BI Anatomy

Authentication, semantic layer, dimensions/measures, API integration.

What is BI made of?

The old nursery rhyme asks what little boys and girls are made of. This lesson asks the same about a BI platform. Take any enterprise BI tool apart (Tableau, Power BI, SAP BusinessObjects) and you find the same building blocks:

Component	What it does
Authentication and authorization	Who can log in, and what they can see
Client (Tableau Desktop)	Where developers and analysts build reports
Server (Tableau Server)	Where reports are published, shared and refreshed
Semantic layer	Business-friendly model on top of raw tables
Visualization templates and tables	Ready-made chart types
Dimensions, metrics, calculated fields, filters, hierarchies, parameters	The fields and controls every report is built from
Mailing (email, shared folders)	Delivering reports to people
Schedule	Refreshing data and sending reports on a timer
Alert	Notifying users when a metric crosses a threshold
Database, service and file connections	Access to the data sources
API	Automating the platform from code
Statistical methods	Forecasts, trend lines, clusters
Python/R integration	Running custom models inside a report
Server and client logs	Troubleshooting and monitoring

Tools differ in names and UI, but the parts are the same. Learn them once in one tool and you’ll find them in any other.

BI client

The client is the desktop or web app where you build reports and dashboards. In Tableau it’s Tableau Desktop: you connect to data, drag fields onto a sheet, and combine sheets into a dashboard. A typical example is the Superstore “Executive Overview – Profitability” dashboard: KPI tiles (sales, profit, profit ratio), a profit map by state, and monthly sales by segment and category.

The client is where BI developers and analysts spend most of their time. Business users usually never open it. They consume the finished dashboard on the server.

Authentication

Authentication answers “who are you?”. Enterprise BI servers plug into the company’s existing identity systems:

• Active Directory
• LDAP
• Kerberos
• OpenID Connect
• SAML (single sign-on)
• Trusted tickets (for embedding dashboards in other apps)
• Local accounts

In a company, you almost never create BI users by hand. They come from Active Directory or a single sign-on provider, so an employee who leaves loses access everywhere at once.

Authorization

Authorization answers “what are you allowed to do?”. It works on two levels:

• Content: who can view, explore, download, publish or delete projects, workbooks and data sources. In Tableau Server this is set per project with permission templates (View, Explore, Publish, Administer) for users and groups.
• Data (row-level security): two users open the same dashboard and see different rows. For example, a regional manager sees only their region.

Give permissions to groups, not individual users, and lock them at the project level. Per-user exceptions quickly turn into a mess nobody can audit.

BI server

The server hosts published content and does the heavy lifting: rendering views, refreshing extracts, sending subscriptions.

	SAP BusinessObjects	Tableau Server
Web layer	Web/app server (IIS or Tomcat with BO Java components)	Gateway
Core services	CMS, job and processing servers	User tier: VizPortal, VizQL Server, Data Server, Ask Data, Metadata Services
Background work	Job servers	Management tier: Data Engine, Backgrounder, Prep Conductor, Elasticsearch/Cache Server/SOLR, Cluster Controller
Storage	FileStore folders, system repository database	Storage tier: Repository, File Store

You don’t need to memorize these services as an analyst. But when a dashboard is slow or an extract refresh fails, knowing that the Backgrounder runs refreshes and VizQL renders views tells you where to look.

Connection to database or service

A BI tool connects to two kinds of sources:

• Files: Excel, text/CSV, JSON, PDF, spatial and statistical files
• Servers: Tableau Server, Microsoft SQL Server, MySQL, Oracle, Amazon Redshift and dozens more

For a database you enter the server, port and database name, then sign in with a username and password. Require SSL encrypts the connection, and Initial SQL runs a statement every time the connection opens.

Use a dedicated read-only service account for BI connections, not your personal login. Reports keep working when you change your password or leave the company.

Visualization templates and tables

Every BI tool ships with a gallery of chart types: Tableau’s Show Me panel and Power BI’s Visualizations pane. Both include bar, line and area charts, maps, pie charts, scatter plots, treemaps, tables and more. Power BI can even run R and Python visuals.

Most business questions are answered with a bar chart, a line chart and a table. Reach for exotic chart types only when they make the answer clearer.

Semantic layer

The semantic layer turns raw tables into a model business users understand: tables are joined, fields are renamed, and relationships are defined once.

• In Tableau you join tables in the data source (inner, left, right, full outer).
• In Power BI you build a data model with relationships between fact and dimension tables.

Every report built on a good semantic layer uses the same joins and the same definitions. Without one, each analyst re-joins the tables their own way, and the numbers stop matching.

Dimensions, metrics, calculated fields, filters, hierarchies, parameters

Concept	What it is	Example
Dimension	Something you group or slice by	Customer name, segment, order date, ship mode, country
Metric (measure)	A number you aggregate	Sales, profit, discount, quantity
Calculated field	A new field defined with a formula	DATEDIFF('day', [Order Date], [Ship Date]) = days to ship
Hierarchy	Dimensions stacked for drill-down	Country → Region → State → City → Postal code
Parameter	A user-controlled input value	A “New quota” of 500,000 for a what-if analysis
Filter	Limits which data is shown	Region: Central, East, South, West

These six concepts exist in every BI tool under slightly different names. Once you understand them, switching from Tableau to Power BI or Looker is mostly learning a new UI.

Mailing (email, shared folders)

BI servers deliver reports to people who don’t open dashboards themselves: by email (with the image in the body and PDF or CSV attachments) or by dropping files into a shared folder.

A well-timed email with the three numbers a manager cares about often gets more use than a beautiful dashboard nobody opens.

Schedule

Schedules run tasks automatically: extract refreshes (reload data into the BI tool) and subscriptions (email a report). Each schedule has a frequency (every 15 minutes, hourly, daily, weekly), an execution mode and a next run time.

Schedule refreshes after your data warehouse finishes loading, not before. Otherwise the dashboard shows yesterday’s data all day.

Alert

Alerts send an email when a metric crosses a threshold, for example “Sales is above or equal to 170,000”. You choose the condition, threshold, subject, how often to send and the recipients.

Alerts flip BI from “go and check the dashboard” to “the dashboard tells you when something happens”. Use them for the few metrics that really need action.

API

BI servers can be automated from code. Tableau has a REST API and the tabcmd command-line tool:

tabcmd login -s https://tableau.example.com -u username -p <password>
tabcmd createproject -n "Work Football Team" -d "Tracks the goals scored by players in the work football team, Tableau FC"

Anything you do by clicking in the admin UI (creating projects, publishing workbooks, refreshing extracts, managing users) can be scripted. Keep credentials in a secrets store, never in a script.

Statistical methods and data science

BI tools include built-in analytics: average and median with 95% confidence intervals, trend lines, forecasts and clusters. For example, Tableau can forecast 15 months of monthly sales from the history, detect a 12-month seasonal pattern, and show the confidence band around the prediction, with a quality rating for each segment.

Built-in forecasts are good for a quick look. For anything the business will plan budgets on, build and validate the model properly, then show the results in BI.

Python/R integration

When built-in functions aren’t enough, BI tools can call Python or R. In Tableau a calculated field sends data to a Python or R service and gets results back. This example scores the sentiment of customer comments with the VADER library:

SCRIPT_REAL("
from vaderSentiment.vaderSentiment import SentimentIntensityAnalyzer
vs = []
analyzer = SentimentIntensityAnalyzer()
for i in range(0, len(_arg1)):
    a = analyzer.polarity_scores(_arg1[i])['compound']
    vs.append(a)
return vs
", ATTR([Comment Text]))

Each comment gets a score, and the table is colored from green (positive, e.g. “This is an awesome product”) to red (negative, e.g. “Customer service was unresponsive”).

Server and client logs

Both the server and the client write detailed logs. On Tableau Server, the logs of each service (the Backgrounder, Tomcat and others) show every scheduled job, refresh and error with a timestamp.

When an extract refresh fails or a dashboard is slow, the logs are the first place to look. Admins also analyze them to find unused dashboards and heavy queries.