASK] items answered; receiver can answer the five questions |
Where documents go to clients, do not send a report you have not read in full. A client reading "all on track" will remember it in the dispute, whoever wrote it. For the general pattern of reviewing AI output without slowing the work, how to set up human review for AI work lays out lightweight ways to do it.
When the reports said green and the project was not
Illustration. A delivery lead's weekly status for a client had been green for six weeks. The notes, however, said "waiting on landlord" every week, and the assistant, asked only to "write a positive, concise update", never called it a risk. The work reached the installation weekend with no approval and the date slipped by a fortnight. The client's complaint was that nobody had warned them. The fix was threefold: the colour rules above, a standing "Risks and issues" section that the prompt must fill, and a rule that anything appearing in the notes for two weeks running is listed as a risk with an owner. A "positive, concise update" is the wrong instruction for any report: ask for accurate and short, and the assistant stops smoothing.
Where the tools sit
Everything here runs in ChatGPT, Claude or Microsoft Copilot with notes pasted in, so you can start without changing your project tool. If your project tool has its own AI, treat it as another route to the same documents, with the same checks. Asana's Starter and Advanced plans, for example, include five AI requests per user per month (up to 50 per account), with extra requests charged at $0.50 to $0.60 each, so a status report written there is not free. Check the current allowance on the vendor's page.
Use a business plan where your content is not used to train the model by default, and strip client names, addresses and contact details from what you paste. Keep any project containing personal data of the client's customers out of a chat tool unless your contract and the tool's terms both allow it; if you are unsure, ask the client or your adviser first.
How to tell whether it is working
Pick three numbers and write them down before you begin. The time you spend on status reports, risk reviews and handovers in a typical week. The number of times in a month a client or colleague asks "what is the status of X?" between reports. The number of calls you take after a handover that the checklist should have answered. For the fit-out lead, an illustrative before and after: weekly reporting from 90 minutes to 35, the risk register moving from "updated when someone shouts" to every Friday, and aftercare calls in the first fortnight falling from seven to two. Your numbers will differ.
The cost is small. A chat assistant on an individual plan is about $20 a month, and a team seat on ChatGPT Business or Claude Team is around $20 to $25 a month with a two-seat minimum, at the list prices checked for this series. The weekly cost that decides whether it works is your own ten minutes of checking. If you skip those, the first wrong sentence in a client report costs more than a year of subscriptions.
Start with the status report for one project, for four weeks, and nothing else. When that is boring, add the risk review. Add the handover checklist the next time a project changes hands, while the need is clear. For managers who also run a team, the weekly routine in a weekly AI routine for small-team leads covers the people side that sits beside delivery.
Questions delivery leads ask before letting AI near project reports
Can AI decide whether a project is green, amber or red?
It should not. Write the rules for each colour in advance, for example amber when any milestone is more than three days late, and have the AI check your facts against them. The status that goes out is yours, because it is a statement to a client or an owner and you answer for it.
What should I feed the AI for a weekly status?
Your own raw notes from the week: what finished, what slipped, decisions made, open issues and costs. Add the plan dates. Do not paste whole email threads with client contact details, and never share confidential client data in a personal account. Keep names to roles where you can.
How many risks should a small project have on its register?
Usually five to ten live risks is enough for a small project. A register with fifty items is never read. Keep only risks with an owner, a trigger and a next action, and move the rest to a closed list when they pass or stop mattering.
Who signs off an AI-drafted handover?
The delivery lead signs it as accurate and the person receiving it confirms they can run the project from it. A handover is complete when the receiver can answer five questions without asking you: what is live, what is owed, who the people are, where the files are and what could go wrong next.
Further reads
- How to Turn Meeting Notes Into Tasks Automatically With AI — Feed meeting notes straight into your task list.
- AI for Interior Design Studio Admin: Sourcing, Quotes, Updates — Sourcing, quotes and updates for a design studio.
- AI Consultant Handover Checklist: What You Need Before They Leave — What to collect before an outside helper leaves.
- AI Error Log: Track Mistakes and Stop Them Happening Again — Log AI mistakes so they stop repeating.
- AI for HR in Small Businesses: 12 Tasks You Can Hand Over Safely — Twelve people-related tasks safe to hand over.
- Shift Handover Template: Let AI Turn Operator Notes Into a Report — A fillable shift handover template and an AI prompt that turns rough operator notes into a report, with safety items protected.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: general project-management practice (status reporting, risk registers, handover checklists); Asana, ChatGPT Business and Claude pricing pages (checked 5 November 2026). The scenario and figures are illustrative.