SE 395-2: AI Tools for Social Impact
Workflow Mapping and Solution Design. Map your partner's real workflows. Write the recipe for how AI fits in. Find the gaps before you build.
"Every big change starts as one small, deliberate action."
First RippleStudent Access
Token provided in class or on Canvas.
Professor Meirelles van Vliet
Fall 2026 · T/R 3:30-4:45 PM · KWR 310
SE 395-2: AI Tools for Social Impact
Week 6 · Thursday · AI Lab
Map your partner's real workflows. Write the recipe for how AI fits in. Find the gaps before you build.
Labs 1-3 built your research and collaboration skills. This lab turns that work into something actionable. Before you write your project proposal, you need to understand your partner's workflows well enough to explain them to someone who has never seen the organization.
This lab is based on a simple idea: if you cannot describe a workflow clearly enough for a stranger to follow it, you cannot design an AI solution for it. You will map a real workflow, write a recipe for how AI could fit into it, test whether your description actually works, and identify what should never be handed to AI.
What you produce here feeds directly into your project proposal, due the same day.
Work through these in order. Steps 1-4 happen in class. Steps 5-6 are homework.
In class, 10 minutes.
As a team, choose ONE recurring workflow from your partner organization. Not the whole operation. One specific, repeatable task that has inputs, steps, and an output that goes to a real person.
The right size: "check my email" is too small. "Run the whole quarter" is too big. You want the middle: something that recurs, has a few inputs, involves some judgment, and produces something a person receives.
Sizing Template
Too small: Posting one social media update. Answering one email. Filing one document.
Right size: The monthly donor report. The weekly volunteer schedule. The client onboarding process. The grant application review cycle.
Too big: "Everything they do with data." "Their entire communications strategy." "All of their operations."
In class, 15 minutes.
Before you imagine what AI could do, write down how your partner actually does this workflow today. No AI in the picture. Just the current state.
Each team member writes independently for 5 minutes. Then compare. You will be surprised at how differently each person understood the same workflow.
Four Questions (borrowed from Jake Van Clief's framework)
In class, 10 minutes.
Hand your team's workflow description to another team. They have 3 minutes to read it and try to explain it back to you. No clarifications allowed during those 3 minutes.
Where they get confused is where your understanding is thin. Where they misinterpret is where your writing is ambiguous. Both are problems you need to fix before building anything.
After the swap, take 5 minutes as a team to revise your workflow description based on what the other team got wrong.
In class, 20 minutes.
Now write the recipe. Same workflow, but as if AI were already part of it tomorrow morning. Use the recipe format below.
Recipe Format
Ingredients (what goes in): List every input with real quantities. Not "emails" but "about 15 donor emails per week." Not "data" but "a spreadsheet with 200 rows of volunteer availability."
Method (steps in order): What the AI does. What the human does. In order. Be specific about handoffs.
Time and Serves: How long does the current version take? How long would the AI-assisted version take? Who receives the final output?
Chef's Note: The one substitution you must never make. What should AI never do in this workflow, and what breaks if it does?
The chef's note is the most important part. It forces you to name what is non-negotiable about human judgment in this workflow. If you cannot name it, you do not understand the workflow well enough.
Homework, 30 minutes.
Take your recipe and your cold reality description and have Claude (or ChatGPT) interview you about the gaps. Use the prompt below. This is not asking AI to generate a solution. It is asking AI to ask you questions you have not thought of.
Claude Interview Prompt
I am going to describe a workflow at organization ______ and a recipe for how AI could fit into it. Interview me so that we end with a clear picture of what works, what is missing, and what could go wrong. Ask one question at a time and wait for my answer.
Start with my description of how they do it today. Then ask about my recipe for the AI-enhanced version. For each part of the recipe, ask: where does the information come from, what decisions require human judgment, what comes out and who receives it, and where it breaks.
Push on the chef's note. Ask me to defend why that part should stay human. If my answer is weak, say so.
Do not suggest tools or fixes. After eight to ten questions, write back a document with these headings: What they do now. Where AI fits. Where AI does not fit. The biggest risk. Questions I could not answer. Keep my words.
Homework, 30 minutes.
Using everything from Steps 1-5, draft the core of your project proposal. This is not the full proposal. It is the foundation.
Write 2-3 pages covering:
| Strong (full credit) | Weak (minimal credit) |
|---|---|
| Your workflow is specific, recurring, and the right grain size | Your workflow is vague ("improve their communications") or too broad to act on |
| The cold reality reflects what the partner actually does, with gaps flagged honestly | You describe what you think they should do, not what they actually do |
| The swap test produced revisions and you can name what the other team got wrong | You skipped the swap or did not revise based on feedback |
| Your recipe has real quantities, clear handoffs, and a defended chef's note | Your recipe is generic ("AI processes the data") with no specifics |
| The Claude interview surfaced questions you could not answer, and you documented them | You used AI to generate answers instead of to generate questions |
| Your proposal seed connects directly to the workflow analysis | Your proposal seed is disconnected from the lab work |
Due: Thursday, October 8 by 11:59 PM. Upload team deliverable and individual deliverable to Canvas.
Length: Team report 6-8 pages. Individual AI conversation log 10-15 pages.
Exemplary Submission
This is what a strong Lab 4 submission looks like. It includes the team workflow analysis (cold reality, swap test revisions, recipe) and the individual proposal seed. Use it as a reference for structure, depth, and the level of specificity expected.
Exemplary Submission
This shows what a strong AI conversation log looks like for Lab 4. It covers the full process: researching the workflow, drafting the cold reality, building the recipe, running the Claude interview, and formatting the final report. Notice how the student uses AI as a thinking partner throughout, not just for one step.