The Baker's Dozen · Workshop 2 · session plan
For data owners and the leadership who use the numbers. One prep pass to set up the data room, then two sessions of two hours each. We build with Claude Code and Claude Cowork, so the company needs a Claude Max plan.
A Claude Max plan, since we'll be writing small custom apps for dashboards and querying. Read access to the data sources: Power BI, the GobbleCube exports, and the production and supply sheets. Get access granted a week before prep; requests have a way of sitting in inboxes.
No cohort. Shalin and your data owner pull everything into one place.
Ask the data direct questions, and build the guardrails that keep answers honest.
A real problem worked to a decision, a dashboard built live, and a test that keeps the system honest.
Before anyone books a room: just Shalin and your data owner, laptops open. We pull everything into one place and make sure it refreshes on its own.
See what exists
List every dashboard, export and sheet the company actually uses. Get read access working.
One place, refreshed daily
Pull it all together with a daily refresh. Boring plumbing on purpose; if it updates every morning without anyone thinking about it, it's working.
First questions
Ask it real questions and fix what breaks: wrong joins, missing fields, columns nobody can explain.
Stage the room
Load the examples and set up the accounts so Session 1 starts with a data room that already answers.
Today the numbers arrive as dashboards and reports from GobbleCube and Power BI. This session replaces waiting for reports with asking directly, and builds the guardrails that keep the answers honest.
By the end, everyone in the room has asked the data room a real question and checked the answer against their own instinct.
Recap of Cultivated AI Teach
The short version for anyone who missed Workshop 1: the loop of choose, communicate, evaluate, and why a person stays in it.
How this worked before Discuss
Someone asks why a city dipped last week. The question waits for whoever knows Power BI, becomes an Excel export, and comes back half-answered days later. AI makes the finding and assembling nearly free. Deciding what to do about it stays with you.
Your most-asked questions Hands-on
Everyone writes down the three questions they ask most, where the answer lives today, and how long it takes to get. This list drives the rest of the workshop and becomes the test set later.
One story from the field Teach
Swiggy built exactly this for their quick-commerce business. At first the answers were right about half the time. Once they taught the AI their metric definitions, it climbed past nine in ten. The lesson travels: the dictionary does the work, not the chatbot.
The metric dictionary Hands-on
Together we define the words the company uses every day: what counts as a sale, which date field is the real one, where returns go, what a stockout means. Ten definitions, argued until one version wins. Then the honesty rules: every answer shows the numbers it used, and when the data isn't there it says so instead of guessing.
What to ask, and what not to Teach
The questions this setup answers brilliantly, the ones it answers slowly, and the ones it shouldn't touch. We turn the most-asked list from earlier into the team's golden questions.
Free play Hands-on
Everyone drives. Compare this month with last across platforms. Chart returns by product. Find which items went out of stock in which city. Draft the Monday morning summary and see how close it lands.
Session 1 built trust in the answers. This one turns the data room into things the team uses every week without us.
The workshop ends with a dashboard the team built, a test that scores the system monthly, and a named owner.
Warm-up Hands-on
Back in the data room. What did you ask during the week, what worked, what surprised you.
One problem, worked to a decision Hands-on
Pick a real, current problem: a slipping city, a margin leak, a stockout pattern. Work it from first question to a decision the room agrees on, with the AI doing the digging and the people doing the judging.
Build a dashboard Hands-on
With Claude Code we build a small custom dashboard app around the golden questions, live. Change it on the spot when someone says "can it also show". This is the part that used to need a vendor and a quarter.
The answer key Hands-on
Automated evals, in plain terms: twenty questions with known answers, run against the system, scored. Rerun it monthly so trust is measured, not assumed.
The roadmap Commit
What gets added next, in what order, and who owns the data room after we leave. Small group, full access, one habit: cross-check any surprising number before acting on it.
Bonus: showing up in AI answers Teach
When someone asks ChatGPT for the best banana bread in Mumbai, does The Baker's Dozen come up? A quick look at how AI search picks its answers, and what nudges a brand into them.
Data access decides the schedule. It's the one thing that can delay the workshop, so we ask for it a week before prep.
This group is small and trusted, so everyone gets full access. Permissions can come later if the room grows.
Build, don't admire: every block after the recap has hands on keyboards.
Covers the data room prep and both sessions. The Claude Max plan is a subscription on your own account, billed separately, and it stays with you after the workshop.