| Course | BU 443 Marketing Analytics |
|---|---|
| Unit | Unit 3 |
| Paper type | Channel performance analysis |
| Length | About 910 words, 6 pages |
| Format | APA 7 student paper |
| School | Herzing University |
| Program | BS in Marketing |
| Updated | September 2026 |
Free sample paper for BU 443 Unit 3
Paid Social Goes From Last to First When the Credit Moves: A Channel Performance Analysis for a Composite Online Houseplant Retailer Under Two Attribution Models
Student Name
Business, Herzing University
BU 443: Marketing Analytics
Instructor Name
February 25, 2026
Paid Social Goes From Last to First When the Credit Moves: A Channel Performance Analysis for a Composite Online Houseplant Retailer Under Two Attribution Models
The Model Comes First
Greenline Plant Co., the composite Portland houseplant retailer followed in this course, wants to know which marketing channels earn their cost before deciding its spring budget. A channel comparison is meaningless until it says how credit for a sale is assigned, so the model is named here before any figures appear. The main analysis uses last non-direct click attribution: each purchase is credited to the last marketing source a buyer arrived from, ignoring visits where the buyer typed the address or used a bookmark, unless no other source exists. The same comparison is then repeated under first-click attribution, which credits the source of the buyer's first recorded visit, to see which conclusions survive a different rule.
Everything else is held constant. The period is March 2026, after the label and duplicate-purchase fixes from the tracking audit. Purchases are the 3,314 de-duplicated orders recorded by analytics; revenue is gross order value, $214,600; spend is the money paid to each channel in March, including the affiliate commission paid to plant bloggers and the email service's monthly fee. The figures belong to the composite case.
Results Under Last Non-Direct Click
Cost per purchase is spend divided by purchases; return on ad spend is revenue divided by spend.
| Channel | Spend | Sessions | Purchases | Revenue | Cost per purchase | Return on spend |
|---|---|---|---|---|---|---|
| Paid search | $14,200 | 21,300 | 690 | $43,500 | $20.58 | 3.06 |
| Paid social | $9,500 | 27,100 | 360 | $23,800 | $26.39 | 2.51 |
| Organic search | $0 | 38,900 | 820 | $51,600 | Not applicable | Not applicable |
| $1,100 | 14,600 | 910 | $57,400 | $1.21 | 52.18 | |
| Blogger referral | $1,780 | 6,200 | 280 | $17,800 | $6.36 | 10.00 |
| Direct | $0 | 34,200 | 254 | $20,500 | Not applicable | Not applicable |
| Total | $26,580 | 142,300 | 3,314 | $214,600 |
Read this way, email is the most efficient channel by far, blogger referrals are cheap, paid search is solid, and paid social is the weakest paid channel, costing about $26 per purchase. The owners' first instinct on seeing this table was to move money from paid social to paid search.
Results Under First Click
The second model credits each purchase to the source of the buyer's first recorded visit. Spend is unchanged; only the credit moves.
| Channel | Purchases | Revenue | Cost per purchase | Return on spend | Change from last click |
|---|---|---|---|---|---|
| Paid search | 610 | $38,300 | $23.28 | 2.70 | Slightly worse |
| Paid social | 640 | $41,200 | $14.84 | 4.34 | Much better |
| Organic search | 900 | $57,000 | Not applicable | Not applicable | Slightly better |
| 380 | $24,100 | $2.89 | 21.91 | Much worse, still efficient | |
| Blogger referral | 430 | $27,400 | $4.14 | 15.39 | Better |
| Direct | 354 | $26,600 | Not applicable | Not applicable | Better |
| Total | 3,314 | $214,600 |
Which Conclusions Survive
Three conclusions hold under both models. Organic search is a large source of sales under either rule, about a quarter of purchases. Blogger referrals are cheap under both, costing less than $7 per purchase. And paid search performs steadily, around $21 to $23 per purchase, so its value does not depend on the rule chosen.
One conclusion reverses. Under last click, paid social is the weakest paid channel; under first click, it is the most efficient, at under $15 per purchase. The pattern makes sense: social ads introduce people to Greenline, who later return through search or email to buy. Last click gives that sale to the channel that closed it; first click gives it to the one that opened the relationship. The owners' plan to cut paid social rested entirely on the first model.
Email shifts in the opposite direction. It closes many sales, but it rarely starts a relationship, since people must already know Greenline to be on its list. Its efficiency under last click partly reflects credit for customers other channels found.
Comparing Channels of Very Different Size
The channels differ greatly in volume, from 280 blogger-referred purchases to more than 900 from email, and small channels produce less stable figures. A cost per purchase built on a few hundred orders can move sharply from one month to the next through chance alone. To check whether blogger referrals are genuinely cheap or merely had a lucky month, the same calculation was run for January and February. Their cost per purchase stayed between about $5 and $8 in all three months, which suggests the result is real rather than noise. Paid social's last-click cost ranged from $24 to $31 across the same months, so its poor showing under that model is also stable. The difference between the models, not month-to-month noise, is what drives the conflicting readings.
How the Owners Should Read the Two Tables Together
The practical lesson is to read the two tables as bounds rather than as rival answers. For a channel that mainly introduces customers, such as paid social or blogger referrals, first click is closer to its upper bound and last click to its lower. For a channel that mainly closes sales among people who already know the business, such as email and direct visits, the reverse holds. A decision that looks right under both models, such as keeping paid search and blogger partnerships, can be taken with some confidence. A decision that flips between them, such as cutting paid social, cannot be taken from attribution data at all and needs a different kind of evidence.
What This Analysis Cannot Say
Neither model measures what a channel adds. Both divide up sales that happened; neither asks how many would have happened without the channel. Li and Kannan (2014) compared attribution rules with experimental evidence and found the rules could misjudge how much a channel really adds, and Lewis and Rao (2015) showed that measuring the true return on advertising is statistically very hard, because advertising effects are small relative to the natural variation in purchases. For a business of Greenline's size, that means no attribution table can settle whether paid social pays for itself.
The analysis therefore recommends no budget shift yet. It recommends holding paid social steady until a controlled comparison can be run, such as pausing it in some regions while keeping it in others, which the evaluation and testing units take up. Farris et al. (2010) caution that attribution reports are allocations, not measurements, and this analysis treats them that way. The retailer and its March numbers were invented for the course.
References
Farris, P. W., Bendle, N. T., Pfeifer, P. E., & Reibstein, D. J. (2010). Marketing metrics: The definitive guide to measuring marketing performance (2nd ed.). Pearson Education.
Lewis, R. A., & Rao, J. M. (2015). The unfavorable economics of measuring the returns to advertising. The Quarterly Journal of Economics, 130(4), 1941-1973. https://doi.org/10.1093/qje/qjv023
Li, H., & Kannan, P. K. (2014). Attributing conversions in a multichannel online marketing environment: An empirical model and a field experiment. Journal of Marketing Research, 51(1), 40-56. https://doi.org/10.1509/jmr.13.0050
What the BU 443 Unit 3 instructions ask for
BU 443 Unit 3 asks students to analyze the performance of marketing channels. Prompts usually require a stated attribution model, a fixed period and consistent definitions, a comparison of channels on cost and return, and a second analysis under a different model to see which conclusions hold. Many instructors ask how to compare channels of very different size and what to conclude if the second model barely changes anything. Students should explain that attribution divides credit rather than measuring causal effect. Composite data are acceptable. Two comparison tables and a discussion in APA 7 are typical, with a conclusion that states which findings are robust and which are not.
How the BU 443 Unit 3 example is put together
The analysis opens by naming the model and fixing the period, purchase count, revenue and spend definitions. A last-click table gives spend, sessions, purchases, revenue, cost per purchase and return for six channels, followed by the owners' first reading. A first-click table repeats the comparison with a column showing how each channel moved. A section sorts conclusions into those that survive both models and the one that reverses, explaining why paid social introduces customers who buy later through other channels. Further sections check stability across months and show how to read the two tables as bounds. The close explains what attribution cannot measure and recommends a controlled test.
Where the points sit in the BU 443 Unit 3 rubric
Herzing's channel analysis rubric generally rewards a named attribution model, consistent definitions, accurate calculations, a robustness check under a second model and conclusions limited to what the data support, with writing and APA. Naming the model earns credit only if it comes before the results. Consistency scores highest when period, spend and outcome definitions are identical for every channel. Calculations are judged on accuracy and clear formulas. The robustness criterion rewards showing which conclusions hold and which reverse. The conclusions criterion pays for recognizing that attribution allocates rather than measures, and for proposing how the open question could be answered. Instructors often credit stability checks across periods for small channels.
Common BU 443 Unit 3 mistakes, and how to avoid them
Channel analyses commonly lose points by presenting results without naming the attribution model, or by naming it only at the end. Another frequent problem is mixing periods or spend definitions across channels. Students often treat attribution results as the effect of each channel. Some run a second model but do not explain what changed or why. Others rank channels by revenue alone without considering spend. Small channels are frequently judged on a single month. Finally, recommending large budget shifts from attribution data alone overreaches; the stronger answer states which decisions the evidence supports and which need a test. It also helps to show one worked calculation, such as cost per purchase for a single channel, so the instructor can follow the arithmetic behind every table.
Get a BU 443 Unit 3 example written to your instructions
Send the Unit 3 instructions and the BU 443 rubric your classroom shows, plus the dataset or case the assignment provides. We write a custom example that declares its model first, holds one period and one definition set, runs the comparison under two models and keeps only the conclusions surviving both. First custom sample free, back in 24 to 48 hours.
More BU 443 papers and marketing samples
- BU 443 Unit 1: Metrics definition exercise
- BU 443 Unit 2: Data collection and tracking audit
- BU 443 Unit 4: Customer and cohort analysis
- BU 443 Unit 5: Campaign evaluation
- BU 443 Unit 6: Test design and reading
- BU 443 Unit 7: Dashboard or visualization
- BU 443 Unit 8: Analytics recommendation
- BU 423 Unit 3: Business model canvas write-up
- BU 427 Unit 3: Readiness assessment record
- BU 323 Unit 3: Segmentation and targeting
- BU 439 Unit 3: Content strategy per platform
BU 443 Unit 3 questions, answered
Which attribution model should the analysis use?
Whichever your instructions specify. Where the decision is left to you, take the model your own data can actually support and state its consequences plainly. What matters more than the selection is that you say what the model does with a path involving several sources, and that you show at least one alternative reading so the ranking is not offered as a fact about the world.
What if the second model barely changes anything?
That is a result worth reporting and it strengthens the paper. Say the ordering held, print both tables anyway and explain why the stability makes your conclusion sturdier. Stability usually means one channel dominates the paths in this data, which is itself a finding. What loses credit is running a single model and then asserting the ranking would have held regardless.
Can I compare channels with very different volumes?
Yes, provided the counts are on the page and the commentary refuses to let a tiny base win. Report the volumes beside the rates, say which comparisons are too thin to lean on, and consider grouping the smallest channels rather than granting each a row that implies more precision than this data holds.
Where can I find a free BU 443 Unit 3 sample paper?
The whole BU 443 Unit 3 channel performance analysis is published free on this page, with both attribution tables, the robustness discussion and margin notes. If your channels or period differ, share your instructions and we write a matching analysis free, usually within two days.
Which attribution model should BU 443 Unit 3 use?
Name one before the results, then repeat the analysis under a second. The sample uses last non-direct click and first click, and shows that paid social's ranking reverses between them while other conclusions hold.