Navigation: Part Index | Main Index | Pitching a Data-Science Idea -->

CRISP-DM Retrospective: From Coursework to Real Projects

Requires: CRISP-DM · Working as a Team on a DS Project

Motivation: Across this course, CRISP-DM was the guiding process. We considered various techniques and process phases. Now, what changes when the CRISP-DM process leaves the classroom and enters an organization with real stakeholders, budgets, and consequences?

Here we'll zoom out to the whole course arc, map every part back onto the CRISP-DM phases, and name what shifts when moving from personal/student projects to professional data-science work: more stakeholders, negotiated expectations, and genuine accountability.

The Course Arc in One Map: CRISP-DM Revisited

We introduced CRISP-DM in 🖝 CRISP-DM as a live orientation tool for the major parts of the course. Let's sum up the whole arc:

CRISP-DM phase	Where you practiced it
Business Understanding	🖝 Part I: The Big Picture, 🖝 Part II: Data-Science Projects
Data Understanding	🖝 Part III: Data Understanding
Data Preparation	🖝 Part IV: Data Preparation
Modeling	🖝 Part V: Supervised Learning, 🖝 Part VII: Unsupervised Learning, 🖝 Part VIII: Deep Learning
Evaluation	Parts V-VIII, and the principles in 🖝 Part VI: Principles That Transfer (Reflection)
Deployment	This part

You have traversed the "inner loop" (Data Understanding, Preparation, Modeling, Evaluation) many times, but Deployment was not much of a topic yet. However, most value of a solution can only unfold when it gets deployed and does actual work. Therefore, deployment is where a model or a system stops being something you own alone and becomes something other people rely on.

What Changes with Multiple Stakeholders: Roles, Expectations, Accountability

For small personal project, you set the goal and own everything from idea to solution. In this spirit, 🖝 Part II: Data-Science Projects was a deliberately light pass at Business Understanding. Real projects widen this:

More stakeholders, more interests. Beyond project-internal functions as we discussed in 🖝 Working as a Team on a DS Project, real projects depend on outside people: funding sponsors, end users who must adopt it, the people affected by its decisions, and functions like IT, compliance, and finance. Each has different interests, and some may resist the outcome.
Expectations must be negotiated. In 🖝 Academia vs. Business Data Science you saw that business success means value for someone, and that the success metric has to be agreed with stakeholders. A recurring obstacle is that stakeholders often carry misconceptions about what AI can and cannot do. (Nussbaum, 2023) contains some of the most common myths: that AI is always accurate, that it learns on its own, that it needs no maintenance. Surfacing and clarifying these upfront is usually helpful.
Accountability rises with the stakes. In coursework, a wrong model costs little. In production, it can cost money, safety, or someone's rights: recall the deployed-system asymmetry from 🖝 AI Ethics: A Primer. Someone needs to be accountable for what the system does. We return to this in 🖝 Ethics and Accountability in Production.

Discussion: When a deployed model produces a bad decision, who should be accountable, and how should that be decided before anything goes wrong?

Plan, Pitch, Perform: From Data Science Idea to Funded Project

This part draws from the "Plan, Pitch, Perform" report (Nussbaum, 2025). This is some of my own work which serves a framing on top of CRISP-DM, aiming at the organizational reality of getting a project funded and kept alive. It groups the procedure into three acts:

Act	What it does	Course connection
Plan	Understand the business context and estimate value cheaply, filtering weak ideas early	🖝 Part II: Data-Science Projects, plus a feasibility pass through Parts III-V
Pitch	Win support and resources with a business case	🖝 Pitching a Data-Science Idea
Perform	Execute, iterate, deploy, and maintain the funded project	surfaced in this part

The guiding principle behind the framework is: Keep investments low while uncertainty is high.

Especially during planning, cheap research and small experiments make sure that money and time are committed only once a project looks feasible and worthwhile. This is essentially 🖝 Reality-Checking Your Idea, scaled up from a "ten-minute check".

The pitch is covered in the next nugget 🖝 Pitching a Data-Science Idea. From there, we move into Perform: taking a working model out of a notebook and keeping it trustworthy over time.

Summary

The prior course parts trace the CRISP-DM phases from Business Understanding to Evaluation. In this part, we also consider deployment.
Aspects of real projects: more stakeholders with divergent interests, success criteria that must be negotiated, and accountability that scales with the stakes.
Non-technical stakeholders often hold misconceptions about AI. Aligning expectations early is part of the job.
The Plan-Pitch-Perform framing wraps CRISP-DM's early phases and keeps investment low while uncertainty is high, extending the reality-check habit from Part II.

As always: Happy learning, happy life! 🫶