Unit 4004 AI SOP Generator Project Sample Answer

1. Project introduction and justification

Standard operating procedures are the written record of how a task is done safely and repeatably. In a small manufacturing cell, they are usually produced by hand from a company template, copied from an older SOP, and corrected after an audit. The process is slow, and its quality depends on whoever has the time to write. Documented procedures also sit at the centre of a quality management system (International Organization for Standardization, 2015), so errors in them carry a real cost.

This project asks one question: to what extent can an AI-assisted SOP generator consistently and reliably improve the SOP drafting process without unacceptable errors, extra review effort or disruption? The application will accept a company template and process notes, produce a draft, flag unclear or missing information for the user to resolve, and export an edited version for printing. The AI draft is never treated as approved. A named human reviewer signs off every SOP, because the tool supports the engineer and does not replace the engineer’s responsibility.

The project is justified on measurable grounds. If drafting time falls while accuracy and reviewer workload stay acceptable, the company gains capacity. If they do not, the finding is still a valid outcome for LO3 and LO4, provided the testing is honest and the evidence is recorded.

2. Stakeholders and current baseline

The baseline must be measured before any AI is used, otherwise there is nothing to compare against. I will time the current manual process for five existing SOPs, record the number of reviewer corrections per SOP and note how often information is missing at first draft. The figures in the table below are illustrative values showing the format; real values replace them after observation.

3. Research question, aim and objectives

Aim: to evaluate whether an AI-assisted SOP generator can reduce drafting effort while maintaining accuracy, completeness and usability, with a human reviewer retained throughout.

1. Establish the manual SOP baseline for time, corrections and completeness by week 3.
2. Specify requirements and acceptance criteria agreed with the quality lead by week 4.
3. Develop a working prototype using AI-assisted development, with version control, by week 8.
4. Test the prototype on at least ten process descriptions and record accuracy, consistency and missing-information detection by week 10.
5. Run at least two improvement cycles, retesting after each, by week 12.
6. Evaluate results against the baseline and report conclusions and recommendations by week 14.

Each objective is specific, has a measure and has a deadline, which is what makes it SMART rather than a general intention.

4. Scope and success criteria

In scope: text process descriptions, a user-supplied template, draft generation, a list of flagged gaps, editing and export to a printable file. Out of scope: automatic approval, integration with the company document control system and any live production use before reviewer sign-off. Photo and voice-note input is a stretch feature, added only if the core objectives are met on time.

The accuracy criterion matters most. Generative models can produce fluent text that is unsupported by the source, a failure widely described as hallucination (Ji et al., 2023). In an SOP, one invented torque value or a skipped isolation step is a hazard, not a typing slip, so any invented content is recorded as a critical defect regardless of the percentage score.

5. Feasibility study

Option C is feasible because the development assistant and cloud credit are available to students, but both have limits, so the plan includes a usage log and a fallback of reducing the test set rather than overspending. Assumptions are stated openly: the company agrees to share non-confidential process notes, a reviewer is available weekly, and the model service stays available during testing.

6. Research methodology

Secondary research will cover documented-procedure requirements, AI risk guidance and the limits of language models. Primary research will use three instruments: timed trials of drafting with and without the tool, a reviewer scoring sheet, and a short operator questionnaire on readability. Because the population is small, I will use purposive sampling of the people who actually write, review and follow the SOPs, and I will say plainly that the findings are indicative and not statistically generalisable.

Quantitative data (time, error counts, flagged gaps) will be analysed with simple descriptive statistics and compared with the baseline. Qualitative comments will be coded into recurring themes. Each test run will log the input, the prompt version, the model setting, the output and the reviewer score, so any result can be repeated and checked.

7. Project planning and management

Dependencies, milestones and the critical path follow standard project management practice (PMI, 2021). The critical path runs A, B, C, E, F, G, a total of 14 weeks on these illustrative durations. Task D has slack because it can run alongside the build, so a delay of a week there does not move the end date. The Gantt chart will show these bars with milestones at the proposal and feasibility review, the logbook review and the interim report. The resource plan covers my own time, the reviewer’s weekly hour, tool subscriptions, hosting credit and a small budget line for usage overruns.

A dated logbook will be kept throughout, with an entry for every decision, test and change to the plan.

8. Legislation, ethics and risk management

Sample logbook entry (Phase 2)

Week 9, test round 1. Ten process descriptions were run through the prototype. Seven drafts were usable after light edits. Two omitted a lock-out step that the source notes had mentioned only indirectly. One invented a cleaning interval that appeared nowhere in the input, which was logged as a critical defect. Action: the prompt now instructs the model to output “information required” instead of estimating, and the gap-flagging check will be extended to safety-related steps. Plan change recorded: cycle 2 starts one week early, and the usability survey moves to week 12.

Evaluation plan

The conclusion will be written from whichever result the evidence supports. If review time cancels out the drafting time saved, the honest finding is that AI helps only partly, and a Distinction-level report says so and explains why, with recommendations on where human review effort should be concentrated. Your own report should map its evidence to LO1 to LO4 in the same way.