An independent study reference written by Dr Phuc V. Nguyen. It is not official subject material — for assessment requirements always follow your subject outline and vUWS.
The analytics workflow
Analytics work follows a recognisable sequence: frame the decision, acquire data, assess and clean it, analyse, interpret the result against the original question, communicate, act, then monitor what happened. The sequence is not a one-way pipeline. Interpretation regularly sends you back to acquisition, and a first look at the data often rewrites the question. Two stages carry most of the risk. Framing decides whether the exercise is aimed at anything, and preparation consumes far more effort than newcomers expect, because operational data is recorded for billing and compliance rather than for analysis.
Why it matters
It resembles cooking for people who have not said what they want to eat. The mistake is to start chopping. First find out who is eating and when, then see what is actually in the fridge, then throw out what has gone off, and only then cook. Serving the dish is part of the job. So is noticing afterwards whether anybody ate it.
Halfway through building a churn model, the analyst finds that the customer status field was only populated correctly from March onwards. What is the best next step?
Worked examples
A university wants to reduce first-year withdrawal and asks for an analysis. Walk the workflow through it.
Framing turns the topic into a decision. The support team can make about three hundred outreach calls in the first month, so the question is which three hundred students. Acquisition pulls enrolment records, learning system activity and submission records. Preparation finds activity data missing for one delivery mode, which has to be handled openly rather than quietly dropped. Analysis ranks students by estimated risk. Interpretation tests the ranking against the constraint: if the top three hundred contains 84 eventual withdrawals against 21 for a random three hundred, the list earns its place. Communication hands over names with the reason each appears. Monitoring compares the following year.
A month into a project, the analyst discovers the data cannot answer the question that was asked.
This is normal and it is not failure. The workflow loops. The right move is to go back to framing with the decision owner and say precisely what the data can and cannot support, then agree either a narrower question the data can answer or a plan to collect what is missing. The failure mode is to answer the original question anyway using data that does not support it, because the answer will look complete and nobody downstream will know which part was improvised.
Common mistakes
- ✗The workflow runs once, from start to finish. It loops constantly. Nearly every project revisits its question after seeing the data, and interpretation frequently sends the analyst back to collect more.
- ✗Framing is a short administrative step before the real work. Framing is where most projects are won or lost, because a precisely stated decision determines what data is relevant and what result would count as useful.
- ✗Data preparation is a chore that can be handed off and forgotten. Cleaning choices are analytical choices. How missing values, duplicates and outliers are treated can move the result more than the choice of model does.
- ✗The work ends when the findings are presented. Without action and monitoring you never learn whether the analysis helped, so the next project starts with no evidence to build on.
Revision bullets
- •Frame, acquire, prepare, analyse, interpret, communicate, act, monitor
- •The sequence loops; interpretation often sends you back to the data
- •Framing converts a topic into a decision with a constraint attached
- •Preparation dominates effort because operational data was recorded for other purposes
- •Monitoring is what lets the next project learn from this one
Quick check
Halfway through building a churn model, the analyst finds that the customer status field was only populated correctly from March onwards. What is the best next step?
What best distinguishes a well framed analytics question from a poorly framed one?
Connected topics
More in What Business Analytics Is
Sources
- Fayyad, Piatetsky-Shapiro & Smyth (1996)Fayyad, U., Piatetsky-Shapiro, G., & Smyth, P. "From Data Mining to Knowledge Discovery in Databases." AI Magazine, 17(3), 37-54, 1996.An early statement that discovery is an iterative multi-stage process in which data selection and preparation dominate.
- Wirth & Hipp (2000)Wirth, R., & Hipp, J. "CRISP-DM: Towards a Standard Process Model for Data Mining." Proceedings of the Fourth International Conference on the Practical Application of Knowledge Discovery and Data Mining, 29-39, 2000.Describes the widely reused cycle from business understanding through to deployment, with loops built in.