BACKGROUND: Info & Data Preparation

The below questions are related to the case described in “Chapter 20: Local Mobile Rewards“ from Cutting-Edge Marketing Analytics, and reference the following documents: http://ntrda.me/2yoRDZ5 (HW #4 instructions) and http://bit.ly/2h7M4H3 (gsheet of the data).

For easy reference, below is the code and commentary, which includes background details regarding cleaning, prep and analysis for the presented commentary, models and recommendations. The results are described in the response section for each question. While multiple visuals and techniques were applied for Q4, this RMD highlights the most straightforward results – which are the figures more easily digestible to executive stakeholders.

When running this RMD, key elements to remember are: 1) set working directory, 2) load the below libraries (e.g. coefplot, ggplot2 & gsheets etc.) and 3) read the CSV into the sved data frame.

## Loading required package: coefplot
## Loading required package: ggplot2
## Loading required package: gpairs
## Loading required package: gsheet
## Loading required package: quantmod
## Warning: package 'quantmod' was built under R version 3.4.1
## Loading required package: xts
## Warning: package 'xts' was built under R version 3.4.1
## Loading required package: zoo
## 
## Attaching package: 'zoo'
## The following objects are masked from 'package:base':
## 
##     as.Date, as.Date.numeric
## Loading required package: TTR
## Warning: package 'TTR' was built under R version 3.4.1
## Version 0.4-0 included new data defaults. See ?getSymbols.
## Loading required package: readxl

QUESTION 1: Mobile Coupon Success

QUESTION: When is a mobile coupon considered successful for the retailer, the consumer, and the mobile-coupon-provider?

RESPONSE: Definitions of mobile coupon usage success, or rather a mobile coupon strategy success, varies by audience (e.g. retailer, consumer, mobile-coupon-provider) and subgroups within those audiences – and their corresponding key performance indictors.

For example, certain a consumer base may define success as having a wide range of coupon options while others just want the very best options tailored to their affluent tastes. Even still, a different consumer base may greatly value the ease of use of the technology (e.g. frictionless, few bugs, doesn’t spam with ads or slow down processing power) while others have a higher tolerance for bugs but want a firehose of new coupon options each time they engage. As such, success greatly varies and needs to be customized through specific goals, understood needs, shared definitions by stakeholders and a defined window of time (e.g. success in the short term vs. long term).

However, broadly and more holistically, mobile coupon usage –and the strategy executed around it– can be successful when it:

  • Retailers
    • Spreads aweareness of a retailer
    • Spreads targeted messaging and offerings from a retailer
    • Attracts new consumers and lapsed consumers
    • Works in tandem with other marketing efforts (and does not cannabilize or conflict with other offerings)
    • Provides an additional understanding of a retailer’s consumers via the redeemption rate of served coupons
    • Provides an additional understanding of a retailer’s consumer base via the third-party mobile-coupon provider
    • Doesn’t solely provide discounts to consumers who would have paid full price
    • Fosters brand loyalty and, hopefully, brand advocacy
    • Leverages an easy-to-use and accurate platform for a retailer to access
  • Consumers
    • Is used and meets a customer’s expectactions/satisfaction
    • Provides an ease of use experience to access deal or an offering of interest
    • Desired volume and type of coupons (e.g. industry )
    • Desired type of coupon as relating to geogrpahical proximity
    • Develop a loyalty for a retailer
    • Feel their time using the coupon, engaging the retailer or engaging the mobile-coupon provider was not a waste of time
  • Mobile-coupon providers
    • Proves return-on-investment value to advertisers and retailers (which varies, but could include # of views, # conversions, # of downloads)
    • Engages their coupon-loving consumer base
    • Serves and matches retailers and consumers (% of success varies based on goals) resulting in purchases or awarenesss etc./end goal of the retailer
    • Provides accurate tracking, cookies internally for product development to leverage, insights/use cases for the sales org

Ideally, the key metrics by each defined audience overlap – and there is real synergy between the mobile coupon provider, consumers and retailers. Trust, reliability and usefulness are all in lock step.

QUESTION 2: Retailer & Consumer Adoption

QUESTION: What aspects of Cardagin determine retailer adoption and consumer adoption?

RESPONSE: Cardigan has two groups it needs to balance (as they sometimes have competing interests) and work to optimize its offerings toward: retailers and consumers. A possible third group could be outside advertisers. Size of one may influence the size of the other. If there are too few consumers or too few retailers, it is unappealing for both groups. In order to attract a sizeable amount of retailers and users (and advertisers), many platforms look to leveraging deep discounts, exciting product offerings or swaying thought leaders to use/endorse/review their product – a few first great interactions are key in adoption. Generally, the larger the better (however, as noted above, this is not necessarily always the priority set). Great interactions vary by retailer and consumer needs. Examples below.

  • Great interactions that lead to adoption for retailers could include: adoption from/exposure to new or lapsed customers, ability to track/key analytics from Cardagin data and hopefully increase share-of-voice.
  • Great interactions that lead to adoption for consumers could include: relevance of coupons offered (e.g. tailored or quality of the coupons), engaging with brands (particularly aspirational or new brands), easy-to-understand offerings and hopefully a great mobile coupon experience that turns them into an advocate for the Cardigan platform.

Additionally, the platform itself needs to work well and be reliable. Investment in updating functionality, design and security are imperative. Attrition in the mobile coupon space is incredibly easy, and best practice –anecdotally at least– is to provide the best services.

QUESTION 3: Returns of Mobile Coupons, Local Newspaper Advertising & Coupons

QUESTION: How do you compare the returns from mobile coupons and local newspaper advertising or coupons?

RESPONSE: Comparing advertising in a multi-touch attribution strategy is a difficult endeavor. Factors regarding the return on investment –such as the long-term engagment vs. short-term ROI– are very important. The sheer growth and customization of mobile coupons is worthwhile to track, but assessing these figures alone (or against simple newspaper ad spend) do not provide a holistic view on swaying consumer opinion. For example, consumers may have had multiple exposures to marketing content before they decide to purchase sometihing, so it is not necessarily fair to attribute success to solely mobile coupons if a consumer used a mobile coupon or signed up for the Cardagin Daily Deal. The saying “half the money I spend on advertising is wasted; the trouble is I don’t know which half” is very applicable here.

One way to provide additional context –aside from tracking literal returns from mobile coupons vs. newspaper coupons– would be to do a consumer survey, where respondents indicate if they’ve seen newspaper, mobile etc. ads and/or ever used a newspaper clipping, mobile coupon, etc. This then can provide a fuller story as to what ‘half’ might be working and showcase that, indeed, newspaper advertising isn’t working or that there is great value. That said, it’s important for consumer surveys need to be very thoughtfully designed, executed and analyzed – as one doesn’t want fatigued respondents or junk data to inform budgetary decisions.

QUESTION 4: Regression

QUESTION: Run a regression of the log of number of redemptions (lnumredeem) on industry category dummies, free shipping, all-free, discount percent, and coupons offered by competition weighted by spatial distance

## Warning: Ignoring unknown aesthetics: xmin, xmax

RESPONSE: Below is a LM regression of the log of the number redemptions on industry category dummies, free shipping, all-free, discount percent and coupons offered by spatial distance (all). Observed data oddities were transformed where appliacable (e.g. percent discounts above a hundred –> hundred, free proudcts other than zero transformed into 1s).

As noted in the code, other models –GLM with Poisson, leveraging spatial 5, etc.– produced very similar results to the model presented below. The primary takeaways are the same.

According to the dataset provided, a coupon’s industry category are less of a factor than, for example, a coupon offering free shipping (coefficient: 0.8393552) or a free product (0.4972148). Auto and nightlife are the ‘stickiest’ (or, in this case, they are the least negative) of the categories (-0.0871407 and -0.2095280 respectively); however, this isn’t as meaningful given the coefficients are negative. It’s interesting to note that spatial lag (0.0001960) and discount (0.0131560) also have positive coefficients, but they are very close to zero.

The “Q4 Residual vs. Fitted” plot shows a generally random pattern, indicating it’s a decent fit for a linear model regression. Again, while there is slight downward trend and slight clustering, the overall results remain consistent based one the results compared. Some endogeneity may still be impacting the results, but –based on data investigation conducted– the findings remain unimpacted, and therefore the most straightforward, replicable model was selected.

plot_coeffQ4(Q4)

coefplot(Q4, color="#748c63")
## Warning: Ignoring unknown aesthetics: xmin, xmax

plot(Q4$fitted.values, Q4$residuals, ylab="Residuals", xlab="Fitted Values", main="Q4 -- Residuals vs. Fitted", col="#748c63")
abline(0, 0) # the horizon

#alternatives
#pairs(logred ~ arts + auto + beauty + food + health + hotel + localflavor + nightlife + profesh + shopping + sportsfit + freeship + allfree + discountpercent + spatiallagall, data=card)

QUESTION 5: Coupon Design Recommendations

QUESTION: What recommendations can you give retailers regarding coupon design from your regression results?

RESPONSE: Similar to defining success, recommendations should be customized to the particular needs and goals of a retailer, which factors in timing, ROI, their consumer base and other elements previously described above. However, given Cardagin’s results in Q4 and current platform configuration, it can broadly recommended that free shipping should be included in any mobile-coupon strategy within its platform. In fact, free shipping is often the expectation among consumers –especially when trying a new offering– as reported by HuffPost, NerdWallet and Reuters. Additionally, including a free product within the mobile-coupon offering could also be broadly recommended.

In specific instance, say a health retailer was interested in leveraging the Cardagin platform but did not want to use free shipping an offer a free product, I would consult them to put their efforts elsewhere or, more likely, request additional data and information to run a mini and targeted test.

That said, I’d also highly advise anyone looking to make decisions by these findings to circle back with Cardagin teammates, if possible, who developed this dataset to better understand the variable definitions – in order to better understand the nuances and assumptions initially made within the dataset’s categories. Also, the observed redemptions need to be closely scrunitized regarding endogeneity, such as factors that are not present in the model, ommitted causes, measurement error, common method variance or overly correlated results. As discussed by John Antonakis, a more holistic analysis via experiments can reduce potential spurious findings and highlight the true relationships within a dataset.

NOTE: This HTML RMD appears to print fewer than 7 pages | http://rpubs.com/tatoonie/MAHW4