Business fit vs technical fit, and what "deploy" means here

The most-confused pair on the paper

Two tasks sound alike and sit in different phases. Assess model is the last task of modelling: does the model fit technically? Is R² high, are the group sizes sensible, did the pivot sum to the grand total? Evaluate results is the first task of evaluation: does the result serve the business question — does it meet the success criterion fixed back in business understanding, and would it change the decision?

A model can pass the first and fail the second, and that is the case the paper builds its stems on.

Worked example — a line that fits and a line that fails

Maple & Birch's analyst fits a line predicting monthly sales from advertising spend across six stores. Its R² is 0.83 — the line explains most of the variation, group sizes are fine, nothing is wrong with the fit. Assess model: pass. Then she reads the slope: it is negative — more advertising, less sales. The business question was whether raising the advertising budget would lift sales; a model whose slope has the wrong sign cannot answer it, however well it fits. Evaluate results: fail. The right move is not to deploy the line and not to fiddle with it, but to go back to business understanding: is the question right, is there a confounder (the biggest stores advertise least because they do not need to), is a different variable needed? That is the outer loop from Lesson 1.3.

The evaluation phase has three tasks: evaluate results against the business criterion; review the process (was anything skipped or done badly?); and determine the next steps — deploy, iterate, or stop.

What "deploy" means in this course

Nobody in AFM 112 ships software. Deployment is putting the answer in front of the decision-maker in a form they can use, with a plan to keep it usable. Its tasks: plan the deployment (what form, for whom), plan monitoring and maintenance (who refreshes it, how often, what would make it stale), produce the final report, and review the project. The crew dashboard is the course's deployment — and the marks are for the dashboard with its refresh plan and owner, not for the file alone.

A one-off chart pasted into a memo is also deployment. Deployment is not defined by the technology; it is the phase where the result leaves the analyst's hands.

One deliverable per phase

Each phase produces a thing you can hand over, and the crew project is marked on those things:

• business understanding → the analytic question with its criterion
• data understanding → the data dictionary (plus a quality note)
• data preparation → the prepared dataset
• modelling → the model description (the pivot layout or the fitted equation)
• evaluation → the evaluation memo (result vs criterion; next step)
• deployment → the dashboard with a monitoring plan

Tag four statements

(1) "The pivot's grand total matches the sheet's SUM" — assess model. (2) "The Alberta gap is $650, well past the $100 criterion, but rests on two orders, so it cannot change the decision yet" — evaluate results. (3) "R² is 0.83" — assess model. (4) "This answers a question the owner no longer cares about" — evaluate results (and a return to business understanding). The tell: a statement about the numbers' quality is assessment; a statement about the decision is evaluation.