Introduction to Marketing and Strategy
IME_51456_EP · Management course · Year 1 · Period 1 · mandatory
Contents
| Module | Introduction to Marketing and Strategy (IME_51456_EP); French title Introduction au marketing et à la stratégie |
| Period | M1 · Period 1 |
| Volume | 36 h (SynapseS 2026–2027 sheet) |
| ECTS | No value given for this unit: the curriculum wiki’s general rule is 4.5 ECTS per 36 h M1 course, but it gives its Period 2 management units 3 ECTS; the 2026–2027 sheet shows none |
| Status | Mandatory · management |
Stakes (Extrapolating)
Official unit for 2026–2027: IME_51456_EP — Introduction au marketing et à la stratégie (Philippe Ginier-Gillet, 36 h, Period 1, listed in LLGA 1 – S1). Its sheet starts from a fact: most innovative products and most start-ups fail, often because their founders misread what users and partners need before adopting something new, or misread the dynamics of their industry; the course teaches students to think like a strategist and a marketer rather than to recite the 4P or Porter’s five forces, through business case studies.
The curriculum wiki lists the same unit in English as Introduction to Marketing and Strategy, the title of this page; the home tree shortens it to “Strategy / Marketing”.
The rest of this page is our own framing. In period 1, the technical core teaches how to model. This management block asks why ship: positioning, innovation, and market reading. Empirically, a Graph×LLM pipeline only “counts” in industry if it sits inside an offer, a segment, and go-to-market constraints — otherwise it stays a demo.
In this course you learn to tie a technical capability (graph RAG, fraud GNN, …) to a value hypothesis and to metrics that go beyond accuracy.
- Separate competitive advantage, product differentiation, and a mere ML feature.
- Read a strategy grid (competition, barriers, adoption) without empty jargon.
- State a value proposition for a structured-AI case (graphs, multimodal, agents).
Place in the 2-year path
- M1 · P1: alongside the scientific core.
- M1 · P2: the 2026–2027 management units are IME_52441_EP Simulation d’entreprise (a one-day business serious game) and IME_52470_EP Digital Transformation and AI in Organisations and Society; this site does not cover them. They replace the wiki’s choice between Entrepreneurship for Sustainability and Case studies on Innovation; our Startup & entrepreneurship notes on that slot are kept for reference.
- M1 · P3: the research internship tests a scientific question and its relevance.
- M2: the ethics seminar and end-of-studies internship close the loop — strategy → evidence → responsibility → deployment.
Industry and research links
Industry: AI product managers, tech strategy consulting, platform teams deciding build vs buy. Research / transfer: situating a paper (GNN, LLM, OT) in an application landscape and adoption bottlenecks — useful for a research internship or CIFRE thesis.
How to approach it
- Keep one concrete track case and force every strategy/marketing concept to answer it. Two such cases appear in the end-of-year projects: the fraud-detection GNN of Y2.5 and a GraphRAG offer as in Y1.1 and Y2.1; recommendation works too.
- Prefer falsifiable hypotheses (segment, cost of error, latency, data moat).
- Later connect to Ethical Issues and Novel Applications of AI. Our Startup & entrepreneurship notes are kept for reference only: that unit is not in the 2026–2027 list.
Metrics beyond accuracy
Next to accuracy, a deployed model is judged on three numbers: the cost of its errors in euros, its latency, and its adoption. This is the hook the projects reuse for objective O7 (hands-on and industry practice).
Cost of error. A missed fraud and a false alert each have a price, c_{\mathrm{FN}} and c_{\mathrm{FP}}. Take a calibrated probability p of fraud (Machine Learning ch. 4). An alert is worth raising when the expected loss it avoids, p\,c_{\mathrm{FN}}, exceeds the expected cost it creates, (1-p)\,c_{\mathrm{FP}}, that is when
p > \frac{c_{\mathrm{FP}}}{c_{\mathrm{FP}} + c_{\mathrm{FN}}}.
With a missed fraud at 500 € and a false alert at 5 € of analyst time, the threshold is
\frac{5}{5 + 500} = \frac{5}{505} \approx 0.0099,
about 1 %, far below the default 0.5. A confusion matrix becomes a business case once each of its cells has a price.
Latency. The delay the process tolerates: a card payment waits a fraction of a second, a compliance report waits a day. It decides between a large model, a small one and a cached answer.
Adoption. Whether people act on the output: the share of alerts an analyst actually opens, or of answers a user accepts. A model nobody uses is worth nothing, whatever its accuracy.
The fraud capstone Y2.5 asks for a cost-in-euros report; every other capstone may add these three numbers as an optional paragraph (deliverables).
Extrapolating Framing notes for this repo; detailed content remains that of the official semester teaching.