HomeAboutProjectsHow I workContact

1Problem+2Data+3Storytelling

From problem to decision.Clear data, well told.

I’m Diego Cayupil, a data analyst. I turn data into clear answers so teams can make better decisions.

Open to senior data analyst & BI roles
  • Santiago, Chile
  • Open to remote work
  • English and Spanish
  • SQL
  • BigQuery
  • Power BI
  • Python
  • Excel
  • Tableau
  • Snowflake
  • dbt
  • Git
  • Claude
OperationsWeek 42Example figures
1

Why did on-time delivery fall?

OTIF: orders delivered on time and in full · 3 warehouses

2
OTIF92.4%▼ 1.1 ptsvs 93.5% over the prior 4 weeks
Weekly OTIF · weeks 31–42929394target 94%normal range92.43134374042
The drop, by warehouse−1.1 pts
North · 70%
North −0.77Central −0.22South −0.11
3

The North warehouse explains 70% of the drop.

1Check its stock-outs this week

02What I solve

Where does it hurt?That’s where I start.

Four pains. Pick yours.

✗Today
Duplicates40+No order87+Impossible date12

Errors nobody sees.

What I do

5 questions to the data before the first chart

primary key · orphan rows · ranges · reconciliation

I test the data before charting it.

✓After
Errors that reach the dashboard0✓ source = model: 48,000 orders

Every number, with its proof.

✗Today

14 charts, no conclusion.

What I do
14→3What decision is behind it?

I keep only what answers the decision.

✓After
The North warehouse explains 70% of the drop.→ Check it today

The conclusion on top, then the next step.

✗Today
LMXJV3 days copying and pasting

The report arrives old.

What I do
Sources→BigQuery→Alert if it fails

I connect the sources and automate the load.

✓After
8:00✓ Ready on Monday, by itself

Those 3 days now go to analysis.

✗Today
How much did we sell in October?412398405

Same question, three answers.

What I do

Net sales = invoiced − credit notes

With a description, for people and AI

I define each metric once and test the AI.

✓After
How much did we sell in October?October net sales: 412. That’s the official measure ✓

The AI gives the dashboard’s number.

03How I work

It all starts with defining.The rest is the result.

The same six phases make AI answer right.

01 · Problem framing

“I need a dashboard”is not a problem.

What comes in

“I need a delivery dashboard, with everything.”

What goes out

“Which warehouse do I fix first?”

02 · Data quality

I don’t build on dataI haven’t tested.

What comes in

“The data is fine, trust me.”

What goes out

2 definitionsof “on time”: 94.1% and 92.4%. The customer’s wins.

03 · Data modelling

One number,one single version.

What comes in

94.1 · 96.0 · 92.4Three figures for the same OTIF

What goes out

92.4%One measure. Everyone reads the same

04 · Insights

I don’t show you data.I tell you what to do.

What comes in

“OTIF fell 1.1 points.”

What goes out

“The North warehouse explains 70%. Check it first.”

05 · Storytelling + UX

If it needs explaining,it isn’t ready yet.

What comes in

8 pages, 60 visuals, no conclusion.

What goes out

One page. The title already states the conclusion.

06 · Documentation

If it isn’t written down,it doesn’t exist.

What comes in

A finished dashboard only I understand.

What goes out

Every metric with its card, in Git and in the model.

04How I use AI

I use AI every day.I never trust it blind.

This is how I build every Power BI project.

Define

1Me
Which warehouse do I fix first?

The business question

Builds

2AI
OTIF % = DIVIDE(
  [OTIF orders],
  [Orders] )

The SQL in BigQuery and the DAX measures

Check

3Me
Source48,000=Model48,000✓

Every number against the source

Decide

4Me
The North warehouse explains 70%

What the dashboard says

if it doesn’t add up, it gets redone
WithClaude CodePower BI MCP serverBigQuery

✕ Never write access · personal data in a chat · a number without its source

05Projects

One complete case.Nine on the way.

OperationsSimulated data

Fill rate to the chains fell from 95% to 88%. Where is it lost?

From the first meeting to the plan: seven steps, with the live dashboard.

See the case→
RetailThe Cyber promotions: did they make money, or just move sales from one week to another?Coming soon
MarketplaceThey buy once and never come back. Which categories bring customers who do return?Coming soon
1 complete case·9 on the waySee projects →
06Contact

What problemdo you want to solve?

Tell me in your own words, even if you don’t know what data you have. I reply within one business day.

To start, this is all I need
01What hurtsIn your own words, no jargon.
02Who it hurtsWho suffers it and who decides.
03What you’d decideIf you had the answer tomorrow.
diego.cayupil2@gmail.comLinkedIn ↗It’s 10:42 · UTC−3