City University of Hong Kong · FACULTY OF INFORMATION TECHNOLOGY

IS6400 Chap.1 Business Analytics Workflow

- one subject, every graph, every model, every mark
6 Chapters2-page Bible
Our own words - no uploaded lecturer files
Updated for this semester
Chapter 1 of 10 · IS6400

Business Analytics Workflow

Begin with the decision

The course places business understanding, data understanding and preparation before model building. That order prevents a team from optimising a score for a target that does not match an operational choice.

Define who acts, what is decided, when it is decided, what one row represents and what later outcome will test success.

Backtracking is part of the method

Testing and evaluation may expose an earlier mistake. Leakage sends the project back to preparation; unstable segment results may send it back to data understanding; a technically strong score that changes no action sends it back to business understanding.

Deployment therefore includes a threshold, owner, action, monitoring measure and exception route.

Write the information boundary

Place the decision timestamp on a timeline and classify every candidate field as available before, at or after that point. Fields created after the outcome are leakage even when they are stored in the same convenient table.

Historical aggregates need their own cutoff rule so a later transaction cannot enter an earlier prediction. Keep the rule with the feature definition, because a clean train-test split cannot repair a feature that already contains future information.

Evaluate the action, not only the score

Choose a baseline that represents the current process, then test the candidate on later or otherwise independent rows.

Translate its errors into workload, missed opportunities, waste or another consequence the organisation recognises. A threshold determines how many cases receive action, so report performance at that threshold and check capacity.

If the model changes no decision, the project needs a new action rule or a return to business understanding rather than another tuning cycle.

Make deployment reversible

Name the operational owner, monitoring frequency, inputs that may drift and outcome that will arrive later. Define an exception path for cases outside the training range and an escalation trigger for workload or harm.

Save the model version, preprocessing rules, threshold and baseline comparison. These details let the team distinguish a changed population from a coding failure and let it pause or roll back the system without reconstructing the original analysis from memory.

In this chapter

What this chapter covers

  • 01

    Business understanding and an operational owner

  • 02

    One row and one prediction horizon

  • 03

    Data understanding before feature construction

  • 04

    Preparation without future leakage

  • 05

    Testing against a baseline

  • 06

    Deployment, monitoring and backtracking

Worked example · free

Define a replenishment decision before modelling

Q [8 marks]. AskSia-authored practice. The listed marks are AskSia study allocations, not a University rubric. A retailer asks for a demand model for a promoted product. Define the row, target, time boundary, baseline and deployment action without choosing an algorithm.
The return path is the important edge.The return path is the important edge. Evaluation identifies the earliest stage that can repair a decision, rather than sending every failure back to model tuning. A model is one stage inside a business decision cycle BusinessquestionPrepareddataModel andtestDecision anddeploymentEvaluation sends weak results back to the earliest stage that can repair them
The return path is the important edge. Evaluation identifies the earliest stage that can repair a decision, rather than sending every failure back to model tuning.
  • 2Specify the decision unit and horizon. The marks shown here are not an official University mark allocation.
  • 3Name fields that would leak future information.
  • 3Define the baseline, action and monitoring trigger.
Use one store-product-order-date as the row and forecast demand over the following selling week. All fields must exist before the order cut-off, so later sales, later footfall and eventual stock-out status are excluded. Compare with a recent uncensored average and evaluate forecast error together with waste and shortage cost. The output changes the order quantity or sends a case for review. Monitor signed error and shortage rate by promotion type; a sustained directional miss after a policy change triggers a data and feature audit.
Sia tip — Write when each feature becomes available. A useful field that arrives after the decision is leakage, not intelligence.
Glossary

Key terms

Decision Unit
The entity and moment to which one prediction and one action apply.
Target Horizon
The future interval between the decision point and the outcome being predicted.
Data Leakage
Information entering a model that would not be available when the live decision is made.
Baseline Model
A simple reference rule that a more complex candidate must beat under the same evaluation.
Monitoring Trigger
A stated condition that starts investigation, retraining or escalation after deployment.
FAQ

Business Analytics Workflow FAQ

Why should the business question come before the model?

Because the question fixes the row, target, horizon, costly error and action. Without those choices a model can optimise a technically valid score that cannot be used, or can learn from information that arrives after the real decision.

Does a weak test result always mean the algorithm is too simple?

No. It may reveal a vague target, censored outcome, missing segment, bad join or time leakage. Return to the earliest stage contradicted by the evidence before increasing model complexity.

What turns a prediction into a deployed analytics process?

A deployment identifies the operational owner, operating threshold, action, exception route, monitoring measure and escalation trigger. A score delivered without those pieces is an analytical output, not a controlled business capability.

Study strategy

Exam move

Take one familiar business request and write seven lines: operational owner, action, row, target, horizon, costly error and baseline. Audit every feature against the decision timestamp. Then sketch the six process stages and place each possible failure at the earliest stage able to repair it. Finish by naming what would be monitored after release and which change would trigger retraining rather than a threshold adjustment.

Rehearse one failure at every stage: a vague action, an ambiguous row, a censored target, a leaking field, a mismatched metric and an unowned deployment. For each, point to the earliest repair and explain why later model tuning cannot solve it. End with a one-minute spoken summary that connects the baseline comparison to a reversible operational decision.

A+Everything unlocked
Unlocks this Bible + all 31 of your subjects at City University of Hong Kong, plus 1,000+ more Bibles across the AskSia library.
Sia - your IS6400 tutor, unlimited, worked the way the exam marks it
The full 2-page Bible + practice bank with worked solutions
Chrome extension - sync your LMS so Sia knows your deadlines
Bilingual EN / Chinese on every Bible and every Sia answer
$0.99 Trial
30-day money-back · cancel in one tap · how it works