01
What “data-driven” really means
It doesn't mean big data or fancy tools. It means one thing: when you make a decision, you can point to the evidence behind it.
“I think customers love the new plan” is a hunch. “Customers on the new plan cancel half as often” is evidence. Only one of them lets you bet the quarter on it.
Myth
“You need lots of data.”
A few hundred orders already tell a story. Small data, looked at weekly, beats big data nobody opens.
Myth
“You need a data team.”
Not anymore. An AI agent can do much of what an analyst did. See chapter 6.
Myth
“Data replaces your gut.”
It sharpens it. Your gut asks the question; the data checks the answer.
In practice it's a loop. Each lap makes the next decision less of a guess.
02
The four questions
Every question you'll ask your data is one of four kinds, and each builds on the last: you can't explain why something happened until you know what happened.
What happened?
“Revenue was $48k last month.”
The foundation. Start here.
Why did it happen?
“Revenue dipped because one-time orders fell after free shipping ended.”
Where an AI agent shines: it can try ten angles while you try one.
What will happen?
“At this pace we'll hit $60k a month by December.”
No machine learning needed. A steady trend, extended with care, goes a long way.
What should we do?
“Bring back free shipping, but only on first orders.”
The human part. Data frames the options; you make the call.
03
How business intelligence fits together
Business intelligence is a big name for a simple idea: turning your data into answers you act on. Every setup, from a one-person shop to a bank, has the same four layers.
- 01
Where data is born
Every tool you run the business with records something.
Your appPaymentsYour storeAdsSpreadsheetsIn Orcabase: Connect it or upload it → - 02
One home for all of it
A data warehouse keeps it together, so one question can cover everything.
Orders next to ad spendHistory keptIn Orcabase: Data Warehouse → - 03
What the numbers mean
“Revenue” written down once: paid orders, minus refunds.
RevenueActive customersChurnIn Orcabase: Metrics → - 04
What you see
Dashboards to glance at, and an agent to ask.
DashboardsQuestionsAI agentIn Orcabase: Dashboards and the Data Agent →
Why this matters
04
The startup metrics that matter
A metric is a number you track with a fixed recipe. You don't need fifty. You need the handful that match the question your company is answering right now.
Just launched
Do people want this?
Activation rate
New signups who reach the “aha” moment
= activated ÷ signups
Weekly active customers
People who did the key thing this week
= count of active customers
Early retention
Signups still active a month later
= still active ÷ started
Growing
Can I grow this profitably?
Monthly recurring revenue
What subscriptions bring in each month
= sum of monthly plan prices
Cost to win a customer
Marketing spend per new customer
= marketing spend ÷ new customers
Churn rate
Customers who leave each month
= customers lost ÷ customers at start
Average order value
What a typical order is worth
= revenue ÷ orders
Scaling
Is every customer worth more than they cost?
Lifetime value
What a customer is worth over their whole stay
= revenue per month × months they stay
Value vs. cost
Lifetime value against the cost to win them
= lifetime value ÷ cost to win
Gross margin
What's left of each dollar after direct costs
= (revenue − direct costs) ÷ revenue
Rule of thumb
05
Vanity numbers vs. useful ones
Some numbers only ever go up, so they feel great and tell you nothing. Useful numbers can go down, and when they do, you know what to fix.
The Monday test
06
Where AI agents come in
Turning a business question into an answer used to need an analyst. Now an AI agent can do it: it looks at your data, digs, checks what it found, and explains it, showing every step.
Put an agent inside the place your data lives and you get an agentic data warehouse: a data team for a company of one.
What agents are great at
- Finding things: “who hasn't ordered in 60 days?”
- Digging into why a number moved
- Building dashboards and drafting definitions
- Explaining it all in plain words
What they need from you
- Your data in one place
- Clear definitions of your key numbers
- Good questions (that's you)
Remember
07
Your first month
You don't need a big project. Four focused weeks and one weekly habit.
Week 1
Connect
- Connect your app's data, or upload your key exports
- Ask the agent five questions you've always wondered about
Week 2
Define
- Pick five numbers from chapter 4
- Let the agent draft them, then certify the ones you agree with
Week 3
Build
- One dashboard, about six charts
- Your five numbers at the top
Week 4
Make it a habit
- A 30-minute review every Monday
- For each number: up, down, and why?
08
Words you'll hear
The terms in this guide, in one sentence each.
- Activation
- When a new customer first gets real value from your product: the “aha” moment.
- Agentic data warehouse
- A data warehouse with an AI agent inside that reads your data and answers questions about it.
- AI agent
- AI that takes steps to finish a task, like looking at your data and checking its answer, instead of replying from memory.
- Certified metric
- A metric you've checked and approved. The one to trust when two numbers disagree.
- Churn
- Customers who leave. Churn rate is the share who leave in a month.
- Dashboard
- One page with the handful of numbers you check regularly.
- Data warehouse
- One home for all your business data, set up for answering questions.
- Metric
- A number with a fixed recipe, like revenue = paid orders minus refunds.
- MCP
- A standard connector that lets AI assistants like Claude and ChatGPT use other apps.
- Retention
- The share of customers who stick around over time. The opposite of churn.
- Your own AI key
- Your account with an AI provider. You pay them directly for what you use.
- Token
- How AI providers measure text, about ¾ of a word. They charge per token.