FIN306 CW1: Responsible Data Science in FinTech
Assessment Brief — 2025/26 Semester 2
| Aspect | Details |
|---|---|
| Weighting | 30% of module mark |
| Format | Recorded presentation (5 minutes) + slide deck (6–8 slides) + data risk register (1–2 pages) |
| Submission deadline | 6 March 2026, 23:59 via Blackboard |
| Submission method | One ZIP file containing all three deliverables |
| Topic confirmation deadline | Week 5 (email or Blackboard discussion) |
Purpose
This assessment develops your ability to think critically about data quality and bias in a real FinTech application, the foundation of responsible data science practice. You record a concise 5-minute pitch supported by a slide deck and a data risk register.
The assessment serves two purposes:
- Summative (30%): Demonstrate your understanding of data generating processes, selection bias, and responsible practice using Week 2 concepts.
- Formative: Develop the habit of systematically auditing data quality before analysis, a skill applied directly in the data quality section of Coursework 2.
Why this format is worth 30%
The deliverables are deliberately compact but cognitively demanding. The 5-minute cap forces you to prioritise: you must filter a large amount of possible content into a clear, evidence-based argument. That discipline (deciding what to leave out, ordering ideas, staying within time) is part of what we assess.
The risk register then carries the systematic audit that does not fit into 5 minutes. It shows you can document multiple data quality risks with specific mitigations in a format used in professional practice. Together, the recording demonstrates oral communication and conceptual clarity; the register demonstrates structured analytical thinking. Both are needed for responsible data science. We assess the combined package, not only the length of the video.
Intellectual judgement in a world of generative AI
The tight 5-minute limit is not an arbitrary constraint. It is a test of something that generative AI cannot do reliably: deciding what matters. A language model can produce pages of relevant-sounding content in seconds. The skill this assessment targets is knowing which of those pages to keep and which to discard, and being able to defend that choice clearly and confidently.
In financial services, this matters enormously. Analysts, risk officers, and compliance teams operate under time pressure with incomplete information, and they are ultimately accountable for their judgements in ways that AI tools are not. As AI becomes more embedded in financial processes, the professionals who can evaluate AI-generated output critically, spot what is missing, and communicate a clear position will be the ones with durable careers. Producing content is no longer the scarce skill. Evaluating it is.
The recording is your argument; the risk register is your evidence. Together they demonstrate the kind of considered, strategic judgement that financial services will always require from humans.
What You Need to Submit
One ZIP file containing all three deliverables, uploaded to the FIN306 CW1 assignment on Blackboard by 23:59 on Friday 6 March 2026.
| Deliverable | Format | Length |
|---|---|---|
| Recorded presentation | MP4, or Kaltura/Panopto link in a .txt file | 5 minutes (±30 seconds) |
| Slide deck | 6–8 slides | |
| Data risk register | PDF or Word | 1–2 pages (8–12 risks, table format) |
File naming (use your student number):
B00123456_FIN306_CW1_Recording.mp4(or a .txt file with the link)B00123456_FIN306_CW1_Slides.pdfB00123456_FIN306_CW1_RiskRegister.pdfB00123456_FIN306_CW1.zip(contains all three)
Topic Areas
Choose one of the four applications below. Each maps directly to Week 2 concepts.
| Application | Core data quality issue | Week 2 connection |
|---|---|---|
| Robo-advisor backtests | Survivorship bias in historical fund/strategy data | Survivorship bias simulation and Bloomberg lab |
| AI credit scoring | Selection bias in alternative data (digital footprints, transaction history) | Selection mechanisms, validity vs reliability |
| Algorithmic trading strategies | Look-ahead bias in signal construction | The Golden Rule of backtesting |
| Fraud detection systems | Class imbalance as a form of selection bias; fairness | Selection bias, DGP, responsible practice |
Confirm your topic by sending a one-sentence description via email or Blackboard discussion by Week 5. You will receive brief feedback before you start the main work.
The Recorded Presentation (5 minutes)
Address three things in sequence:
1. Application context (c. 1 minute)
What problem does the FinTech application solve, what data does it use, and who relies on it?
2. Data quality analysis (c. 3 minutes)
Focus on one primary bias type from Week 2 (survivorship, look-ahead, or selection) and answer:
- How does this bias arise in the data generating process?
- What is the direction and rough magnitude of the distortion?
- Who is harmed if the bias goes undetected?
Depth on one bias is better than breadth across many.
3. One responsible practice recommendation (c. 1 minute)
What would reduce or disclose this bias? Reference a professional standard or academic finding where possible (FCA guidance, Bloomberg survivorship-free databases, a cited paper).
Recording guidance:
- Screen-record your slides with your voice over them
- Speaking pace: approximately 120–130 words per minute
- Rehearse at least twice; you can re-record as many times as you like
- You do not need to appear on camera, but may if you wish
- Submit as MP4 or include a link if using Kaltura or Panopto
The Slide Deck (6–8 slides)
Your slides support the recording and are assessed separately. They must stand alone as a readable document.
| Slide | Content |
|---|---|
| 1 | Title, student number, date |
| 2 | Application overview (what it is, what data it uses) |
| 3 | Data generating process: where does the bias enter? |
| 4 | Bias analysis: direction, magnitude, who is affected |
| 5 | Evidence: one academic finding or industry example |
| 6 | Responsible practice recommendation |
| 7 | References (Harvard style) |
Keep each slide to 4–6 bullet points. One key idea per slide.
The Data Risk Register (1–2 pages)
A data risk register is a structured tool used by data scientists, quants, and risk officers to document known data quality concerns before analysis begins. Produce a register of 8–12 distinct risks for your chosen application.
The register is not a repeat of your slides. It is the place for the systematic audit that cannot be fully conveyed in 5 minutes. Strong presentations reference the register explicitly (e.g. “as set out in my risk register, R1 and R3 are the two most critical risks”).
Template:
| Risk ID | Risk Description | Bias Type | Data Source Affected | Likelihood (H/M/L) | Impact (H/M/L) | Mitigation / Rationale |
|---|---|---|---|---|---|---|
| R1 | Historical fund returns exclude funds that closed | Survivorship | Internal returns database | H | H | Use survivorship-free database (Bloomberg EQQQ); include delisting returns. Justifies H/H: backtests otherwise systematically overstate returns. |
| R2 | … | … | … | … | … | … |
Guidance:
- Aim for 8–12 distinct risks, not variations of the same one. Breadth and specificity both matter.
- The Mitigation / Rationale column should (a) state a specific mitigation and (b) briefly justify your likelihood and impact ratings where non-obvious.
- At least three risks should relate to the primary bias you discussed in your presentation; the others should cover secondary quality concerns (missing data, measurement error, validity, reliability).
Assessment Criteria
| Criterion | Weight | What we look for |
|---|---|---|
| Content | 40% | Accurate bias identification, DGP understanding, quality of risk register analysis, appropriate scope |
| Communication | 30% | Clarity and confidence of recording, logical slide structure, readable risk register |
| Visual aids | 20% | Slide quality, use of evidence and examples, professional presentation |
| Academic rigour | 10% | Engagement with sources (FCA guidance, academic papers, Bloomberg documentation); Harvard references |
Key differentiator: Intellectual depth — can you explain why a bias arises in this specific application, not just that it exists?
Tips for Strong Work
On the recording: Practice timing out loud before you record. Most students run long on first attempts. If you hit 5:30, cut the context section, not the analysis.
On the analysis: Week 2 is your primary resource. The survivorship bias simulation, the Bloomberg UK banking lab, and the DGP checklist in the chapter are directly relevant. Cite them.
On the risk register: Aim for 8–12 distinct risks. Entries like “Data may be inaccurate” or “Model could be biased” score poorly. Strong entries name the specific data source, the mechanism, a concrete mitigation, and a brief rationale for likelihood/impact ratings.
On sources: For a 5-minute task, 3–5 sources is sufficient: one academic paper, one industry or FCA document, and the module chapter. Quality matters more than quantity.
Common Pitfalls
- Describing the business model instead of the data quality concern
- Listing bias types generically rather than applying them to your application
- Risk register rows that are vague or repeat the same concern
- Recording that runs over 6 minutes or under 3 minutes (both lose marks)
- Slides with dense paragraphs that duplicate the recording word for word
Frequently Asked Questions
Q: Why is the presentation only 5 minutes?
A: The 5-minute limit is deliberate. Deciding what to include and what to leave out is part of what we assess. The risk register carries the systematic detail that does not fit in the recording. Think of the presentation as your argument and the register as your evidence.
Q: Does the risk register need to be covered in the presentation?
A: No. Reference one or two of your highest-priority risks to show the deliverables are connected, but the register stands on its own as a separate document.
Q: Can I choose a topic not on the list of four?
A: Only with prior approval. Email a one-paragraph justification by Week 5. Unapproved alternatives will not receive credit.
Q: Do I need to show Python code?
A: No. This is conceptual analysis. If you ran code in the Bloomberg lab that produced a relevant figure, you may include it as evidence, but code itself is not assessed.
Q: Can I use personal or workplace examples?
A: Yes, with two conditions: publicly available information only, and supported by an academic or industry source.
Q: What if I cannot submit on time?
A: Request an extension via the University system before the deadline. Late penalties apply (−5% per working day; 0% after five working days unless mitigating circumstances are approved).
Academic Integrity
This is individual work. AI tools may assist with grammar checking or source finding, but all intellectual content (the risk register, slide commentary, and recorded explanation) must be your own. Materials will be checked via Turnitin. The module coordinator reserves the right to orally examine any student when unauthorised use of AI is suspected.
Contact: Professor Barry Quinn CStat, PhD · b.quinn1@ulster.ac.uk