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.
Why analytics needs a methodology
A methodology is an agreed sequence of steps, each with a named output, that analytics work is expected to pass through. It buys three things that talent alone does not. It forces the business question to be settled before anyone touches a table. It makes the work traceable, so a second person can see which rows were dropped and why. It makes iteration deliberate rather than accidental, because the loop back to an earlier step is part of the design instead of a sign of failure. Without one, analytics tends to start at whatever data is convenient and drift towards whatever result is interesting.
Why it matters
A methodology is closer to a pilot checklist than to a recipe. It does not tell you what to cook. It tells you which things have to be true before you take off, and it makes you say them out loud. Most analytics work that goes wrong does not fail at the modelling step. It fails because nobody wrote down, in one sentence, whose decision this was meant to change.
A team delivers a customer segmentation with strong technical validation and clear charts. Nobody uses it. Which methodological step most likely went missing?
Worked examples
A retail team is asked to "do something useful with the loyalty data". Two months later they present a six-segment customer clustering. The room nods politely and nothing changes.
The work skipped the first step. A methodology would have required a sentence naming a decision, a decision-maker and a date: the category manager chooses next quarter's promotion mix by the end of month two. From that sentence the segmentation either earns its place or it does not. Note what was not wrong here. The clustering was probably correct, the code probably ran, and the charts were probably attractive. Technical quality cannot rescue an unowned question.
An analyst removes about 4,000 transaction rows with impossible timestamps, ships the report, and leaves the organisation. Four months later finance disputes the revenue figure.
Nobody can reconstruct the exclusion rule, so the number cannot be defended or corrected. It has to be rebuilt from scratch, and the rebuilt figure will not match. A methodology fixes this cheaply: every step has a written output, and the data preparation output records the filters applied and the reason for each. The cost of that record is a few minutes. The cost of not having it was a month of rework and a loss of trust in every number the team had produced.
Common mistakes
- ✗A methodology is bureaucracy that slows delivery down. The steps themselves are cheap. What is expensive is discovering at the end that the question was wrong, and that is precisely the failure a methodology is designed to catch early.
- ✗Following a methodology guarantees a useful result. It protects against process failures such as an unowned question, an undocumented filter or an unreviewed assumption. It offers no protection at all against data that is unfit for the question, or against a question nobody cares about.
- ✗A methodology means a fixed forward march through numbered stages. Every serious analytics methodology is a cycle with explicit backward arrows. Returning to an earlier step because you have learnt something is the intended behaviour, not a mistake.
- ✗Small pieces of work do not need one. On a two-day task the methodology collapses to a few sentences, but the sequence still applies. Naming the decision, checking the data can support it, and recording what you excluded takes about ten minutes on any size of job.
Revision bullets
- •Methodology = agreed steps, each with a named output
- •Buys three things: question settled first, traceable work, deliberate iteration
- •Its cheapest saving is catching a wrong question before the build
- •Backward arrows are designed in, not a sign of failure
- •Undocumented data decisions cannot be defended later
- •It cannot rescue unfit data or a question nobody owns
Quick check
A team delivers a customer segmentation with strong technical validation and clear charts. Nobody uses it. Which methodological step most likely went missing?
An analyst excludes rows with impossible timestamps and does not record the rule used. The published figure is challenged months later. What has actually been lost?
Connected topics
More in How Analytics Gets Built
Sources
- 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.Sets out the argument for a documented, industry-neutral process rather than ad hoc project practice.
- 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.The earlier KDD process framing, which already treats discovery as an iterative sequence with explicit loops rather than a single modelling act.