OACadvanced
Machine Learning & Auto Insights in OAC
Explain OAC's built-in machine learning capabilities — Auto Insights, Explain, and custom ML models within Data Flows — and how they augment traditional BI analysis.
These are the features that most clearly separate OAC from classic OBIEE: the platform proposing things rather than answering what you asked.
Auto Insights is like a smart photo app that automatically flags the best, most interesting shots from thousands of vacation photos instead of making you scroll through every single one yourself — useful for surfacing candidates quickly, but you still decide which photos actually make the album.
Key Concepts
1
Auto Insights examines a dataset and surfaces what is interesting.
trends, outliers, correlations, clusters
-> presented as generated visualisations with a short narrative2
The value is in the questions you did not ask. An analyst looking at revenue by region will not notice that
one product category behaves differently in one month unless they happen to slice that way.
3
Explain goes the other direction. Select a column or a value and ask why:
4
"Explain" on Customer Churn
-> which attributes most distinguish churned customers
-> how the rate differs across segments
-> what is anomalous5
It runs statistical analysis and presents the drivers. That is a first pass at a question that would
otherwise be an afternoon of slicing.
6
Custom models are trained in a data flow.
train a dataset -> a model (classification, regression,
clustering)
inspect quality metrics: accuracy, precision, recall
apply score new data in another flow7
So a churn model is built, evaluated and applied without leaving OAC and without a data scientist.
8
Which is the caveat worth stating plainly. These tools make modelling accessible, and they do not make
the reasoning behind it accessible.
9
a correlation is not a cause
a model with 95% accuracy on imbalanced data may be predicting
"no" every time
a driver identified by Explain may be a proxy for something
you already knew10
The quality metrics are shown, and reading them is the user's responsibility. An accessible tool producing a
confident wrong answer is a real risk, and an interviewer asking about these features is usually checking
whether you treat them as a starting point or as a conclusion.
11
Where they genuinely earn their place. Exploration — pointing you at where to look — and scoring at
scale once a model has been properly evaluated. Not as a replacement for understanding the data.
12
What the interviewer is probing.1. "What does auto insights do?" Probing: the feature. Stalls: "Predicts." Moves up:
suggests visualisations and highlights patterns in a dataset automatically, as a starting point for
exploration.
13
2. "What machine learning is actually available?" Probing: the breadth. Stalls: "Deep
learning." Moves up: training and applying standard models — classification, regression,
clustering — within OAC, plus registering models from elsewhere.
14
3. "What is the caution you would raise?" Probing: the judgement. Stalls: "None." *Moves
up:* a model trained by someone without the statistical background can be confidently wrong, so
validation and a named owner matter more than the tooling.
15
4. "Where does it genuinely help?" Probing: the practical angle. Stalls: "Everywhere."
Moves up: surfacing a correlation nobody thought to look for, which a human then investigates —
not as an unattended decision maker.