Quick start: build a dashboard

Build a dashboard that shows me how the business performed last month.

The reader is the CEO and has ninety seconds before a leadership meeting.
They want to know whether last month was good, and what to ask about.

Across the top, four headline numbers, each with the change versus the same
month last year and versus plan if plan data exists:
- revenue
- gross margin percent
- operating expenses
- cash collected

Below that:
- a rolling thirteen-month revenue trend with a margin line on a second axis
- revenue split by <product line / division / region> for the month, ranked,
  showing which moved most against last year
- the five largest expense categories with the change versus last year

Rules:
- our fiscal year starts in <month>
- exclude <intercompany transfers / any entity you do not consolidate>
- treat revenue as recognized revenue, not bookings
- color anything more than 5% below last year in red, and more than 5%
  above in green

When it is built, show me the result and then list every assumption you had
to make about our data: which fields you mapped to each concept, and
anything ambiguous you resolved by guessing.

Build a dashboard that shows me whether we are going to hit this quarter.

The reader runs sales and looks at this every Monday morning. They need to
know the gap to target and which specific deals close it.

Show:
- closed revenue this quarter to date, the current target, and the gap
- weighted pipeline for the remainder of the quarter, using stage
  probability, with a simple projection of where we land
- a funnel by stage: count of deals, total value, and average age in stage
- the ten largest open deals closing this quarter, with owner, value, stage,
  days since the last activity, and close date
- deals that have slipped their close date more than once, flagged

Rules:
- our quarter is the calendar quarter
- exclude deals marked <lost / on hold / disqualified>
- treat anything with no activity in 21 days as stalled and mark it
- if we have a stage probability field, use it; otherwise tell me and use
  a straight stage-based estimate

When it is built, show me the result and list every assumption you made,
especially which field you used for deal value and which for close date.

Build a dashboard that shows me whether we are delivering on time and where
we are losing days.

The reader is the COO. They look at this before the operations standup and
need to walk in knowing which two things to raise.

Show:
- on-time delivery percentage for the last 90 days, as a trend, with our
  target of <your target>% marked as a line
- current open orders bucketed by promised date: overdue, due this week,
  due next week, later
- the ten most overdue open orders, with customer, promised date, days late,
  current stage, and value
- average days from order to ship, broken down by stage, so I can see where
  the time goes
- on-time percentage by <plant / product family / customer>, ranked worst
  first

Rules:
- measure on-time against the <original promise date / current promise date>
- exclude orders that were cancelled or that the customer asked us to hold
- count partial shipments as late until the order is complete

When it is built, show me the result and list every assumption, especially
which date fields you used for promised and actual ship.

Build a dashboard that shows me our workforce cost and where we are losing
people.

The reader is the CEO. They want to understand headcount cost trends and
spot the departments with a retention problem before it becomes a
recruiting problem.

Show:
- current headcount, total annualized labor cost, and average cost per head,
  each with the change versus twelve months ago
- headcount by department, with the twelve-month change
- rolling twelve-month voluntary turnover rate, as a trend and by department,
  ranked worst first
- average tenure by department, and the count of people in their first year
- open positions and average days open, if we have requisition data

Rules:
- separate voluntary from involuntary departures and label them clearly
- include employer-paid benefits and taxes in labor cost if that data exists,
  and tell me if it does not
- exclude <contractors / seasonal staff / interns>
- suppress any group with fewer than five people so no individual is
  identifiable

When it is built, show me the result and list every assumption, especially
how you classified departures and what is included in labor cost.

Do not include names, salaries of individuals, or any field that identifies
a single employee anywhere on this dashboard.

Review the charts on this dashboard as a data visualization specialist would.

For each one, tell me:
- whether the chart type suits the question it is answering
- whether it is readable at a glance on a laptop and on a phone
- anything misleading: a truncated axis, a dual axis implying correlation,
  a color scale that is hard to read, or a scale that exaggerates a change

Recommend changes, with the reason for each. Change nothing yet.

Create a metric dictionary for this workspace and keep it up to date as we
build.

Go through every dashboard we have built so far. For each distinct metric,
record:
- the metric name as it appears on screen
- the plain-English definition, written so a new controller could reproduce it
- the exact calculation
- the source system and table it comes from
- every filter and exclusion applied
- which date field it uses and on what basis
- the grain
- who owns the definition

Where the same metric is calculated differently on two dashboards, flag it
as a conflict rather than picking one.

Then show me the dictionary and read it at the start of every future build,
so you stop asking me the same questions.

Add controls to this dashboard so it serves more than one question.

- A date range picker at the top, defaulting to <the last complete month>,
  with quick options for this month, last month, this quarter, this year,
  and last twelve months. Everything on the page respects it
- Dropdown filters for <division / region / team>, with "All" as the default
- Clicking any bar or chart segment shows the underlying records in a table
  below, with a way to get back
- A toggle switching the comparison between prior month, prior year, and plan
- An export button that produces the current view as a spreadsheet, with the
  filters applied and stated in the file

Keep the default view exactly as it looks now, so someone who opens it and
touches nothing sees what they see today.

Remember my last filter selection when I come back.

Turn this dashboard into a scheduled email digest.

Send it <every Monday at 7am / on the first working day of each month> to
<recipients or role>.

The email contains:
- a subject line with the single most important number and its change,
  so it is readable from a phone notification
- the headline numbers with their changes, as text and simple visuals that
  render in an email client
- the three largest movements since the last send, each with one sentence
  of plain-English explanation, clearly labeled as commentary
- anything that crossed a threshold we care about
- a link to the full dashboard

Keep it under one screen. If nothing material changed, say so in one line
rather than padding it.

Show me a preview of what the next send would look like before you schedule
anything.

Add an export that turns this dashboard into a document for our board pack.

The document should:
- open with a one-page summary: the headline numbers, the change, and three
  bullets on what drove them
- follow with the detail sections, each chart on its own with a caption
  stating what it shows and the period covered
- state on every page the period covered, when the data was pulled, and
  which source systems it came from
- clearly separate calculated figures from AI-written commentary, so a
  reviewer always knows which is which
- carry a version and a date so two copies can never be confused

Generate it for <last month> so I can review the format, then tell me how to
produce it for any period.

Write a summary of this dashboard for <who and what they need it for>.

Five paragraphs, plain English, no jargon:
1. What happened in the period, in one sentence, then the supporting numbers
2. What drove the largest change, with the evidence
3. What is trending in a direction worth watching
4. What the numbers cannot tell us, and what we would need to answer it
5. The two or three things worth a decision

Cite the figure behind every claim so anyone can check it. Where you are
inferring a cause rather than reading it from the data, say so in the
sentence rather than in a footnote.

Set up alerts on this dashboard.

Alert me when:
- <metric> goes <above / below> <value>
- <metric> moves in the same direction for <3> consecutive <weeks>
- any source feeding this dashboard has not refreshed successfully in
  <24 hours>

Send alerts to <who and how>. Each alert states what happened, the current
value, the previous value, and a link to the relevant view.

Rules that keep alerts worth reading:
- alert once when a condition starts, and once when it clears, rather than
  every time you check
- hold off re-alerting on the same condition for <7 days> after I acknowledge it
- group everything that fires in the same hour into one message

Show me what would have fired over the last ninety days, so I can tell
whether the thresholds are set sensibly before we turn them on.

Given everything connected to this box, propose the eight numbers I should
look at every Monday morning as <your role> of a <your industry> company.

For each one, tell me:
- the number and how you would calculate it from what we actually have
- the decision it would change
- how often it meaningfully moves, so I know whether weekly is the right cadence
- what a healthy value looks like for a company our size, and say so plainly
  when you are estimating rather than reading it from our data

Rank them by how much they would change my decisions, and tell me which two
you would drop if I only wanted six.

No code yet. I want to agree the list first.

This dashboard is ready to share with the leadership team.

Give it a clear title and a one-line description of what question it answers.
Add a footer showing when the data last refreshed and which systems it came
from. Make it readable on a phone.

Then tell me the link to open it, who can currently see it, and what I need
to do to give <name or role> access.

The <metric name> figure for <period> shows <value>. I expected roughly
<your expected value>.

Walk me through it:
1. Show me the exact query behind that number
2. Show me the underlying records it summed, so I can spot anything unexpected
3. Tell me which filters and exclusions are applied
4. Tell me the freshness of the source data
5. Give me your best three explanations for the gap, ranked by likelihood

Change nothing yet. I want to understand it before we adjust anything.