This course provides students with an in-depth knowledge of the design and management of applied research projects. The course prepares students for the undertaking of the final project in year four of the degree programme.
The course teaches one method — the scientific method in the form most usable inside an organisation — and applies it end to end across fifteen weeks to a real dataset, with a 30-minute computational lab in every session from Session 3 onwards. The method is McKinsey’s seven-step problem-solving process, set out in full in Conn and McLean’s Bulletproof Problem Solving: define the problem, disaggregate the issues, prioritise, build a workplan, conduct the analyses, synthesise the findings, communicate the story. It is the same engine as Rasiel’s issue-based problem solving and as Friga’s TEAM FOCUS; §4.1 gives the mapping.
The seven steps are carried by four techniques, and it is the techniques the course drills:
• MECE decomposition (step 2) — a logic tree whose branches do not overlap and whose leaves reconstruct the metric exactly.
• Pareto prioritisation (step 3) — concentration arithmetic performed on the metric fixed in step 1, which is what licenses a student to ignore most of their own tree.
• Hypothesis testing (step 5) — design that could prove the student wrong: power, pre-registration, leak-free validation, causal identification.
• Minto’s Pyramid Principle (steps 6–7) — governing thought first, MECE key line beneath it, evidence beneath that.
Hypothesis testing is restricted to quantitative evidence: machine learning across its four paradigms, big-data methods, causal inference, and generative AI used strictly as a co-scientist rather than as a source of facts.
The course serves the three Final Degree Project (TFG) options — the Academic Thesis, the New Venture and the Management Case — which between them contain six named routes. Every route runs the same seven steps; what changes is the audience and therefore the evidence standard.
Every student must have a formal advisor, internal or external, in place by the first week of December.
By the end of the course students will be able to set realistic research objectives, design accurate and appropriate research methods, undertake comprehensive literature reviews, analyse quantitative and qualitative data and write-up their findings. This course also aims to practically apply the theoretical concepts and develop management skills necessary for the successful undertaking of a research project.
Examples of interaction between the framework and the types of project are provided in the folowing table
FOCUS stage | Academic Thesis | New Venture | Management Case |
Frame | Research question and the gap in the literature it addresses | The riskiest assumption in the business model | The client’s key question, and whose decision it is |
Frame (the M) | Operationalised dependent variable | The unit-economic driver: CAC, LTV, conversion, payback | The KPI the decision-maker is judged on |
Organize | Conceptual model and hypotheses H1…Hk | Assumption tree of the business model | Issue tree of the value drivers |
Organize (prune) | Which hypotheses the available data can test | Which assumption, if false, kills the venture | Pareto on the KPI: where the value actually sits |
Collect | Secondary data, survey or panel, with provenance | Landing pages, interviews, pilots, market data | Client systems plus open benchmarks |
Understand | Estimation, robustness, limitations | Experiment results and cohort economics | Driver analysis with quantified impact |
Synthesize | Discussion and stated contribution | Investor narrative and roadmap | Recommendation and implementation plan |
Audience | An examining board | A sceptical investor or venture committee | A decision-maker who must act |
Skills acquired include: planning, organisation, time management, interpersonal communication, teamwork, leadership. Students will have the opportunity to apply knowledge developed in several subject areas including marketing management, global marketing, human resource management, cross-cultural management, project management and should provide experience in preparation for the final project.
Titular Professors
Understanding of the standard data methods used to analyze data.
Step 1 — Define the problem (LO1–LO3)
• LO1. Convert an ill-defined organisational symptom into a single-sentence SMART problem statement naming a decision-maker, a decision, a lever, a measurable metric, a threshold and a horizon.
• LO2. Specify that metric to a professional standard — construct, formula, grain, source, threshold, known distortions — and implement it as a tested, documented function.
• LO3. State falsifiable hypotheses with named killer tests, select a research design that can support the intended claim, and compute the statistical power and minimum detectable effect before collecting data.
Steps 2–4 — Disaggregate, prioritise, plan (LO4–LO5)
• LO4. Decompose a metric into a MECE issue tree that reconstructs it exactly, and detect the standard decomposition failures (silent overlap, method branches, mixed levels).
• LO5. Prioritise branches by Pareto arithmetic performed on the chosen metric, together with effort and value-of-information, and produce a workplan in which every row ends in a named end product.
Step 5a — Gather the evidence (LO6–LO7)
• LO6. Assess provenance, coverage, sampling and missingness; produce a reproducible data-quality audit and a data statement that lets a stranger re-pull the data.
• LO7. Work at the scale a question requires — columnar storage, out-of-core querying, deliberate sampling designs — and explain why volume does not remedy bias.
Step 5b — Conduct the analyses (LO8–LO11)
• LO8. Select among unsupervised, supervised, semi-supervised, causal and sequential methods on the basis of what each can legitimately claim, and validate each to the standard its claim requires.
• LO9. Build leak-proof supervised pipelines, choose a metric and an operating threshold from the decision’s cost structure, and report calibration and subgroup performance.
• LO10. Design and analyse experiments and quasi-experiments, and match the verb in a conclusion to the strength of the design.
• LO11. Use generative AI as an instrumented co-scientist — generation, criticism and ranking — behind a verification harness, and declare that use fully.
Steps 6–7 — Synthesise and communicate (LO12–LO14)
• LO12. Assemble findings into a Minto pyramid: a governing thought with a number, a MECE key line, and evidence beneath each point.
• LO13. Defend the resulting argument under questioning, bound it with specific, directional limitations, and state what would prove it wrong.
• LO14. Deliver a repository a stranger can clone and reproduce, with a complete AI-use appendix and an environment stamp.
Working practice (LO15)
• LO15. Secure and run a formal advisor relationship — scope agreed in writing, a documented meeting cadence, and issues escalated early — and evidence it in a monthly log.
The course’s primary vocabulary is the seven-step process. Each step has one output, and no step may begin before the previous step’s output exists.
Step | What it produces | Technique | Sessions |
1. Define the problem | One SMART problem statement: decision-maker, decision, measurable metric (the M), threshold, horizon | SMART | 2–3 |
2. Disaggregate the issues | A logic tree whose leaves reconstruct the M exactly | MECE | 4 |
3. Prioritise the branches | The same tree, ranked by share of the M per unit of effort, and pruned | Pareto | 5 |
4. Build the workplan | One row per surviving branch: hypothesis, analysis, data, owner, date, end product | Workplan table | 5 |
5a. Gather the evidence | A fact pack with provenance, sampling and known distortions documented | Sourcing, sampling | 6–7 |
5b. Conduct the analyses | Findings with effect sizes and honest uncertainty | Hypothesis testing | 8–13 |
6. Synthesise the findings | A governing thought supported by a MECE key line | Minto’s Pyramid | 14 |
7. Communicate the story | A document and a defence the audience can act on | Minto’s Pyramid | 15 |
Note the shape of the course: four of the seven steps produce no findings at all. Steps 1 to 4 are design, and they occupy Sessions 2 to 5 because that is where final-year projects are won or lost.
4.1.1 One method, four vocabularies
Students will be supervised, employed and examined by people who use different words for the same seven moves. The mapping below is what makes their skills portable; it is taught once, in Session 1, and then the seven-step column is used for the rest of the term.
7-step (this course) | Scientific method | FOCUS spine | Issue-Based Problem Solving |
1. Define the problem (SMART) | Observe, then state a testable question | F — Frame | State the issue |
2. Disaggregate (logic tree, MECE) | Deduce consequences | O — Organize | Issue tree |
3. Prioritise (Pareto, 80/20) | Decide which test is decisive | O — Organize | Prune the tree |
4. Build the workplan | Plan the experiment | O — Organize | Workplan |
5a. Gather · 5b. Conduct analyses | Gather evidence; test and falsify | C — Collect · U — Understand | Fact pack; prove / disprove |
6. Synthesise the findings | Draw a conclusion | S — Synthesize | So-what |
7. Communicate the story | Defend it publicly | S — Synthesize | Storyline |
On FOCUS and TEAM. Friga’s FOCUS — Frame, Organize, Collect, Understand, Synthesize — is a five-letter compression of the same sequence, and this course keeps it only as a structural label for its five parts (see §7). His TEAM half — Talk, Evaluate, Assist, Motivate — concerns how a group works rather than how a problem is solved. This course therefore treats it as working practice rather than as method: it appears as the advisor relationship and a short monthly log (§8, component 6), and is not examined as a framework. Students who want the full treatment will find it in Friga, chapters 2–3.
4.2 The Pareto–SMART link, stated explicitly
This course rests on a claim that is made explicit in Session 5 and used from then on:
Pareto prioritisation is arithmetic performed on the M in the SMART problem statement.
“80% of the effect comes from 20% of the causes” is meaningless until one says 80% of what. That “what” is the Measurable success criterion fixed in Session 2. The chain has five links, each requiring the previous one:
1. M is defined — a formula, a grain, a window, a source; one number per unit.
2. Units are enumerated — customers, products, branches, countries, failure modes.
3. Contribution is computed — the amount of M attributable to unit i. This requires mutual exclusivity: no unit’s contribution may be counted twice.
4. Units are ranked — sorted descending; the cumulative share is the Lorenz/Pareto curve.
5. The cut is chosen — the smallest set of units covering the share of M that matters.
Change the metric and the ranking changes, so the shortlist changes, so the recommendation changes. This is why Frame must precede Organize: prioritisation is not a judgement call bolted onto the tree, it is a computation whose input is the metric. Students verify this empirically in the Session 5 lab, and are required to report a measured concentration statistic (units-to-80%, top-decile share, Gini) rather than asserting an 80/20 split — following Juran’s own 1975 correction of the label he had popularised.
4.3 Communication: the Pyramid Principle as steps 6 and 7
Minto’s Pyramid Principle is not a writing add-on; it is steps 6 and 7 themselves. A finished argument is a governing thought containing a verb and a number, supported by a MECE key line of three to five reasons, each resting on evidence the team produced. Its four rules — ideas summarise the ideas below, ideas in a group are the same kind, ideas in a group are logically ordered, groups are MECE — are the same discipline applied in Session 4 to the issue tree, now applied to the argument.
5. Three TFG options, six routes, one method
Students declare their Final Degree Project (TFG) option and route in Session 3. All six routes are assessed against the same rubric spine, because all six run the same seven steps.
5.1 The six routes
Route | Audience whose scepticism must be survived | The M is typically | Evidence standard |
A1 · Academic — independent | An examining board with no stake in the topic | An operationalised dependent variable | Replicable design, effect sizes with uncertainty, stated limitations |
A2 · Academic — research group | The group’s supervisor and co-authors, then the board | The variable the group’s instrument already measures | The group’s own protocol, plus demonstrated fit with its published work |
B1 · New Venture — existing own venture | An investor who can read the student’s real numbers | The unit economic that is currently breaking | The venture’s own operating data, cut honestly, against a control or benchmark |
B2 · New Venture — family-business spin-off | The family board, who are also the incumbent | Incremental margin of the spin-off, net of cannibalisation | Incumbent data plus an arm’s-length test of genuinely new demand |
B3 · New Venture — business plan | A venture committee, with no data of the student’s own | The riskiest assumption, expressed as a number | Primary tests the student ran: landing pages, interviews, pricing tests, a pilot |
C · Management Case | A decision-maker inside one organisation | The KPI that person is judged on | Client systems plus open benchmarks; quantified impact on the KPI |
A note on the options. They are not three difficulty levels and they are not three subject areas. They are three audiences, and the audience sets the evidence standard. A student who cannot name their audience has not yet chosen a route.
5.2 What distinguishes the routes within each option
Academic Thesis — A1 versus A2
Both routes end at the same examining board and are marked against the same rubric. They differ in where the question comes from and how much scaffolding the student inherits.
A1 — Independent research. The student identifies the gap themselves, from the literature, and carries full responsibility for the theory, the instrument and the data access. It suits a student with a question they cannot stop thinking about and a realistic route to data. The characteristic failure is an under-specified gap discovered in March; the mitigation is that the Session 5 prioritisation must cite the papers that make the gap real.
A2 — Attached to a research group. The student joins an ongoing line of work in one of the University’s research groups. The question is scoped with the group, and the instrument, the protocol and often the data already exist. It suits a student who wants depth, a co-author network and a realistic prospect of publication. Two conditions apply:
1. A named contact inside the group must be secured before the advisor deadline. Places fill early; groups should be approached in Sessions 3–5, not in November.
2. The student’s own contribution must be separable and attributable in writing, agreed with the group in Session 3. The board will ask which part is the student’s, and a vague answer is treated as a failure of the contribution criterion, not as a presentational problem.
New Venture — B1, B2 and B3
The three routes differ in whose numbers the student can reach, and that determines what step 5 can actually be.
B1 — Existing own venture. The student already trades and has real revenue, churn and acquisition cost. The M is the unit economic that is currently breaking, and the evidence is the venture’s own operating data against a benchmark or a control period. The characteristic risk is advocacy: the student is both founder and analyst. The design must be capable of returning the answer “stop”, and the pre-registration in Session 3 is where that is committed to.
B2 — Family-business spin-off. A new line, product or market launched out of an incumbent family firm. The M is incremental margin net of cannibalisation — the number the family board will actually argue about — and the evidence combines incumbent data with an arm’s-length test of genuinely new demand. The characteristic failure is counting switchers as new customers; the switcher rate must be measured, not assumed.
B3 — Business plan. No venture yet and no data of the student’s own. The M is the riskiest assumption expressed as a number, and the student must generate primary evidence: landing pages, structured interviews, pricing tests, a pilot. A B3 built only from desk research does not pass, and the Session 7 memo must name which primary test is already running.
Data-access note. Routes B1 and B2 must state a data-access agreement in the Session 7 evidence memo; route C must do the same for client systems. Where the data is personal or commercially confidential, §10 applies in full.
5.3 The advisor
Every student must have a formal advisor — internal faculty or an approved external professional — formally in place by the first week of December. A student without a filed advisor agreement by that date may not defend in this cycle.
An internal advisor is a member of faculty. An external advisor is an approved professional supervisor from the host organisation, the venture or the research partner, and must be countersigned by an internal academic who remains accountable for academic standards. Appendix I is the agreement form.
When | Milestone | Evidence |
Sessions 1–3 | Shortlist three plausible advisors whose work touches the question | Three names and one line each on what they bring |
Session 3 | Declare TFG option and route; name the first advisor to be approached | Course declaration form |
Sessions 3–5 | Approach, with a one-page brief attached: draft problem statement, intended M, and what is being asked of them | Sent email with the brief attached |
Sessions 5–8 | Agree scope: meeting cadence, what the advisor will read, and the boundary of the student’s own contribution. In writing for A2 and for all external advisors | Written scope note |
First week of December | Formal advisor agreement filed, with internal countersignature for external advisors and the data-access agreement where the route requires one | Appendix I, signed |
December onwards | At least one substantive meeting per month | A dated note per meeting, sent to the advisor within 24 hours |
Students are told plainly in Session 1, and reminded in Sessions 3, 5 and 8, that a friendly conversation in October is not an agreement. If an advisor stops responding, or promised data access does not materialise, the student must raise it with the programme coordinator within two weeks; waiting out of courtesy is the most common way a term is lost.
5.4 The same steps, different artefacts
Step | Academic Thesis (A1, A2) | New Venture (B1, B2, B3) | Management Case (C) |
1. Define | Research question and the gap it addresses; for A2, the increment on the group’s prior work | The riskiest assumption in the business model, or the unit economic that is breaking | The client’s key question, and whose decision it is |
1. Define (the M) | Operationalised dependent variable | CAC, LTV, conversion, payback; for B2, incremental margin net of cannibalisation | The KPI the decision-maker is judged on |
2. Disaggregate | Conceptual model and hypotheses H1…Hk | Assumption tree, or the unit-economic identity | Issue tree of the value drivers |
3. Prioritise | Which hypotheses the available data can actually test | Which assumption, if false, kills the venture | Pareto on the KPI: where the value actually sits |
4. Workplan | Chapter plan with data and dates | Test plan with cost per test | Analysis plan agreed with the sponsor |
5a. Gather | Secondary data, survey or panel, with provenance | Landing pages, interviews, pilots, own ledger, market data | Client systems plus open benchmarks |
5b. Analyse | Estimation, robustness, limitations | Experiment results and cohort economics | Driver analysis with quantified impact |
6. Synthesise | Discussion and stated contribution | Investor narrative and roadmap | Recommendation and implementation plan |
7. Communicate | Defence to an examining board | Defence to a sceptical investor | Defence to a decision-maker who must act |
Typical length | 40–80 pages | 30–50 pages plus a model | 30–50 pages plus exhibits |
The class sessions will involve a dynamic combination of theory and practice. Students will be required to read before coming to class and be prepared to discuss relevant issues in the group, as well as being able to apply their knowledge to practical examples. Throughout the course students will work in groups on an ongoing research project that they will design, execute and evaluate. Every theoretical step taken will be tested in practical terms through these projects. Students will be expected to develop these projects outside of class. The practical research project management dimension of the course is designed and executed by the students as managers of the project under the supervision of the professor. The 'classes' are redefined as 'meetings' and the students must learn to plan and organize these meetings effectively in order to achieve the objectives they establish. Students are guided by the professor in terms of theoretical contents, which is expected to be done in preparation for meetings and should influence the actions and its deliverables. The ultimate aim is to simulate as realistically as possible how a research team would work on a consultancy project.
Weekly sessions of 3 hours: concept block, workshop, and a 30-minute Python lab.
For further details, see the contents, section.
# | Component | Weight | Due | Assesses | Steps |
0 | Lab notebooks (13, Sessions 3–13) | 15% | Weekly from Session 3 | LO2, LO6–LO11, LO14 | 1, 5 |
1 | Problem statement, issue tree and workplan | 10% | Session 4 | LO1, LO2, LO4, LO5 | 1–4 |
2 | Evidence and design memo and Analytical core | 25% | Midterm week | ||
3 | Final pyramid deliverable and defence | 20% | Session 13 | LO6–LO11 | 5 |
4 | Participation Participation and Attitude | 10% | Final session | LO12, LO13 | 6–7 |
5 | Final Exam | 20% | Finals Week | all | all |
AI: Students who opt to use an AI in their work (if authorized by the professor) are required to include a paragraph at the end of the task or activity in which the student has used said tool and explain why you have decided to use it and the result obtained from it. The highest use level is L3. Failure to include this information clearly will be considered as an attempt at fraud or cheating as it deceived the evaluation system. In such case the penalties set out in Salle Campus Barcelona - Copies Regulation will be applied. https://www.salleurl.edu/en/copies-regulation Retake policy: Retake will be a final exam and the maximum retake grade for a student would be a 6.
Level of AI use: This following table describes the 5 levels with a synthesis of the guidelines
for the use of AI and the requirements it entails for the student.
Level | Meaning | Use of AI | Student Requirements |
1 | No AI | The use of AI is not allowed in any phase of the task. The work must be carried out entirely without artificial assistance. | The activity must be completed independently, demonstrating the student’s own knowledge and skills, without the use of AI tools. |
2 | AI for Idea Generation and Structure | AI may be used for brainstorming, outlining, initial guidance, or organizational suggestions, but not for producing the final content. | AI may be used only as preliminary support. The final submission must be entirely human-authored and must not include AI-generated content. |
3 | AI for editing | AI is allowed to improve writing, clarity, grammar, punctuation, or phrasing, but not to generate new content. | Students must submit original work and include the initial (pre-AI) version as evidence of the editing process. |
4 | AI to perform tasks, with human evaluation | The use of AI is authorized to complete specific parts of the task. The focus is on critical analysis and human evaluation of the generated content. | Students must use AI for the indicated parts, critically analyse and evaluate the outputs, and clearly reference any AI-generated content. |
5 | Full use of AI | AI may be used extensively throughout the process as a support tool or "co-pilot" to develop the work. | Students may freely integrate AI to support their work according to the task objectives. In this version of the scale, specifying which parts were generated with AI is not mandatory. |
0. Lab notebooks (15%). Thirteen short notebooks, one in each of Sessions 3 to 15, marked pass/fail on three criteria only: (a) the notebook runs top to bottom from a clean kernel with no manual steps; (b) the exercises marked your code here are attempted, and the final “apply it to your own project” exercise is written for the student’s own TFG route; (c) the interpretation paragraph is present and states what the output does and does not establish. Eleven of thirteen passes earn full marks, so two may be missed without penalty.
Each lab is published in two versions. Students receive the student notebook, in which the exercise cells are scaffolded stubs. A worked-solutions notebook — complete code plus a solution note for every exercise, explaining what the answer shows and what students typically get wrong — is released after the submission deadline for that lab, and is the instructor’s teaching copy during the session.
1. Problem statement and issue tree (10%). Two pages plus one paragraph. Page 1: the SMART statement; the M with its six-line specification; the annotated MECE tree. Page 2: the Pareto evidence (curve, units-to-80%, Gini), the pruning decision, and the workplan table. Plus one marked paragraph stating what the team is not analysing and why.
2. Evidence and design memo and Analytical core (25%). Data statements for every source; population–frame–sample analysis with the direction of bias at each gap; the missingness regime with its evidence; the pre-registered analysis plan (dated); the scale and sampling paragraph, including the pinned snapshot and the multiplicity policy. The two prioritised branches analysed to a defensible standard. Method chosen for the question; validation contract respected; effect sizes with uncertainty; the verb in every conclusion matched to the design; a subgroup and fairness table wherever a model allocates anything. Fully reproducible. AI-use appendix complete to date.
3. Final pyramid deliverable and defence (20%). A Minto-structured document in the format and length of the student’s TFG route (§5.4), plus a ten-minute defence and five minutes of questions. The container differs by route; the logical object — one governing thought, a MECE key line, evidence beneath each point — does not.
4. 15% Participation and Attitude (10%): Participating in the class discussions, active on Q&A, provide examples, active & engaged team member. Appropriate behaviour, avoiding unnecessary disruption and side conversations. Points will also be taken out for miss-use of phone, being disruptive, student been expelled from class, etc. Attendance - Pending assignments will only be accepted if your absence is justified by your tutor.
5. Final Exam: Measures Proficiency in the Methodology.
AI: Students who opt to use an AI in their work (if authorized by the professor) are required to include a paragraph at the end of the task or activity in which the student has used said tool and explain why you have decided to use it and the result obtained from it. The highest use level is L3. Failure to include this information clearly will be considered as an attempt at fraud or cheating as it deceived the evaluation system. In such case the penalties set out in Salle Campus Barcelona - Copies Regulation will be applied. https://www.salleurl.edu/en/copies-regulation Retake policy: Retake will be a final exam and the maximum retake grade for a student would be a 6.
Saunders, M et al. (2006) Research Methods for Business Students, 4th edition, FT Prentice Hall. Other readings will depend on the theme of the project.
Conn, C. & McLean, R. (2019). Bulletproof Problem Solving. Wiley. — the seven steps, with worked cases.
Friga, P. N. (2009). The McKinsey Engagement. McGraw-Hill. — TEAM FOCUS in full.
Conn, C. & McLean, R. (2019). Bulletproof Problem Solving. Wiley. — the seven steps, with worked cases.
Rasiel, E. M. (1999). The McKinsey Way. McGraw-Hill. — the origin of issue-based problem solving.
Minto, B. (2009). The Pyramid Principle (3rd ed.). Pearson. — read the first 60 pages now, the rest before Session 14.
Popper, K. (1959/2002). The Logic of Scientific Discovery. Routledge. — falsifiability, chapters 1 and 4.
Google Research (2025). Accelerating scientific breakthroughs with an AI co-scientist. research.google/blog — the multi-agent architecture discussed today.