All videos
Antvia Intelligence · Episode 1 of 8

Intelligence in one tour

A tour of Antvia Intelligence for a business or analytics leader. It starts on the home page, visits the command center, opens a dashboard that marks its own untrustworthy tiles, asks Echo questions it answers and one it refuses, shows the semantic models and alert rules underneath, and reads the decision register and its measured results. It ends by repairing the broken feed in Antvia Data and asking Echo the refused question again.

7:488 chapters888 wordsSilent, with captions

Also on YouTube, with the rest of the Antvia Intelligence series. The whole episode is written out below, with the timing, if you would rather read it.

Chapters

What happens, in order.

0:00
Introduction
Title card.
0:07
What the business runs on
The Intelligence home page: the month-to-date figures and which ones are not current, what the dashboards read, what was decided, and the ranked list of what needs attention.
1:23
The command center
A short visit to the command center: revenue against the plan due, the ranked items that need the business head, and the replay.
2:13
Dashboards that admit doubt
The Lens register with readership and broken feeds, the Business pulse dashboard, and tiles that name the table behind their warning.
3:20
Answers you can check
An Echo conversation: what it understood, the tables it read and their health, the SQL, a result table and chart, a reasoning answer, and a question it refuses because its table is stale.
4:31
Underneath and around
The semantic models with a measure defined three ways, the alert rules firing now, the decision register and its admission gate, and the measured results in Leap.
6:08
Fix the feed, ask again
Repairing the channel stock feed in Antvia Data, then asking Echo the same question again and getting a checked answer, with the dashboard current.
7:25
Close
Closing card.
Transcript

The whole episode, in words.

The videos are silent, with the captions burned in. This is the same text, with the timing, so you can find the part you need without scrubbing for it.

Introduction

0:00
Antvia Intelligence. Intelligence in one tour. Dashboards, answers and decisions that say what they are built on, and what happens when that changes.

What the business runs on

0:07
What the business runs on. The home page for the people who read the numbers.
0:12
Antvia IntelligenceAntvia Intelligence turns governed data into dashboards, answers and decisions. This is the home page at Acme Home Appliances, a demo company.
0:22
August to dateRevenue is ₹20.3 crore, 96% of the plan due by now. Gross margin is 32.4%, under its 34% plan.
0:31
Not currentTwo figures carry a warning. Channel stock is nine days old, and customer satisfaction is unverified. The numbers stay; the doubt is on the tile.
0:42
What it readsEight semantic models and eighteen dashboards. Two model joins multiply numbers, and eight dashboards are fed by a table with a problem.
0:52
What we decidedThe decision register: two proposals in review, three decisions being measured, and ₹4 crore of value attributed after costs.
1:02
Needs somebodyNeeds somebody ranks what is wrong underneath the numbers: joins that inflate totals, one measure defined three ways, and alerts that are firing.
1:15
Most readOn the right, the most-read dashboards. The top two are marked stale, because they read tables with problems.

The command center

1:23
The command center. The whole business on one page, for whoever owns the P&L.
1:30
Command centerThe command center is one page for the head of the business: sales, margin, plants, supply, marketing, launches, service and people.
1:42
Same numbersIt reads the same company. August to date is ₹20.3 crore against ₹21.1 crore due, and West is well short of its line.
1:54
Ranked by rupeesWhat needs the business head is ranked by rupee impact, each with an owner and the sources it was read from.
2:04
ReplayAnd a month of history can be replayed a day at a time. Episode two covers it in full.

Dashboards that admit doubt

2:13
Dashboards that admit doubt. Lens, and a dashboard opened.
2:21
Who reads themLens lists every published dashboard with how many people opened it in thirty days. One has had no viewers at all.
2:30
ReadershipReadership is the column the page is built around. A dashboard nobody opens is still a maintenance cost somebody is paying.
2:39
What feeds themEight are fed by a table the catalog does not stand behind.
2:46
Two unhealthyBusiness pulse, the most read, reads two unhealthy tables: channel stock, nine days old, and customer feedback, failing three checks.
2:57
Said up frontBusiness pulse says above its tiles that two of them are not showing current data.
3:04
Named causeEach flagged tile names its table and why: fact_channel_stock, last refreshed nine days ago.
3:12
The bridgeBelow, the charts: what moved EBITDA against plan this month, with material cost taking the most.

Answers you can check

3:20
Answers you can check. Echo restates the question before it answers it.
3:27
Understood firstEvery Echo answer starts with what it understood, in real tables, columns and time windows, before any number.
3:36
Read fromThen the tables it read, with their health. Billing and the distributor master are current.
3:44
The resultThen the result: areas that grew revenue ten percent or more last quarter, Coimbatore first.
3:53
A chartAsked for a line, Echo draws one: Coimbatore's monthly revenue for twelve months, with the step in January annotated.
4:02
WeighedAsked why January jumped, it weighs three explanations and says which one to verify before anyone believes it.
4:20
RefusedAsked about channel stock, Echo declines. The table it would read is nine days old, and an answer from it would look current and would not be.

Underneath and around

4:31
Underneath and around. Models, alerts, decisions and what they returned.
4:40
One word, three sumsData models is the semantic layer. net_revenue means three different things in three models, and the page says so before two dashboards disagree in a meeting.
4:52
Fan-outAnd two joins will silently multiply a measure: a many-to-many join counts each row once per match, so the total comes out high.
5:05
AlertsAlerts: three rules are firing now, two of them critical, including the freshness rule on channel stock.
5:13
Never firedOne rule is armed and has never fired, so nobody knows whether it works. Severity and status are kept as two separate facts.
5:27
ArcArc is the decision register. A proposal is only admitted if it names a metric the platform computes and a scope, and one approved decision can never be measured, because it applies everywhere.
5:41
MeasurableMeasurable is the tier a scope earns: a synthetic control, a comparison group, or movement only.
5:52
LeapLeap reads what each decision actually did. Every settled decision had an effect nobody predicted, and two of them cost money.
6:01
ValueThree decisions measured, worth ₹4 crore between them after the costs that were found.

Fix the feed, ask again

6:08
Fix the feed, ask again. One repair in Data, and Intelligence answers.
6:14
Over in Antvia DataThe feed belongs to Antvia Data, where the connectors live. The portal has issued a fresh key.
6:25
Six stagesConnect, extract, land, rebuild the models, run the quality gates, publish.
6:45
AnsweredThe same question now answers: air coolers and fans hold most of the excess, and purifiers are twelve days short of target.
6:55
CheckedAnd it says why the answer can be trusted: every table behind it is current and passing its checks.
7:05
Stood downIn Alerts, the freshness rule on channel stock has stood down by itself. Two rules are firing now, and nobody touched the third.
7:17
CurrentBusiness pulse shows channel stock as current. One fix, and every surface that reads the table followed.

Close

7:25
Numbers that say what they are built on. Dashboards that flag their own stale and unverified tiles. Answers that restate the question and refuse when data is stale. One definition per measure, and joins that cannot inflate quietly. Decisions measured against the number they were meant to move.
Next step

See it on your own data.

Thirty minutes on the half you came for. Bring the report that breaks every month, or the decision you never settled.