MT362 Unit 9 AI Vendor Comparison Matrix Example

Contract Terms Before Demonstrations: An MT362 Unit 9 Vendor Comparison Matrix for Churn Prediction

[Student Name]

Purdue University Global

MT362: Artificial Intelligence Applications for the Marketing Professional

Unit 9 Vendor Comparison Matrix

[Instructor Name]

[Date]

The vendors are composites described by type, and their terms, prices and claims are illustrative; the regulations and sources cited are real.

Weights Fixed Before the First Demonstration

The weighting memo is dated two weeks before the first sales call and signed by the director of lifecycle marketing, and the matrix shows it first so a reader can confirm that no criterion was reweighted after a vendor impressed the team. Data use and training restrictions carry 25 percent of the score. Deletion, subprocessors and data location carry 10 percent. Score explanations carry 15 percent, access for the chain's own bias audits 15 percent, integration effort 15 percent and three-year cost 20 percent. Each cell holds a score from 1 to 5 and the clause or page that supports it.

The Three Vendors

The vendors are described by type. Vendor E is a churn module inside the email platform Fernbrook already uses. Vendor S is a standalone retention-prediction firm. Vendor D is a predictive add-on to a customer data platform. Nothing in the matrix describes how any vendor's model works internally; every cell rests on a document the vendor supplied.

Who May Learn From the Chain's Data

Data terms carry the most weight because the Unit 4 review made them the condition of using customer records at all. Vendor E's terms allow it to use de-identified customer data to improve products for all of its clients. Vendor S's terms forbid any use of Fernbrook's data beyond Fernbrook's own service. Vendor D's terms permit use for its own product improvement with an opt-out. Under California's regulations, a service provider generally may not use a business's personal information to build or improve services for other businesses, beyond narrow exceptions (California Privacy Protection Agency, 2023), so a vendor that trains for all clients may not qualify as a service provider at all. The matrix routes that question to counsel and scores Vendor E a 2, Vendor D a 4 and Vendor S a 5. That single clause moves more points than any price difference.

One clause about whose models may learn from the chain's data decides the comparison more than any demonstration did.

Deletion, Subprocessors, Location

Each vendor's promised deletion window after termination, its list of subprocessors and the location of stored data are quoted directly. Vendor E deletes within 30 days and lists five subprocessors, scoring 5. Vendor S deletes within 45 days and lists four, scoring 4. Vendor D names eleven subprocessors and could not say which of them touch subscriber records, scoring 2.

Scores the Chain Can Audit

The Unit 7 bias review needs every score and its top reasons exported each month. Vendors S and D offer that export and score 4 for audit access. Vendor E shows scores and reasons only inside its own dashboard, which earns it a 2, since the chain could not join them to birth years for the audit. On explanations themselves, Vendors E and S each provide top reason codes for every score and score 4; Vendor D provides only general feature importance and scores 3.

Integration and Cost

Vendor E integrates with no new data connections and scores 5, at about $21,000 a year, $63,000 over three years, also a 5. Vendor S needs about 220 analyst hours to connect order and schedule data and costs $48,000 a year, $144,000 over three years plus the integration time, scoring 3 for integration and 4 for cost. Vendor D needs similar integration work and costs $62,000 a year, $186,000 over three years, scoring 3 on both.

The Totals

Weighted totals are 4.10 for Vendor S, 3.65 for Vendor E and 3.30 for Vendor D. Vendor S leads despite costing more than Vendor E because its data terms and audit access are the two criteria the earlier units made conditions of the project. A sensitivity check confirms the order is stable: even if cost were weighted at 30 percent and data use cut to 15, Vendor S would still finish first, narrowly.

Claims Held for the Pilot

Vendor S's case study reports a 23 percent reduction in cancellations for another retailer, and Vendor D's materials state 92 percent accuracy. Both sit in a separate column marked claim to be tested, and neither enters the totals. In September 2024 the Federal Trade Commission announced a set of enforcement actions against deceptive claims about artificial intelligence (Federal Trade Commission, 2024), and that sweep is the reason vendor assertions are treated as hypotheses here. The Unit 8 pilot is named as the only evidence the chain will accept, consistent with Kohavi et al. (2020), who argue that controlled experiments, not vendor reports, show what a change does in one's own setting.

Recommendation

Proceed to contract negotiation with Vendor S for the pilot only, on a quarter-to-quarter term, with the data-use clause, monthly score exports and deletion window written into the contract. If Vendor S declines quarter-to-quarter terms, Vendor D becomes the alternative, conditional on naming which subprocessors touch subscriber data.

References

California Privacy Protection Agency. (2023). California Consumer Privacy Act regulations (Cal. Code Regs. tit. 11, § 7000 et seq.). California Privacy Protection Agency.

Federal Trade Commission. (2024, September 25). FTC announces crackdown on deceptive AI claims and schemes [Press release]. Federal Trade Commission.

Kohavi, R., Tang, D., & Xu, Y. (2020). Trustworthy online controlled experiments: A practical guide to A/B testing. Cambridge University Press.