DBA 833 Topic 1 prediction goal framing paper example

Predictive Modeling Grand Canyon University Free custom sample in 24 to 48h

This DBA 833 Topic 1 sample paper splits a composite business-software company's request, a model of why customers cancel to see who will cancel next, into two questions. It was written for GCU DBA 833, Predictive Modeling, catalog code DBA-833, a core course of Grand Canyon University's DBA. About 14 percent of 9,000 accounts leave each year. Following Shmueli and Breiman, the ranked call list for six account managers is prediction, judged on held-out renewals, while the onboarding redesign is a causal question passed to a separate comparison study. A table assigns each later decision to its goal, and Ascarza's finding on targeting by risk is noted.

CourseDBA 833 Predictive Modeling
TopicTopic 1
Paper typePrediction goal framing paper
LengthAbout 1,060 words, 2 pages
FormatAPA 7 student paper
SchoolGrand Canyon University (GCU)
ProgramDBA
UpdatedOctober 2026

Free sample paper for DBA 833 Topic 1

1

Two Questions in One Request: Separating Which Customers Will Cancel From Why They Cancel, and Keeping the Predictive Goal

[Author Name]

College of Doctoral Studies, Grand Canyon University

DBA-833: Predictive Modeling

Topic 1 Prediction Goal Framing

[Instructor Name]

[Date]

The organization, setting and figures below are a composite written as a model document. No real employer, client, colleague or patient is described.

What this page is doingThe title names the split the paper makes and the goal it keeps. The company and its request are a composite; the sources are real.
2

Two Questions in One Request: Separating Which Customers Will Cancel From Why They Cancel, and Keeping the Predictive Goal

The Request

The chief revenue officer of a composite business-software company, which sells a scheduling product on annual subscriptions to about 9,000 small and mid-sized firms, sent the analytics team a one-line request: build a model of why customers cancel so we can see who is going to cancel next. About 14 percent of accounts decline to renew each year. The request reads as one task, but it holds two questions, and each serves a different decision. The retention team wants a ranked list of accounts to contact in the ninety days before renewal, so that its six account managers spend their calls where cancellation is likeliest. The product group wants to know whether a redesigned onboarding sequence, introduced eight months ago, reduces cancellations, so that it can decide whether to keep it. This paper separates the two before any model is chosen, keeps one and states what that choice permits and forbids.

Explanation and Prediction Are Different Goals

Shmueli (2010) argued that explanatory and predictive modeling differ not only in purpose but in nearly every design decision that follows. An explanatory model tests a causal theory: its variables are chosen because theory says they cause the outcome, its coefficients are read as effects, and it is judged by whether those effects are estimated without bias. A predictive model aims to forecast outcomes for cases not yet seen: its variables are chosen for what they add to accuracy, whether or not they cause anything, and it is judged by error on new data. A model can predict well while explaining nothing, and a model that estimates a causal effect correctly can predict poorly, because an unbiased estimate of one effect may leave most variation unexplained.

Breiman (2001) offered a second lens, describing two cultures in statistics: one that assumes a stochastic model generated the data and estimates its parameters, and one that treats the process as unknown and judges algorithms by predictive accuracy. The first culture's habits, reading coefficients and testing their significance, are natural for explanation. The second's habits, holding out data and measuring error, are natural for prediction. The request mixes the vocabulary of both.

The Predictive Question

Which accounts will cancel at their next renewal is a predictive question. Success means the ranked list finds cancellations among accounts the model has not seen, measured on renewals that occur after the model is built. Three consequences follow.

First, variables are admitted for what they add to accuracy on new accounts. Declining weekly log-ins, a fall in the number of active users, unresolved support tickets and a recent change of billing contact may all predict cancellation, and none needs a causal story. A variable that is merely a symptom of a decision already made, such as a request for a data export, may predict extremely well; it belongs in the model if it is available ninety days before renewal and excluded if it usually appears only in the final weeks.

Second, no coefficient carries a causal reading. If accounts using a particular feature cancel less, the model may weight that feature heavily, but the product group cannot conclude that pushing the feature would reduce cancellations. Heavy users may simply be the firms that would have stayed anyway.

Third, success is measured on held-out renewals, by how many of the accounts in the top part of the list actually cancel compared with the base rate of about 14 percent, not by fit on the data used to build the model.

The Explanatory Question

Whether the redesigned onboarding reduces cancellations is a causal question. It asks what the cancellation rate would have been for the same customers under the old onboarding, which no predictive model can answer. Comparing customers who joined before and after the redesign would mix the onboarding with everything else that changed in those months, including pricing, the sales team's targets and the economy. The question needs a comparison design, for example randomizing new customers between the old and new sequences for a period, or comparing cohorts on either side of the change with careful checks. This paper passes it to a separate study with its own plan. Keeping it out protects both projects: a predictive model built to answer it would be judged by the wrong standard, and a causal study squeezed into a churn model would inherit variables chosen for accuracy rather than for identification.

Table 1

Later Decisions and the Goal That Governs Each

Later decisionGoverned byConsequence for the churn model
Which variables to includePredictive goalAny variable available 90 days before renewal that improves accuracy
How to judge the modelPredictive goalError and ranking quality on later, held-out renewals
How to read coefficients or importance scoresPredictive goalAs contributions to accuracy, never as effects
Whether onboarding worksExplanatory goalNot answered by this model; separate comparison study
What to tell account managersPredictive goalWhich accounts to call, not why they will leave

What the Choice Permits and Forbids

Keeping the predictive goal permits the team to use any information legitimately available at the time of scoring, to choose among model families by held-out performance and to change the model whenever a better one appears. It forbids presenting the model as an account of why customers leave. Account managers will want reasons, and the model's most influential variables will look like reasons; the team must say clearly that a low log-in count marks an account at risk without showing that log-ins keep customers.

The Claim the Model May Support

If it performs as intended, the finished model may support one claim: among accounts renewing in the coming quarter, those it ranks highest are more likely to cancel than the rest, by a measured margin on renewals it did not see. That claim is enough to direct the retention team's calls, with one caution. Accounts most likely to cancel are not necessarily those a call would save; in field experiments, Ascarza (2018) found that targeting customers by predicted risk was less effective than targeting those most responsive to the intervention. Learning who responds would be a later, experimental step. It is not enough to redesign the product, set prices or judge the onboarding, and the report to the chief revenue officer should say so in its first paragraph.

References

Ascarza, E. (2018). Retention futility: Targeting high-risk customers might be ineffective. Journal of Marketing Research, 55(1), 80-98. https://doi.org/10.1509/jmr.16.0163

Breiman, L. (2001). Statistical modeling: The two cultures (with comments and a rejoinder by the author). Statistical Science, 16(3), 199-231. https://doi.org/10.1214/ss/1009213726

Shmueli, G. (2010). To explain or to predict? Statistical Science, 25(3), 289-310. https://doi.org/10.1214/10-STS330

Reading the GCU DBA 833 Topic 1 instructions

Topic 1 in DBA-833 asks doctoral students to frame the goal of a predictive project before any model is chosen. Strong papers show when a request holds more than one question, name the decision each serves and decide whether each is predictive or explanatory. Students should explain how that choice governs which variables may enter, how coefficients may be read and how success is judged, and pass any causal question to a design able to answer it. The paper should end with the single claim the finished model may support and say plainly what it cannot be used for.

How this DBA 833 Topic 1 example is built

The finished paper takes the request as the chief revenue officer wrote it and shows that it contains two goals. The retention team wants a ranked list of accounts to contact before renewal, which is prediction: success means the list finds cancellations among accounts the model has not seen. The product group wants to know whether a redesigned onboarding sequence reduces cancellations, which is explanation of a causal kind and needs a comparison the paper does not attempt to build. Shmueli's argument that explanatory and predictive modeling differ in purpose, in evaluation and even in which variables belong is used to set the two apart. Breiman's contrast between the data-modeling and algorithmic cultures supplies a second lens. The paper then keeps the predictive goal and states what that choice permits and forbids.

Reading the DBA 833 Topic 1 rubric

Faculty reward a request split before modeling. An Excellent paper separates the call list from the onboarding question using Shmueli on explaining versus predicting and Breiman's two cultures. Credit goes to admitting variables for accuracy at ninety days, reading no coefficient causally and judging success on held-out renewals. The rubric values a table assigning each later decision to its goal and passing the causal question to a comparison study. Strong papers note Ascarza on targeting by risk. Top marks go to stating the one claim the model may support.

DBA 833 Topic 1 help before you submit

Deductions start when a paper builds a churn model and then reports its coefficients as the reasons customers leave. A model tuned to rank accounts can put weight on a variable simply because it travels with cancellation, and that weight says nothing about what would happen if the company changed it. The mirror error treats an accurate but opaque model as a failure because nobody can read a mechanism from it, when mechanism was never the goal. Papers that leave the goal implicit make every later choice defensible for whichever purpose the reader happens to assume. Answering the onboarding question with the retention model promises a causal finding from data never designed to support one. Citing Shmueli as ranking prediction above explanation misreads an argument that was about matching method to purpose.

Get a DBA-833 Topic 1 example written to your instructions

Send the DBA-833 Topic 1 instructions and the rubric from your classroom, with the business problem or case your section assigned. We write a custom example to them, with prediction or explanation chosen as the goal, both questions separated, the consequences for later choices tabled and the permitted claim stated, in 24 to 48 hours. The first one is free.

More DBA 833 papers and DBA samples

DBA 833 Topic 1 questions, answered

Can one model serve both explanation and prediction?

Sometimes, but rarely well for both at once. A model shaped to test a causal claim restricts itself to variables theory justifies and accepts some loss of accuracy, while a model shaped to predict admits anything that improves performance on new cases. Shmueli notes that strong explanatory power does not guarantee good prediction, and the reverse holds as well. The example therefore assigns one goal to each question.

Should an opening framing paper include a fitted model?

Typically the framing comes first and the fitting later. The topic asks for the goal, the decision it serves and the evidence that will count as success, and a full model at this stage tends to bury the framing decision under results. The example shows a framing paper for a composite company, as DBA-833 coursework, and later topics are where preparation, fitting and validation typically appear.

Where can I find a free DBA 833 Topic 1 sample paper?

The framing paper is shown here complete: the request, explanation against prediction, the predictive question, the explanatory question, what the choice permits and forbids and the claim the model may support. Post your DBA 833 Topic 1 prompt and get a free paper.

Why exclude a data export request from the model in DBA 833 Topic 1?

It predicts cancellation extremely well but usually appears only in the final weeks before renewal. A model scored ninety days ahead cannot use information that does not yet exist at that point.

Why not judge onboarding with the churn model?

Comparing customers who joined before and after the redesign mixes the onboarding with pricing, sales targets and the economy in those months. A causal answer needs a comparison design, such as randomizing new customers between sequences.