Deliverable 4 · AI plan¶
Your city is running: the map is drawn, the tech tree is unlocked, the utilities are rented from the cloud. Now the game offers you an automatic advisor. Before you switch it on, a good mayor asks four questions. Which jobs would it do? What does it need to watch, and does it pay for itself? Who else uses one, and what does it change? And what could go wrong?
Deliverable 4 is that decision for your company: an AI plan. Your group proposes where AI would really help, which data it needs, whether it pays, what the sector is already doing, what it would be built with, and the legal risk level of each proposal. One member also builds a working demo in the browser. It is the fourth chapter of your digital transformation dossier.
What you will learn
- Propose concrete processes of a real company that AI could automate or optimise, and measure the improvement
- Say which data each proposal needs and estimate its profitability with a spreadsheet
- Describe the AI landscape of the company's sector, present and future, with cited sources
- Choose languages and tools for the proposals and show a model that works
- Classify each proposal under the AI Act and write rules for its responsible use
What you hand in¶
| Format | One PDF per group, about 6–10 pages without the cover |
| Where | Google Classroom, assignment "D4 · AI plan". The Coordinator hands it in |
| Deadline | Monday 22 March 2027, 23:59 |
| Defence | Individual, in session 19, Monday 22 March 2027 (the same day) |
| Weight | 70% of the RA4 mark. The other 30% are the class activities of sessions 15 to 18 |
Every fact about your company needs a source
Everything you state as a fact about your company or its sector (what it does, what systems it has, what AI it or its competitors already use, any figure) must be found and cited: the company website, annual report, press releases, the press, Eurostat, Ibestat. Each citation has a title, the author or website, the link and the date you consulted it. What you cannot find you may still propose with reasons, but you label it hypothesis. A claim without a source or a label counts as an error. A source that says "forecast" or "automated" does not say "AI": do not add the word yourself.
No generative AI
Generative AI is not allowed in this deliverable. The only AI authorisation of the unit was the error-hunting activity in session 16, and it ended in class. A section written with AI or copied scores 0, and the defence will show it: you cannot explain a text you did not write. The Teachable Machine model in Section D is not an exception to this rule: it is the object you build and test, not a tool that writes for you.
The sections¶
Each member owns and signs one section (the individual part). With 3 members, Section D becomes part of the group part and everybody builds it together.
| Section | Criterion | What to write | Owner |
|---|---|---|---|
| A · Processes to automate or optimise | RA4 a) | Two processes of the company that AI could improve. For each: what it is, who does it today and how (manual, by rules, already with AI); what the AI would do (classify, extract, predict, recommend, generate) and whether it automates or optimises; why a fixed rule is not enough; the indicator that would show the improvement; who supervises the AI and what happens when it is wrong. | Individual |
| B · Data and profitability | RA4 b) | For each process of Section A: which data the model needs, whether the company already has them (and in which system, from your earlier deliverables), whether any are personal data, and how much data there would be. A profitability estimate in a spreadsheet for at least one process: monthly saving or extra revenue, costs (service, set-up, supervision), net benefit, payback period and first-year ROI, plus one sensitivity case. Every number without a source is labelled hypothesis. Link to the sheet. | Individual |
| C · The sector's AI landscape | RA4 c), d), f) | Where the company's sector stands: Eurostat figures and at least two real cases of AI in the sector, one of them in the Balearic Islands if possible, cited. What AI can and cannot do today in this sector, and what is likely to change in the next years (separating facts from forecasts). How AI changes the work of the company's IT department and its developers: tools, new roles, risks. | Individual |
| D · Languages, tools and a working demo | RA4 e) | Which languages, libraries and services the proposals of Section A would be built with, and why (training or calling a model through an API?), supported by what you found on GitHub and in developer surveys. A working demo in the browser related to the company, normally a Teachable Machine model (image or sound): its classes and number of examples, a table of at least 10 tests with new data, the hit rate, its clearest error and why, and its limits. Screenshots and, if you saved or exported it, the link. | Individual (4th member) |
| Introduction | RA4 c) | The company in two lines (link to your earlier chapters), what an AI plan is, and why it makes sense, or not, for this company now. | Group |
| Risk level and responsible use | RA4 c) | For every proposal, its risk level under the AI Act (Regulation (EU) 2024/1689) with a one-sentence reason and what that level obliges the company to do. Five rules of responsible use for this company, at least one specific to its sector. | Group |
| Integration | RA4 a)–f) | One table that joins the sections: proposal · process · data · profitability · risk level · owner. One paragraph saying which proposal you would do first and why. | Group |
| Sources | All | One list of every source cited, in the same format, with the date consulted. | Group |
| Group log | All | Who owned which section, who had which role, what you decided and when (template on the project page). | Group |
Roles in this deliverable¶
Choose one role each and write it in the group log. Choose a section and a role you have not had before, if you can.
| Role | In Deliverable 4 |
|---|---|
| Coordinator | Shares out the sections in session 18, sets an internal deadline (we suggest Friday 19 March), checks the group log and hands in the PDF |
| Source checker | Opens every link; checks that every claim of "this company uses AI" has a source that really says so; labels hypotheses |
| Technician | Checks the spreadsheet formulas and that the demo works; makes sure the sheet is shared with view access |
| Editor | Format, spelling, one voice; the same process has the same name in every section and in the integration table. With 3 members, the Coordinator does it |
How it is assessed¶
| Part | Weight | What is assessed |
|---|---|---|
| Your section | 40% | Content correct for its criterion, real and cited sources, hypotheses labelled, reasons and numbers given, clear writing |
| Your defence | 20% | You explain your section in your own words, answer why, and know how it connects to another member's section and to the risk levels |
| Group part | 40% | Introduction, risk level and responsible-use rules, integration table and priority, coherence between sections, sources list, format and spelling, group log |
Rubric¶
| Part | Excellent | Adequate | Insufficient |
|---|---|---|---|
| Your section | Everything the table asks for, about this company; real cases and figures cited; numbers that add up (the sheet works); hypotheses labelled and reasoned; terms used precisely | Most of what is asked; some generic examples or a missing source; the main ideas and numbers are correct | Missing parts; text that could be about any company ("AI will improve efficiency"); facts without sources; "AI" claimed where the source does not say it; numbers that do not add up |
| Your defence | Answers fluently with reasons and numbers; knows where each fact comes from; explains a teammate's proposal and its risk level | Answers the main questions with some help; knows the sources of the key facts | Cannot explain parts of the section, or does not know where its content or numbers come from |
| Group part | Every proposal has a justified risk level and its obligations; rules are concrete and fit the sector; the integration table joins all sections; the priority is argued; clean format, one voice, complete log | Risk levels given with weak reasons; generic rules; integration with small contradictions | Risk levels missing or wrong; no rules; four separate texts glued together; no log; many spelling or format errors |
Programme rules: RA4 needs at least a 5. Late work is marked out of 5; more than 48 hours late is a 0. Plagiarism or unauthorised AI is a 0. If you fail RA4, the recovery is an individual task: you redo one section of this deliverable alone and defend it, for a maximum of 5.
The defence¶
On Monday 22 March 2027 (session 19) the groups finish the deliverable while the teacher calls each student for 2–3 minutes. No paper and no computer: a conversation.
- Two questions about your own section. For example: "Why is a fixed rule not enough for this process?", "Where does the monthly saving in your sheet come from, and which numbers are hypotheses?", "Does your source really say the company uses AI?", "Show me, in words, one error of your model and why it happened."
- One question about another member's section or the group part, to check that the group really put it together. For example: "What risk level does your teammate's proposal have, and why?", "Which proposal would you do first?"
If you cannot come on 22 March for a justified reason, tell the teacher in advance: your defence is moved to the start of the next session.
Before you hand in¶
| Check | Who |
|---|---|
| Every section has its owner's name | Coordinator |
| Every fact about the company and the sector has a working source; every guess says hypothesis | Source checker |
| No source is quoted as saying "AI" when it does not | Source checker |
| The spreadsheet opens with the link, its formulas work and the numbers match the text | Technician |
| The demo has at least 10 tests with new data, and screenshots | Technician |
| Every proposal of Section A appears in Section B, in the risk table and in the integration table, with the same name | Editor |
| Every proposal has a risk level with a reason; there are five rules of responsible use | Editor |
| The group log is filled in and signed | Coordinator |
| One PDF, uploaded to Classroom before Monday 22 March 2027 at 23:59 | Coordinator |
Start from what you did in class: your Teachable Machine report from What is AI, your rules and risk levels from Generative AI and responsible use, your real case and spreadsheet from Automation and profitability, and your GitHub table from Sectors, languages and software.