Build one focused, feasible proposal for a 2,000-word report and a 10-minute presentation. Reveal examples, diagnose common failures and develop your own work as you progress.
The one alignment chain
Every strong proposal is an argument for a future investigation. Each choice must logically create the need for the next.
Do not copy the scale of the long model answer. It is about 3,170 words before references and appendices. Copy its logic, then compress it to the assessed 2,000-word limit.
Failure clinic
Open each card to see the symptom, the reason it loses marks, a better version, and a quick task.
1. A topic instead of a question
Weak“The impact of AI on business.”
Reveal diagnosis and repair
Why it failsThere is no defined practice, outcome, population, setting or realistic boundary. Almost any evidence could appear relevant.
Improved“How does employee use of generative-AI customer-service tools influence perceived response quality in UK retail banking teams?”
Try it: Which part of the improved question sets the context?
“UK retail banking teams” identifies the bounded context and population.
2. A descriptive literature catalogue
Weak“Smith (2024) says… Jones (2025) states… Ali (2023) found…”
Reveal diagnosis and repair
Why it failsThe reader sees separate summaries but no debate, comparison, evaluation or gap.
Improved“While survey studies associate AI support with faster responses (Smith, 2024; Ali, 2023), Jones’s (2025) interview evidence suggests speed may reduce employees’ checking behaviour. The disagreement may reflect different measures of quality.”
Test: What makes this synthesis rather than summary?
It groups sources around one claim, compares findings and methods, and explains a possible reason for disagreement.
3. A methods shopping list
Weak“The research will use surveys, interviews and secondary data.”
Reveal diagnosis and repair
Why it failsMore methods do not automatically create a stronger study. Each method adds access, ethics, analysis and word-count demands.
Improved“Semi-structured interviews with 8–10 customer-service employees will explore how staff judge AI-generated responses. Interviews are preferable to a closed survey because the question seeks explanations of decision processes.”
4. Generic research philosophy
Weak“Interpretivism is about subjective meanings. Positivism is about facts.”
Reveal diagnosis and repair
Why it failsTextbook definitions do not justify the knowledge this particular study needs.
Improved“An interpretivist position fits the aim because the study examines how employees construct and explain ‘quality’ during AI-assisted work, rather than treating quality as a single objective measure.”
5. Unrealistic access
Weak“A survey will be sent to all UK bank customers and senior executives will be interviewed.”
Reveal diagnosis and repair
Why it failsNo sampling frame, gatekeeper, recruitment route or fallback is identified.
Improved“Participants will be recruited through two professional banking networks accessible to the researcher. If fewer than eight eligible employees volunteer, the study will use publicly available practitioner webinars as a bounded documentary dataset.”
6. Two disconnected submissions
WeakThe report studies employees, but the slides propose a customer survey.
Reveal diagnosis and repair
Why it failsThe presentation and report are two formats for one proposal—not opportunities to redesign the project.
ImprovedUse identical question wording, participants, sample, collection method and analysis plan in both outputs.
7. Naming software instead of analysis
Weak“The interviews will be analysed using NVivo.”
Reveal diagnosis and repair
Why it failsSoftware organises material; it is not an analytic procedure.
Improved“Reflexive thematic analysis will involve familiarisation, initial coding, candidate theme development, review and naming. Themes will explain recurring ways employees assess and correct AI outputs.”
8. Inventing findings or impact
Weak“The research will prove that AI increases customer satisfaction by 20%.”
Reveal diagnosis and repair
Why it failsA proposal has no findings yet, and the design may not support causal or numerical claims.
Improved“The study could help managers identify where staff guidance and checking procedures are needed when AI tools are introduced.”
Research-question lab
Build through a narrowing funnel
Stress-test the question
Useful question patterns
How and why does [practice] influence [outcome/process] within [bounded context]?
How do [participants] experience [phenomenon] in [setting]?
To what extent is [factor] associated with [outcome] among [population]?
How does [case A] compare with [case B] in relation to [criterion]?
Literature: build a debate, not a list
Claim
What is argued?
Group sources that address the same issue. State agreement and disagreement.
Evaluation
How strong is it?
Compare contexts, samples, methods, measures, dates and limitations.
Implication
What remains?
Explain what the pattern means for your question, evidence need and design.
Synthesis paragraph builder
Avoid: “There is no research on…” unless a systematic search can defend that claim. Prefer “Limited evidence explains…” or “Existing work has concentrated on X, leaving Y underexplored.”
Methods alignment builder
Complete each row in one or two precise sentences. If the chain breaks, narrow the project before adding more methods.
Feasibility gate
Write to the 2,000-word architecture
Section
Words
Job of the section
Common trap
1. Introduction and focus
350
Context, problem, rationale, aim, question, objectives and scope.
A long company history or broad topic overview.
2. Critical literature review
650
Synthesise debates, use theory and establish a defensible gap.
One paragraph per author; theory used as decoration.
3. Methodology and ethics
700
Justify approach, design, sample, collection, analysis, quality, ethics and limitations.
Textbook definitions and unconnected method lists.
4. Alignment and feasibility
200
Show the design answers the question; state access, fallback and schedule.
Assuming access or presenting a vague timeline.
5. Anticipated contribution
100
State possible academic/practical value without inventing findings.
Guaranteed impact or unsupported predictions.
Total
2,000
Confirm institutional rules for references, tables and appendices.
Recommended scale
One main question, no more than two or three sub-questions, and three or four observable objectives. Each objective should correspond to evidence the design can collect and analyse.
Concision exercise
Reduce a paragraph by 35–40% while preserving its claim, evidence, critique and link to the question. Explain what every deleted sentence failed to contribute.
The same proposal in 10 minutes
Time
Slide
Essential content
0:00–0:45
1. Title and decision
Topic, case/population and one-sentence proposal.
0:45–1:45
2. Problem and significance
Problem, evidence of relevance and scope.
1:45–2:30
3. Question and objectives
Main question and compact objectives.
2:30–4:20
4. Literature and gap
Two or three debates, framework and unresolved gap.
4:20–7:15
5. Proposed methodology
Approach, design, sample, collection, analysis and justification.
7:15–8:20
6. Ethics, feasibility and plan
Risks, mitigation, access, contingency and timeline.
8:20–9:30
7. Alignment and contribution
Why this design answers the question and expected value.