Information Strategy Flashcards
(189 cards)
1
Q
3 digital macro trends
A
Computational Power; Connectivity of devices; Big Data
2
Q
How to evaluate the impacts of digital technologies on industry dynamics?
A
Porter’s 5 Forces; Changes in value chain
3
Q
3 reasons why some industries are more vulnerable:
A
- Newly easy to enter;
- Attractive to attack;
- Difficult to defend
4
Q
Industry vulnerability as easy to enter condition
A
Regulatory change;
Changing customer needs;
Reduction in minimum scale required in order to compete;
Changes in distributions system;
Market size;
Nature of the product;
Market has been shocked - status quo changed
5
Q
Industry vulnerability as attractive difficult to defend
A
Existing pricing structure;
strategic inflexibility;
lack of vision;
legacy systems;
barriers prevent incumbents from immediate replication.
6
Q
What are the challenges in the technology adoption?
A
Technology challenge - which technologies will take off or die out?;
Business challenge - which startups will push out incumbents?;
Information challenge - storing, analyzing, tapping into data.
7
Q
What is disruptive innovation and explain the disruptive innovation model?
A
Disruptive innovation =
- Targeting fringe market
- Underperform on mainstream attributes & overperform in other
- High improvement potential on the attributes that mainstream customers value
Disruptive innovation model =
Describes a process whereby a smaller company with fewer resources is able to successfully challenge established incumbent businesses. More specifically:
- Incumbents focus on improving their products for their most demanding customers;
- Entrants begin to successfully targeting those overlooked segments; • Incumbents tend not to respond;
- Entrants then move upmarket, delivering the performance that mainstream customers require;
- When mainstream customers start adopting the entrants’ offerings in volume, disruption has occurred.
8
Q
• What is causing the long-tail phenomenon?
A
- Supply side: Virtual shelf space, Aggregation of consumers, Electronic delivery &bMade-to-order production
- Demand side: Search tools, Recommendations systems, web-based tools Customer reviews & online communities
9
Q
• How does long-tail phenomenon affects industry dynamics?
A
- Supply side: Increase incentive to develop new products, Restructuring of marketing strategies & New intermediaries and industry structures
- Demand side: Changes in consumer tastes and demand patterns, Positive feedback & Culture changes from access to more varied source of information
10
Q
What is unbundling?
A
Companies have broken the products by offering just the most valuable part to for consumers. For example, iTunes offering separate songs.
11
Q
What is disintermediation?
A
Companies have broken links of the supply chain by going directly to consumers or by jumping steps in the supply chain. For example, Dell going directly to consumers.
12
Q
What is decoupling?
A
Decouplers identify activities consumers perform in the customer value chain that can be broken so that part of that value can be delivered and captured by decoupler. For example, cosmetics sample boxes, Amazon, Zynga.
13
Q
Long Tail Strategy
A
retailing strategy of selling a large number of different items which each sell in relatively small quantities, usually in addition to selling large quantities of a small number of popular items.
14
Q
What did Netflix do right?
A
- It captured changes in technology by shifting from fragile VHS to DVDs
- Recommendation system
- Served niche interests
- The long-tail
- Unlimited subscription model with no late fee
- Pricing model
- Customer lock-in
- They raised switching cost
- Operational optimization
15
Q
Why do incumbents overlook disruptive innovations?
A
- They emphasize different product or service attributes
- They target different customer segments
- They start out as small and low-margin businesses
- They conflict with existing way of doing business
16
Q
Components of digital business model
A
- Content (What is consumed?) Product information, price, use details, e-books, movies, software.
- Experience (How is it packaged?) Customer-facing digitized business process, community and customer input, expertise for informed decision making, recommendations
- Platform (How is it delivered?) Internal - other business processes, customer data, technology External - proprietary hardware, public networks, partners
17
Q
A digital business model challenges the physical model in three main areas:
A
- Internal power: the one who ‘owns’ the customer’s experience often changes from product groups to the unit that manages the multiproduct customer experience;
- Business processes: these require rethinking, to be seamless across channels;
- Customer data: this is becoming an enterprise-wide resource rather than a resource remaining hidden in one area
18
Q
Three trends have converged to raise the stakes for the effectiveness of your enterprise’s digital business model:
A
- Digitization
- Increasing number of digital natives
- Amplification of customer voice
19
Q
Where should you start on content, experience, platform?
A
- If your goal is driving new digital revenue, then start with strengthening your content.
- If your goal is cross-selling and driving more revenue per costumer, focus first on improving your customer experience.
- If your goal is efficiency and flexibility, then focus first on building and exploiting shared digital platforms.
20
Q
Anomalies in disruptive innovations
A
1. Low-end disruptors vs new-market disruptor:
Low-end disruptors come at the bottom and take hold of an existing value network before moving upwards, while new-market disruptors create completely new value network and appeal to customers who did not have a product (PCs)
2. Disruptors vs incumbents:
When both incumbents and disruptors follow the same plan, incumbents will remain (higher education).
21
Q
Features of information goods
A
- Utility of information is different among consumers
- Utility depends on the time of consumption (newspaper, weather forecast)
- Utility increases with exclusivity but very difficult to maintain
- Indirect network effects (reviewers + restaurants)
- Negative externalities (congestion)
- Information goods are very easily aggregable
22
Q
Physical nature of information goods:
A
INDESTRUCTABILITY
- no second-hand market
- coase conjecture
- no perishing
TRANSMUTABILITY
- Extreme customizability
- Fake news: So, rather than trying to protect content integrity, sellers need to differentiate their products by customizing and updating and selling them as interactive services.
- Business model impact: selling as a service not as products (SAP)
REPRODUCIBILITY
- First unit involves a lot of cost, second unit almost has zero marginal costs
- Monopolies emerge (decreasing marginal costs, high barriers to entry, winner-takes-all markets)
- Scale effects
- Illegal reproduction (to counter: update, change your products)
23
Q
Coase conjecture
A
The loss of market power for durable goods. Like any durable good, a producer of digital products therefore competes with its past sales, therefore forced to charge competitive prices even when there is no competitor. The more software you sell the less customers you’d have to sell to.
The Coase conjecture, developed first by Ronald Coase, is an argument in monopoly theory. The conjecture sets up a situation in which a monopolist sells a durable good to a market where resale is impossible and faces consumers who have different valuations. The conjecture proposes that a monopolist that does not know individuals’ valuations will have to sell its product at a low price if the monopolist tries to separate consumers by offering different prices in different periods. This is because the monopoly is, in effect, in price competition with itself over several periods and the consumer with the highest valuation, if he is patient enough, can simply wait for the lowest price. Thus the monopolist will have to offer a competitive price in the first period which will be low. The conjecture holds only when there is an infinite time horizon, as otherwise a possible action for the monopolist would be to announce a very high price until the second to last period, and then sell at the static monopoly price in the last period. The monopolist could avoid this problem by committing to a stable linear pricing strategy or adopting other business strategies
24
Q
How can producers remedy the coase conjecture and not lower the prices?
A
- Credible and fixed price announcements
- Rental models (licensing/subscription)
- Steady updates of products
- Frequently updating
25
There are two "ways" to price information goods
1. So since marginal costs are close to zero, value-based pricing makes more sense than cost-based pricing, thus you can use interactive determination of prices - auctions, stock markets
2. Or you can set prices and make it "take-it-or-leave-it" offer but differentiate the pricing with respect to: - Buyer characteristics - Usage characteristics - Product quality
26
Producer rent
Revenues above variable costs
[https://www.youtube.com/watch?v=z0wg9ZPyL38](https://www.youtube.com/watch?v=z0wg9ZPyL38)
27
Consumer rent
What consumers would have paid
[https://www.youtube.com/watch?v=z0wg9ZPyL38](https://www.youtube.com/watch?v=z0wg9ZPyL38)
28
Dead Weight Loss (DWL)
The total lost value of trades that didn't occur in a market, but would have occurred if the market was at its equilibrium point.
[https://www.youtube.com/watch?v=z0wg9ZPyL38](https://www.youtube.com/watch?v=z0wg9ZPyL38)
29
The utility of IG differts greatly between consumers -\> Differential pricing.
Three types of differential pricing:
1. Personalized pricing (1st degree or perfect price discrimination) = sell to each user at reservation price -\> auctions
2. Versioning/ Bundling (2nd degree or price & quality or quantity) = offer product line -\> pro vs lite & single vs set
3. Group pricing (3rd degree price discrimintation) = priced on group membership or identity -\> students vs non-students
30
How to version as a developer?
- Design in a way so it's easily versioned
- Create the highest quality and then take out features
- How many # of versions? As many versions as there are consumer segments Offer at least three versions
31
Versioning
The provision of two different versions/quality levels of product aimed at different consumer segments.
Aim: Make consumers self-select to a product depending on their willingness to pay.
Examples: business vs economy flights, lite vs pro versions or **free in-store pickup or paying for express delivery for product**
How: Set prices accroding to WTP & CR to maximize profit
Design to customize, start with HQ version & offer versions for every segment with at least 3 versions extremeness aversion so other versions look better
Let op prijs & Cr als je prijs van versions bepaalt
CR = WTPgroup - Pgroup
When WTP is heteregeneous, make the lower quality even worse to maximize profit
32
Explain bundling and its benefits
Bundling = the joint offer of products that are related to each other to sell them as one unit.
How: find bundle with maximum total consumer valuation to maximize profit
Upsides
1. Cost saving sfor prouction & transportation
Complementarity between bundles
sorting of customers based on valuation
33
4 types of bundling:
1. Pure bundling = products exclusively as a bundle -\> (A,B)
2. Mixed bundling = single components or a bundle -\> A,B or (A,B)
3. Mixed components = bundle or limited number of components -\> A or (A,B)
4. Pure components = No bundling -\> A,B
34
Define information good
An information good is a product, service, experience or artifact that is valuable to people by virtue of the information it contains.
35
Understand the role of transaction costs in shaping the structure of the firm
The higher transaction costs, the more likely you are to be a hierarchy. If you have high asset specificity and high complexity of product description, thus transaction costs would be higher for you, thus you are more likely to go for hierarchy.
36
Understand the role of IT in the evolution of Electronic Markets
IT makes:
- Product description less complex (easy access to large datasets of product features reduce the complexity of finding each piece)
- Assets less specific (wider markets and better access to technologies)
37
Signaling
An action by an informed party to reveal information. In contract theory, signalling is the idea that one party (termed the agent) credibly conveys some information about itself to another party (the principal). For example, a Harvard degree.
38
Recognize and distinguish information exploitation: adverse selection and moral hazard
- Adverse selection happens when sellers engage in opportunistic behavior until the market gets annoyed.
- Moral hazard happens when people are not incentivized correctly through contracts to do the right thing. For instance, for banker if the loan succeeds, they have a high pay-off but if it does not, govt will bail them out.
39
Objective of English Auction
Extract the maximum valuation from the buyers in the short-term. Leads to an unstable market.
40
Objective of Dutch Auction
Extract the maximum valuation from the buyers while keeping prices controlled.
41
Moral hazard
Moral hazard is an ex-post information asymmetry problem
In economics, moral hazard occurs when someone increases their exposure to risk when insured, especially when a person takes more risks because someone else bears the cost of those risks.
Example:
Driver may behave more dangerously after signing a car insurance contract
42
Signaling
An action by an informed party to reveal information. In contract theory, signalling is the idea that one party (termed the agent) credibly conveys some information about itself to another party (the principal). For example, a Harvard degree.
43
Screening
An attempt by an under informed party to filter information. For example, Spartans asking strong and brave slaves to fight for Sparta (= risk screening for revolts, and then killing them OR expensive lung disease insurance for smokers).
44
Features of the Second Price Sealed Bid Auction
- Less transparent: you don't know the fees;
- Gives good insight of market valuation of the good;
- Easier for the buyers, but less revenue for the sellers.
45
Key takeaways of CoolBlue:
- Omnichannel digital strategy
- Focus on customer experience and platform
- A lot of data mining: combine quantittive data of market analysis with qualitative interviews and control supply chain by communicating specs & sweetspots to producers -\> Mac, Lenovo direct mail
\* Direct mail naar inwoners met geschikt dak voor zonnepanelen -\> auto's elektrisch, zonnepanelen, private lease & oude auto inruilen -\> gratis private lease + besparing op energierekening
46
What are the benefits of electronic auctions?
- Determine the value of the product/service when valuations in the market are heterogeneous.
- Reduce search costs for the best transaction partner.
- Increase the procurement reach.
47
3 types of Sales Auctions
1. English Auction
2. Dutch Auction
3. Sealed Bid Auction
- First Price Sealed Bid
- Second Price Sealed Bid
- Generalized Second Price Auction (Google Search)
48
Objective of English Auction
Extract the maximum valuation from the buyers in the short-term. Leads to an unstable market.
49
Objective of Dutch Auction
Extract the maximum valuation from the buyers while keeping prices controlled.
50
Objective of Sealed Bid Auction
Extract the maximum valuation from the buyers while keeping the competition fair and the market price close to the true value of the good.
51
Features of the First Price Sealed Bid Auction
- Extremely transparent: you know what you pay;
- Less straightforward insight on the market valuation;
- Brings higher revenues;
- It is slightly harder for the buyers who have to calibrate their bids with respect to their valuation, their desired surplus, and their competitors' strategy.
52
Features of the Second Price Sealed Bid Auction
- Less transparent: you don't know the fees;
- Gives good insight of market valuation of the good;
- Easier for the buyers, but less revenue for the sellers.
53
Key takeaways of CoolBlue:
- Omnichannel digital strategy
- Focus on customer experience and platform
- A lot of data mining
54
Features of a platform
- Nexus of rules and architecture
- It is open or closed, allowing regulated participation
- Actively promotes interactions among different partners in a multi-sided market
- Scales much faster than pipeline business
- Value is created outside of the firm (community)
55
What is the definition of network effects?
Compare: Network effects vs. Economies of Scale
Network effect = network's value to user depends on number of network users and the WTP for network participation is the cap on platform fees. Network externalities are demand-side economies of scale: rising network effects increase existing and prosoctective user's willingness-to-pay and -participate.
Unit cost reductions that result from network growth should not be labeled network effects
- Supply side EoS: Falling average costs -\> Monopolistic supply
- Demand side EoS: Value grows with volume -\> Monopolistic demand
56
Two-sided 4 network effects:
- Same-side positive = users Facebook
- Same-side negative = competition for dates & auctions
- Cross-side positive = Uber drivers & riders
- Cross-side negative = more advertisements on web/ tv -\> les attractive for readers
57
• Explain and give examples of positive and negative same-side and cross-side network effects
- Same-side positive - AirBnB user reviews help you to choose the appropriate home
- Same-side negative - congestions the more UBER users, the higher prices for you
- Cross-side positive - the more recruiters on LinkedIn the better chance to get a job
- Cross-side negative - more Uber drivers more congestion
58
3 platform flows:
Goods/services, Information & Money
Danger: you can't monetize whatever travels off platform
59
How do you architect the flows in order to charge?
1. Transaction cut = fee per transaction, but not too much (AirBnB, iTunes & Uber)
2. Pay for access = charge the more-needing side (Match.com & LinkedIn)
3. Attention = charge fee for similar match (Adwords, YouTube & Facebook)
4. Pay for tools = freemium model (SAP, LinkedIn & vimeo)
60
How to price a two-sided platform?
Pricing is complicated for two-sided networks due to WTP & other sides's growth. Platform problem is the same as penguin problem: who jumps first?
Choose a price for each side, factoring in the impact on the other side’s growth and WTP
* Subsidy side: attract users and grow network size
* Money side: pay for the privilege of gaining access
* Price-sensitivity vs. quality-sensitivity
Goal: generate cross-side network effects
• User sensitivity to price, quality, brand value, and same-side network effect
61
How do adjacent platofrms (serving similar or overlapping user bases) compete?
Define platform envelopment, give examples and explain threats.
Entry by one platform provider into another's market.
* Combining its own platform functionality with the target's in a multi-platform bundle.
* Leveraging common components and/ or shared user relationships
* Combining in multi-platform bundle
Examples: Windows Media Player (WMP) vs. RealNetwork
Threats: resistent to aggressive pricing; installed customer base, cross-marketing & customer retention
62
What are the results of platform envelopment?
Similar working as bundling.
- Efficiency gains: transaction and production cost savings, improved user experience.
- Price discrimination: reduces heterogeneity in consumer's aggregate valuations
- Strategic advantages: extend market power into complement market; protect core market by foreclosing rivals' access to key component.
63
• Distinguish among different types of platform envelopment and give examples
Foreclosure attack - the target is either a nested network user or a nested component of the attackers platform [complement] (eBay -\> PayPal, RealNetwork and Windows Media Player)
Intermodal attack - the enveloper's platform and its target's are weak substitutes.[weak substitute] (LinkedIn -\> Monster + Netflix & Blockbuster)
Conglomeration attack - the enveloper's platform is functionally unrelated to its target's [functionally unrelated] (Phone & debit card / smartphone + iPod = iPhone)
64
How to defend against envelopment?
- Match the bundle, but bundle vs bundle competition can be fierce
- Find a big brother via mergers or alliances
- Cede and redeploy: change business model -\> Blackberry from hardware to software
- Sue: antitrust litigation
65
• Describe the differences between platform and product based business model
Core Asset
- A platform is an ecosystem with a monetizable core asset while a product works on the platform and helps in monetizing the platform’s core asset. Features of the product are its core assets.
Design
- A platform is designed keeping both producers and consumers in mind while a product is designed keeping only consumers in mind.
Value Creation
- In a platform-based business, value is created through the interactions between users and third parties. The firms charge platform users for accessing the platform. Hence, platform creates an ecosystem value. Products on the other hand create value by charging money for certain features/items based on customer needs. Products, thus create customer value.
Cost
- In products, customer is the one who bears the cost as he is the one who consumes the value created. Platform economy isn’t straight forward. One or more parties pays a transaction cut. For example, a platform may charge for API, based on usage or a platform may incentivize a producer for participation. Similarly, the platform may monetize consumer attention.
Revenue Stream
- Products have a single revenue stream while platforms have many.
Business Drivers
- Enterprises and end users drive the product’s business while consumers and partners drive the platform’s business.
66
• Understand the nature of network effects
Network's value of to a user depends on the number of other network users.
67
• Articulate the effects of platform integration of Instagram & facebook
* Consumers obtain additional utility from using Instagram after its tighter integration with Facebook
* A large fraction of the users Instagram gained are new users, rather than incumbent users of third-party apps
* The integration has positive (negative) impact on consumer usage of big (small) third-party apps
* The integration benefits the Facebook photo-sharing ecosystem
- Overall positive spillover effects on large third-party apps
- And increasing demand for first-party apps
- Negative demand for small third-party apps
68
How to implement a field experiment successfully?
- Decide on the unit of randomization
- Minimize the spillover and crossover effects between experimental treatments
- Decide on complete or stratified randomization (-used with demographic variables)
69
Crossover effect
when an individual from control group deviates from initial assignment and gets to the treatment
70
Spillover effect
when treatment group effects control group
71
Challenges of tapping into Big Data:
DATA
Managing data quality
Availability of data
Getting access to data sources
Managing and integrating data structures
Managing privacy
VALUE
Using analytics for improved decision making
Measuring customer value impact Establishing a business cae
PEOPLE
Analytics and technical skills shortages
TECHNOLOGY
Restrictions of existing IT platforms
PROCESS
Producing credible analytics
ORGANIZATION
Overcoming resistance to change
Working with academia
72
Good Big Data strategy involves
DATA - data assets, ICT strategy
ORGANIZATION - culture, HR strategy
BUSINESS - value proposition, strategy
73
How to monetize data?
Companies can take three approaches to monetizing their data:
(1) improving internal business processes and decisions - process leadership,
(2) wrapping information around core products and services (CapitalOne showing map and logo for customers to easily spot fraudulent transactions) -Product leadership
(3) selling information offerings to new and existing markets - business unit leadership.
74
Explain the implications of physical properties of information goods for supply and demand
Supply:
- Transmutability: It is easy to modify, thus it is very easy to version the product
- Reproducability: Marginal cost of production is almost zero, thus they need to constantly change the product
Demand:
- Coase conjecture (you can't sell as much)
- Indestructibility
75
Explain the beneficial effects of 3rd degree differential pricing for monopolists
Third-degree price discrimination
-Instead of setting one price, a seller could segment their consumer base and sell for two different prices in order to be affordable to both groups and extract their WTP.
76
Explain the different approaches versioning and bundling use to skim off consumer rent.
Versioning aim is to make consumers self-select to a product depending on their WTP.
Bundling facilitates the convenient purchase of several products and/or services from one company.
77
Explain the intuition behind optimized versioning
When consumer WTP is heterogeneous, make the lower quality good even worse to maximize profit. It happens because you can charge people who can afford the better version more.
78
Explain the implications of MC\>0 for bundling
Bundling can become unattractive because you need to sell more items, which cost you.
79
Learn the basics of Incentive Theory: Incentive Compatibility and Contracts. Exemplify contracts against adverse selection and moral hazard.
- As signaling and screening is not always sufficient, designing proper contracts and mechanisms that prevent people from exploiting their information advantage.
- A contract is Incentive Compatible when it reflects the best interest of all participants.
- Each participant should be able to obtain their "best" outcome just by behaving according to their personal preferences.
- The less informed party (principal) needs to find the optimal way to contract with a more informed party (agent) while keeping everyone willing to participate in the transaction.
Contract against adverse selection: compulsory warranties on performance to reveal the true quality of the product
Contract against moral hazard: high extra fees for smoker problems in non-smoker health insurance plans
80
• Explain how information products differ from physical products; and apply theories of pricing, versioning and bundling of digital goods to real-world examples.
.
81
• Examine the key features of digital platforms; and explain the common challenges and strategies in managing them.
.
82
• Articulate the challenges and opportunities in managing data, information, and AI strategies.
.
83
• Distinguish among different forms of online markets and online auctions; identify information-based problems in them and create solutions to solve them.
.
84
• Distinguish among different forms of online markets and online auctions; identify information-based problems in them and create solutions to solve them.
.
85
Machine Learning
Machine learning is the process whereby a computer distills meaning by exposure to training data
86
Cognitive computing
A subset of A.I. that deals with cognitive behavior we associate with ‘ thinking ’ which allows an entity to obtain information from observations , learn and communicate
87
Artificial Intelligence
Broad field of science encompassing not only computer science but also psychology, philosophy , linguistics and other areas
88
5 AI challenges
- Data Management - good data management processes and architecture for quick access to large amounts of data are needed for effective application of AI.
- Data Access - for training classification models, large amount of labelled data is needed; labelling this data is a huge effort.
- Ethical and unbiased AI - Labeled data sets often contain biases, and these biases progress into AI models and automated decisions; automated decision making algorithms have an impact on society and companies that is still difficult to fully grasp
- Talent - Acquiring and retaining top AI talent is a challenge for companies;
- Organizational and cultural changes - effective application of AI requires new ways of working and continuous learning
89
Opportunities for AI:
BANKING AND FINANCE
Credit decisions Risk management Fraud prevention
LOGISTICS
Route optimization Autonomous vehicles
ENERGY
Monitor energy consumption Energy transition
RETAIL
Targeted marketing campaigns Dynamic pricing Virtual trial rooms
HEALTHCARE
Virtual nurses Health monitoring Medication management
90
5 fundamental methods and techniques of AI:
1. Heurisitics = trial & error methods, sufficient but not optimal (vehicle routing)
2. Support Vector Machine = finding boundary line and separation among classes (SPAM filter)
3. Artificial Neural Networks = output generated on input & learning mechanism (Deep Learning & image recognition)
4. Markov Decision Process = framework based on states, actions & probability for complex decision making (self-driving cars, Alpha Go & maintenance)
5. Natural Language Processing = everything concerned with speech recognition and language generation (chat bots, sentiment analysis, Siri, AIME, Google Translate)
91
3 types of transaction costs
- Search and information costs
- Bargaining costs
- Policing and enforcement costs
92
Search and information costs
finding the best goods from the best supplier at the lowest price
93
Bargaining costs
Drawing the best contract at the best conditions with the best supplier
94
Policing and enforcement costs
Making sure everything works out as contracted, and penalize who doesn't respect the agreements
95
5 determinants of transaction costs
- Frequency: of transaction
- Specificity: transaction-specific investments are needed to achieve the least cost?
- Uncertainty: of market
- Bounded rationality: what access to information does the firm have?
- Opportunism: what is the risk of opportunistic behavior?
96
Coordination costs
the cost of all the information processing necessary to coordinate the work of people and machines that perform the primary processes. These costs are decreased in electronic markets
97
Operational risk
Other parties in the transaction willfully misrepresent or withhold information or underperform 'shirk' - their agreed upon responsibilities.
98
Opportunism risk
Lack of bargaining power, or resulting from the execution of a relationship.
99
Move to the middle hypothesis
Despite IT will lead to a greater degree of outsourcing, firms will rely on fewer suppliers and long-term relationships. This is due to:
- Transactional economies of scale: high sunk costs in setting up relationships with suppliers, repeated relationships often more convenient than one-offs
- Incentives to suppliers: easier to incentivize close suppliers to deliver quality, innovation (non-contractible variables)
- Increased costs and reduced benefits of search: suppliers may low-ball their first offer, and raise their prices later. Buyers may thus pay dearly their initial access to lower price form an unknown partner
100
Kuaffman and Mohtadi (2004) expanded the theories from Malone and Clemons and introduced:
Risk Augmented Transaction Costs Trade-Off
Explain the choice between transacting with a limited number of partners or participating in the open market
* (Bigger) companies who want low procurement risk will choose proprietary - expensive platforms.
* (Smaller) companies will choose open platforms to save the procurement costs but will be subject to procurement risk.
* When a market becomes riskier, firms tend to move towards Proprietary Platforms
101
Models of (Electronic) Markets
1. Bilateral Negotiation
2. Sales Auctions
3. Reverse Auction
4. Double Auctions
102
Objective of Reverse Auction
Obtain the best product/service at the lowest price.
103
Objective of double auction
Buy and sell goods which valuation strongly depends by their scarcity and/or market's expectations
104
Industry vulnerability as attractive to attack
- If all participants are earning excess profits
- Profound difference exists in profitability across customers
105
Pareto vs. long-tail
The focus is not on the difference of 80/20 and 82/18. The difference is about where you should focus on. The Pareto states that if 20% of the efford accounts for 80% of the profit you should focus on that 20%. On the other side, the long tail states that if the tail is long enough, you could focus on the tail, which could lead to more profit than is captured in the head. So, they kind of focus on opposite side of the 80/20 distribution.
106
Distinguish unbundling, disintermediation and decoupling
**Unbundling**
- the move from physical to unbundled digital product (iTunes song)
**Disintermediation**
- supply-chain of the business. Dell going directly to customers
**Decoupling**
- identify part of value-chain and distinguish which one you can decouple/interchange the steps
107
What are the axis of the supplier, omnichannel ....
**Knowledge of your end consumer:** value chain vs. ecosystem
**Business design:** simple vs. complex
108
Disruptive innovation phasing
The starting point of the disruptor: low-end, new market
Trajectory: How the incumbent firm reacts?
109
Supplier business model
- Sells through another company
- Potential for loss of power
- Skills: low-cost producer, incremental innovation.
Examples: Sony, insurance via agent.
110
Omnichannel business model
- "Owns" customer relationship
- Multiproduct, multichannel customer experience to meet life events
- Integrated value chain
Examples: CoolBlue, Banks, Retailers.
111
Modular producer business model
- Plug-and-play product/service
- Able to adapt to any ecosystem
- Constant innovation of product/service
Example: PayPal
112
Ecosystem business model
- Provides a branded platform
- Ensures great customer experience
- Plug-and-play third party products (leverage)
- Deep customer knowledge
- Extracts rents
Example: Amazon, Apple, Google, Microsoft
113
The dynamics of two-sided networks
Transactions in two-sided networks always entail a triangular set of relationships. Two user groups interact with each other through one or more intermediaries called platform providers. A platform has an architecture (a design for products, services and infrastructure) and a set of rules (the protocols, rights and pricing terms). These platforms exhibit two types of network effects:
* Same-side effects: increasing the number of users on one side of the network makes it either more or less valuable to users on the same side o Typically negative: e.g. sellers preferring fewer rivals in B2B exchange o Sometimes positive: e.g. Xbox owners value being able to play games with friends
* Cross-side effects: increasing the number of users on one side of the network makes it either more or less valuable to the users on the other side o
Typically positive: e.g. Uber o Sometimes negative: e.g. TV viewers preferring fewer ads
114
Things to take into account while pricing a platform:
Platform providers have to choose a price for each side, factoring in the impact on the other side's growth and willingness to pay.
- Ability to capture cross-side network effects.
- User sensitivity to price: subsidize the network's more price-sensitive side and to charge the side that increases more strongly in response to the other side's growth (Adobe)
- User sensitivity to quality (Video game market and royalty rate)
- Output costs (FreePC)
- Same-side network effects (Autobytel segments geographically to reduce B2B competition)
- User's brand value (marquees)
115
Difference between B2C and B2B commerce
Users: consumers/purchasers, sales reps
Frequency: one-time/repeat
Total order value: low/high
Market: broad & general/small & targeted
Goals: Extra revenue/process optimization
Buying process: simple/complex
Stock status: in-stock/not in-stock
Pricing strategies: standard/specific
Payment process: direct/on account
Fulfillment: 3rd party warehouse/internal warehouse
Strategic position: new line of business/different way of processing orders
116
The move from pipeline to platform involved three key shifts:
1. From resource control to resource orchestration
2. From internal optimization to external interaction
3. From a focus on customer value to value to a focus on ecosystem value
117
Name 5 disruptive technologies (SMACIT)?
Social media,
Mobile,
Analytics (AI),
Cognitive Cloud computing.
Internet of Things
118
Explain the five phases of the Gartner Hype Cycle?
119
What is the difference between digital, digitized and digital disruption?
Digitized = standardizing business processes for cost, reliability & transparency
Digital = introducing speed & connectivity to create new customer value propositions -\> digital offerings
Digital disruption = Digitization of businesses while breaking down industry barriers and creating new opportunities while destroying long-successful business models
120
Explain and exemplify the forces of Porter's five forces?
1. Threat of new entry = specialist knowledge, economies of scale, cost advantages & barriers to entry
2. Buyer power = Number of customers, order size, price sensitivity, ability to substitute & switching costs
3. Supplier power = Number of suppliers, size of suppliers, uniqueness of suppliers' service, your ability to substitute the suppliers & switching costs
4. Threat of substitution = Substitute performance & switching costs
5. Competitive rivalry = Number of competitors, (quality) differences, customer loyalty & switching costs
121
What is decommodization in the travel industry?
Optional checking in bags etcetera to lower price (á la carte pricing)
122
What are 4 reasons incumbents overlook disruptive innovations?
Incumbents:
* Emphasize different product/ service attributes
* Target different customer segments
* Start as small & low-margin businesses
* Conflict with existing thoughts of doing business
123
What is long tail strategy?
Selling large number of different items in small quantities **in addition to** large quantities of popular items
124
Name 4 different business models and explain the matrix?
1. Omnichannel = owns customer relationship, multiproduct, multichannel & integrated value chain (banks and retailers: Coolblue)
2. Ecosystem driver = platform, great customer experience, plug 'n play for third party producers, matches needs with providers & extracts 'rents' (Amazon)
3. Supplier = sells through another company, potential for loss of power & low-cost producer with little innovation (insurance via agent or Procter & Gamble)
4. Modular producer = plug 'n play service, adapt to any ecosystem & constant innovation of service (Paypal)
125
Explain economies of scale effects and its reasons in industrial vs information goods?
126
How can firms actively segment their consumer base (four ways)?
1. Different price sensitivity = difference for groups (business users will be less price sensitive compared to private users).
2. Network effects = low introductory price to benefit from netwerk effects when the utility of a good/ service depends on how many other people are using it
3. Lock-in effect = high switching costs lead to high lock-in effects. Throug low introductory prices or rebates for new customers coupled with higher subsequent prices, lock-ins can be achieved (get 'em while they're young: Oracle etc.)
4. Sharing/ licensing = free access to research journals for university libraries & high prices for industry readers
127
What is a platform?
Platforms unlock new value from spare resources & user-generated content by allowing third party access and communities instead of features
A platform is:
* A nexus of rules and architecture
* Multi-sided: actively promotes (positive) interactions in a multi-sided market
* Scalable: scales way faster than a pipeline business because it does not necessarily bear the costs of external production
* Openness: is either open or closed, based on regulated participation
* Network externalities: leverages network effects
128
How does positive feedback drive Uber's growth (draw) and is every network effect positive?
129
Explain how Adobe prices it platform for PDF writers and readers?
Explain how Operating Systems subsidize developers?
130
How can you use information as a strategy?
1. Name-your-own-price
2. Non-anonymous browsing (reduction in matches)
3. Data integrators (data brokers)
131
What are the two ways to infer causality?
1. More than 2 control conditions (observational data): assumptions and data to deal with the lack of random assignment
2. Experimental data: random assignment to different conditions -\> A/B tests with control and treatment groups are powerfull
132
Explain the role of asset specificity and product description complexity in the Electronic Markets Hypothesis
The hypothesis favors both markets & hierarchies, but in general high asset specificity and high complexity of product description lead firms to shape their structure towards the use of Hierarchies over Market
* Asset Specificity = degree input could be used by other firm
* Complexity of product description = information needed to make purchase decision
133
What is the electronic market hypothesis?
"By reducing the costs of coordination, IT will lead to an overall shift toward proportionately more use of markets rather than hierarchies to coordinate economic activity"
Malone, Yates, Benjamin (1987)
134
Define markets and hierarchies in the EMH?
* Markets = external: high quality from specific supplier at lower price due specialism/ scale
* Hierarchy = in-house production
135
Name the 3 effects of the Electronic Markets Hypothesis?
1. Electronic Communication Effect
2. Electronic Brokerage Effect
3. Electronic Integration Effect
136
Explain the Electronic Communicatin Effect?
* More information communicated faster
* Decrease cost of communication
137
Explain the Electronic Brokerage Effect?
* Increase in the number of alternatives
* Increase the quality of the alternatives
* Decrease the selection costs
138
Explain the Electronic Integration Effect?
* Access to unified data sources allows efficient exchange and use of information
139
Explain market evolution according to the Electronic Market Hypothesis?
140
Explain Clemons' view on IT and Transaction Costs?
* IT reduces coordination cost
* IT reduces operations risk: more information available easier to monitor transactions and partner performance
* IT reduces opportunism risk: reduces cost of access to resources, lower impact of specificity on bargaining power. Also, IT investmens are heavily standardized
All these together reduce the Need for Ownership -\> push firm towards higher degrees of outsourcing
Reality: both long-term relations and open markets co-exist since transaction costs are not static due to change and shocks
141
3 Stages of AI (Deloitte)
1. Assisted Intelligence = requires human assistance & interpretation
2. Augmented Intelligence = Machine Learning augments human decisions
3. Autonomous Intelligence = AI decides & executes autonomously
142
What is asymmetric information?
* An (economic) agent has more information than the principal
* the more informed party willfully exploits their private information to gain (economic) advantage
143
Name the two types of problems regarding information asymmetry
Adverse Selection & Moral Hazard
144
Adverse selection and solves
Adverse selection is ex ante asymmetric information.
Bad risks buy (used car market example) -\> WTP for HQ car decreases -\> no more HQ cars sold -\> P cars decreases -\> WTP decreases -\> no more sales/ buys -\> market collapse
For example: the seller knows the quality of the car, while the buyer does not (both cars look the same). The seller will price a low quality car at the buyer's WTP for a HQ car
Solved by signaling and screening
145
Explain the LemonADS problem
Advertisers move to private marketplaces since open Real Time Bidding (RTB) are too risky.
At the same price point, publishers who engage in both transaction methods (private and Open RTB Marketplaces) offer lower viewabilities on the Open RTB Marketplace!
No collapse due to trust, but trust is not quantifiable
146
How can blockchain and smart contracts solve the trust issue?
* Decentralization and Openness: Blockchain makes transactions and product descriptions trackable and available to all
* Smart contracts promise to allow the transparent execution of contracts without the need for a third-party enforcing system (no more lawyers)
* So, the sellers won't be able to misrepresent the quality of their goods (ideally)
147
Is more information always better?
Informed buyers are not always better off
Uninformed (non-Chinese buyers) paid less for lucky addresses than superstitious informed (Chinese) buyers, because they anticipate the supposed price premium they will obtain in re-sale market
-\> Uninformed botain the higher margin if re-selling
148
Explain the 4 models of electronic markets
149
Explain bilateral negotiations
Buyer and seller find a compromise on the price based on their valuation and their desired surplus, but market information is very inaccurate
150
Why electronic auctions?
1. Determine value when TWP is heterogeneous -\> hard-to-price goods: Art, IPO etc.
2. Reduce search costs for the best transaction partner
3. Increase procurement reach
151
Sales auctions
1 seller, many buyers. Seller extracts the best valuation from competing buyers
152
Name the 3 types of Sales Auctions
1. English auctio: Seller \<-\> Auctioneer/ platform \<-\> Buyers
2. Dutch auction: Seller \<-\> Auction clock \<-\> buyers
3. Sealed bid acution = Seller -\> Auction -\> Buyers & Platform ranks bids and decides which seller and buyer
- First Price Sealed Bid = highest bid pays bid price (houses)
- Second price sealed bid (Vickrey) = hight bid pays the second price (google search) & Generalized if multiple goods
153
Explain the Programmatic Auction Supply Chain
154
Explain the Waterfall mechanism and Header Bidding mechanism in Programmatic Auctions
Waterfall:
1. DSP forwar highest bid received
2. Exchange ranks DSP's on personal preference (% fee)
3. First bid above minimum bid floor is accepted
Header Bidding
1. Waterfall mechanism
2. Extra auction layer to pick highest DSP instead of first in sequence
Exchanges with 2nd price mechanism have a disadvantage with header bidding (lower price) -\> more exchanges move to 1st price
155
Explain gender bias in STEM ads and auction competition spillovers
Women are more likely to convert, which leads to a higher price for ads shown to women, easier to win auctions for men than for women and therefore less ads shown to women
156
Why is the Internet of Things transforming business models? Name three fundamental properties of digital technology
(Iansiti & Lakhani, 2014: Digital Ubiquity)
1. Exact replication: digital signals can be transmitted perfect without error, no matter where (unlinke analog signals)
2. Infinite times: a Facebook page can be shown a billion times without changing
3. Zero marginal costs: no incremental costs for showing page
- \> these properties improve scalability of operations and move industries from product to service revenue. Instead of a features list with pricing and discount caps, we're shaping deals from the group up that are based on the value derived by the customer.
157
5 Ways to approach/ embrace digital ubiquity (digital appearing everywhere):
(Iansiti & Lakhani, 2014: Digital Ubiquity)
1. Apply the digital lens to existing products and services
2. Connect your existing assets across companies: connecting customers' needs by building capabilities to meet them
3. Examine new models of value creation: look which data is needed to extract value with analytics to enable old and new customers
4. Consider new value capture modes
5. Use software to extend the boundaries of what you do: existing capabalities and customer relationships are the foundations for new opportunities
Manage risks:
- Sensitive to economic trends that affect customers -\> absorb risk for its customers
158
What are two important implications and two fields of significant improvement of Machine Learning?
(Brynjolfsson & McAfee, 2017: Business of AI)
Implications:
1. Humans know more than they can tell. We can't explain how we do certain things such as face recognition. Because of ML, we can automate tasks that we couldn't before.
2. ML systems are excellent learners -\< superhuman performance in activities such as detecting fraud and diagnosing diseases
Fields of improvement:
1. Area of perception & cognition: voice and image recognition
2. Complex problem solving: playing poker or establishing efficiency gains
159
Explain Polanyi's Paradox and the risks of overcoming them
(Brynjolfsson & McAfee, 2017: Business of AI)
* Polanyi's Paradox exemplifies that previously coding meant transferring knowledge from deverlopers' heads into a form that machines would understand and enable them to execute it
* This implies that we can not include knowledge which we can't fully explain (we know \> we tell)
* Machien Learning is overcoming those limits
Risks
* Machine may have hidden biases derived from training data
* Neural network systems deal with statistical instead of literal truths. This lack of verifiability will be a concern when it comes to dangerous applications, such as controlling a nuclear power plant
* Difficult to find out what went wrong as the underlying structure can be really complex
Appropriate benchmark is not perfection, since humans make mistakes as well and can not explain why.
160
# Define reinforcement learning
(Brynjolfsson & McAfee, 2017: Business of AI)
* The programmer specifies current state of the system, the goal, lists allowable actions and describes the elements of the environment that constrain the outcomes for each of those actions.
* A reinforcement learning system will optimize for the goal you explicitly reward, but not necessarily the goal you really care about.
161
Hub firms like Google, Amazon, Apple, Facebook and Microsoft are causing competitive bottlenecks and thereby creating monopolies. Exemplify.
(Iansiti & Lakhani, 2018: Hub economy)
Competitive bottlenecks like Google's Android as an OS
162
Explain how hub firms compete.
(Iansiti & Lakhani, 2018: Hub economy)
They take network-based assets that already reached scale in one setting, and then use them to enter another industry and "re-architect" its competitive structure -\> transform it from product-driven to network-driven.
Example:
Google's automotive strategy leverages technologies and data advantages to change the structure of the industry instead of improving the car itself
163
In which principles are economic hubs rooted
(Iansiti & Lakhani, 2018: Managing hub economy)
1. Moore's Law: computer processing power will double approximately every two years
2. Metcalfe's law: network's value increases with the number of connection points or users (network effects)
3. Network theory: digital networks naturally lead to the emergence of positive feedback loops that create increaslingly important, highly connected hubs.
164
Explain competition in a traditional product industry vs digital platform
(Iansiti & Lakhani, 2018: Managing hub economy)
Decreasing returns to scale (marginal revenue goes to zero) with several players
VS
increasing returns to scale with a narrow set of players (digital platform -\> network effects)
165
# Define multihoming
(Iansiti & Lakhani, 2018: Managing hub economy)
A practice enabling participants on one hub's ecosystem to easily join another, like Sonos with integrating Apple Music, Spotify, Pandora, Tidal etc.
This can significantly mitigate the rise of hub power
166
In which tree main areas do digital business model challenge physical models
(Weill & Woerner, 2013: Optimizing digital business model)
1. Internal power: CX often changes from product groups to unit that manages the multiproduct CX
2. Business processes: omnichannel focus
3. Customer data: becoming an enterprise-wide resource rather than resource hidden in one area
167
# Define deep customer-driven innovation
(Weill & Woerner, 2013: Optimizing digital business model)
Hiring anthropologists to identify unmet needs instead of surveys and market research
168
In which two areas do companies need to develop new capabilities
(Weil & Woerner, 2013: Thriving in a digital ecosystem)
1. Learn more about customer
- Learn more about customer's goal and life events
- Emphasize evidence-based decision making
- Develop integrated, multiproduct channel customer experience
2. Become more of an ecoystem
- Become first choise in your space
- Build great partnerships
- Treat efficiency and compliance as a competence
169
Why is Uber not disruptive according to the Disruptive Innovation Theory?
(Christensen, Raynor & McDonald, 2015: Disruptive innovation)
1. Disruptive innovations originate in low-end or new markets that incumbents overlook. They find a way to turn non-consumers into consumers.
- Uber has gone build a position in mainstream market first and subsequently appealing to overlooked segments
2. Disruptive innovations don't catch mainstream customers till quality rises. Otherwise they are "sustaining innovations"; better TV, fifth razor blade
- Ubers innovations are sustaining: better CX, lower price
-\> Uber SELECT does disrupt the limousine car industry, since it focuses on low end market which want to sacrifice a measure of convenience for monetary savings.
170
Name four important points regarding disruptive innovation:
(Christensen, Raynor & McDonald, 2015: Disruptive innovation)
1. Disruption is a process
2. Disrupters build different business models
3. Some disruptive innovations do not succeed (disruption does not equal success)
4. Do not overreact and dismantle your profitable business if disruption is looming, but strengthen relationships with core customers by investing in sustaining innovations and build a seperate unit which focuses solely on growth opportunities that arise from the disruption
171
# Define 'first-copy costs'
(Shapiro & Varian, n.d.: Pricing information)
The first copy is very expensive, but after the first one has been produced it becomes very cheap to produce more (encyclopedia á 1600$ vs CD á $50)
172
Explain competitive pricing strategy in information markets
(Shapiro & Varian, n.d.: Pricing information)
The story of CD phone books is classical: once several firms have sunk the costs necessary to create the product (e.g. a CD or rail line), competitive forces tend to move the price toward **marginal cost**, the cost of producing an additional copy. In other words, competition among sellers of commodity information pushes prices to zero. Because, when the **sunk costs** have been sunk, there is no natural floor to the price except the cost of producing and distributing another product.
173
How to discourage entry in an information category:
(Shapiro & Varian, n.d.: Pricing information)
1. Avoid greed: sacrifice short-term margin to make the market less attractive to would-be entrants to invest the sunk costs necessary to enter your market = limit pricing
2. Play tough: find a way to send a signal that entry will be met with aggressive pricing -\> threat of price cuts (brand value is not everything)
174
How to fight a price war caused by customer issues (price sensitivity), company issues (cost or strategic positioning), competitor issues (rival cost or strategic positioning) or contributor issues (resellers, suppliers, allies or government)?
(Rao et al., 2000: How to Fight a price war)
1. Take inventory
2. Stop the war before ti starts
3. Respond with non-price actions -\> quality, risk of poor quality or seek help from contributors
4. Modify selected prices (fighting brands, bundling & loyalty program)
5. Fight it out
6. Retreat and cede some market share
175
Explain the downside of the cost structure of information goods
(Shapiro et al., 1998: Versioning)
Because the fixed costs are both large and sunk, companies that don’t enjoy market dominance can be caught in devastating price wars. If competition forces a company to reduce its prices to a level near marginal cost, that company will not be able to win back their big investments.
176
Explain the logic of free version
(Shapiro et al., 1998: Versioning)
1. Low marginal cost menas that there is almost no investment to give information away
2. Information is an **experience good**, customers don't know what it's worth until they've actually tried it -\> easy to test
177
Name 7 versions of versioning
(Shapiro et al., 1998: Versioning)
1. Convenience
2. Speed
3. User interface
4. Support
5. Resolution
6. Annoyance (ads)
7. Comprehensiveness: the extent to which information offers a depth of detail
178
Name the three shifts in the move from pipeline to platform
(Van Alstyne, Parker & Chourdary, 2016: Pipelines, platforms)
1. **From resource control to resource orchestration:** the _resource-based view of competition_ holds that firms gain advantage by controlling scarce, valuable and inimitable assets. For platforms, the network of producers and consumers becomes the major asset as the assets that are hard to copy are the community and the resources its members own and contribute.
2. **From internal optimization to external interaction:** pipeline firms create value by optimizing an entire value chain, and platforms create value by facilitating interactions between external producers and consumers. Ecosystem governance is an essential skill.
3. **From a focus on customer value to a focus on ecosystem value:** pipelines want to maximize the lifetime value of individual customers and platforms maximize the total value of an expanding ecosystem in a circular, iterative, feedback-driven process.
179
Explain how network effects influence Porter's 5 forces
(Van Alstyne, Parker & Chourdary, 2016: Pipelines, platforms)
The five forces model doesn’t factor in network effects and the value they create. It regards external forces as “depletive” (decreasing value) and so argues for building barriers against them. In demand-side economies, however, **external forces can be “accretive”** (adding value). Thus, the power of suppliers and customers may be viewed as an asset on platforms. Understanding when external forces may either add or extract value in an ecosystem is central to platform strategy.
180
Explain 'permissionless innovation'
(Van Alstyne, Parker & Chourdary, 2016: Pipelines, platforms)
They let producers invent things for the platform without approval but guarantee the producers will share in the value created (e.g. app store). However, unfettered access can destroy value by creating “noise”, which is misbehaviour or excess or lowquality content that inhibits interaction.
181
# Define big data hubris
(Vidgen et al., 2017: Creating value from BA)
The often implicit assumption that big data is a substitute for, rather than a supplement to, traditional data collection and analysis.
182
Name the four dimensions of organizational change in becoming data-driven
(Vidgen et al., 2017: Creating value from BA)
1. Organization and management
2. People
3. Process
4. Technology
183
The top 3 challenges highlighted by the study represent key steps in the data analytics journey:
(Vidgen et al., 2017: Creating value from BA)
1. Data quality
- Driven by legacy systems and unclear BA strategy
2. Using analytics for improved decision making
3. Creating a big data and business analytics strategy.
184
Explain the RBV on the data analytics challenge
(Vidgen et al., 2017: Creating value from BA)
The resource-based view (RBV) theory proposes that firms are comprised of a set of resources or assets, including those related to data and analytics, that need to work collaboratively together to deliver capability around a given task. Those organizations that perfect this strategic approach will generate a rare, valuable, non-substitutable ability to leverage business value from big data analytics, thereby generating competitive advantage. Given this theory and empirical findings, the researches derived an integrative ecosystem shown in the figure
185
A company’s experimentation personnel can be organized in three ways:
(Kohavi & Thomke, 2017: Online experiments)
* **Centralized model:** in this approach a team of data scientists serve the entire company. The advantage is that they can focus on long-term projects, such as building better experimentation tools and developing more-advanced statistical algorithms. One major drawback is that the business units using the group may have different priorities, which could lead to conflicts over the allocation of resources and costs. Another con is that data scientists may feel like outsiders when dealing with the businesses and thus be less aware of the units’ goals and domain knowledge, which could make it harder for them to connect the dots and share relevant insights. Moreover, the data scientists may lack the power to persuade senior management to invest in building the necessary tools or to get corporate and business unit managers to trust the experiments’ results.
* **Decentralized model:** distributing data scientists throughout different business units. The benefit of this model is that the data scientists can become experts in each domain. The main disadvantage is the lack of a clear career path for these professionals, who also may not receive peer feedback and mentoring that help them develop. And experiments in individual units may not have the critical mass to justify building the required tools.
* **Center-of-excellence model:** a third option is to have some data scientists in a centralized function and others within the different business units. (Microsoft uses this approach.) A center of excellence focuses mostly on the design, execution, and analysis of controlled experiments. It significantly lowers the time and resources those tasks require by building a companywide experimentation platform and related tools. It can also spread best testing practices throughout the organization by hosting classes, labs, and conferences. The main disadvantages are a lack of clarity about what the center of excellence owns and what the product teams own, who should pay for hiring more data scientists when various units increase their experiments, and who is responsible for investments in alerts and checks that indicate results aren’t trustworthy.
186
The best data scientists are skeptics and follow Twyman’s law. Explain.
(Kohavi & Thomke, 2017: Online experiments)
Any figure that looks interesting or different is usually wrong.
187
Explain hetergeneous treatment effects
(Kohavi & Thomke, 2017: Online experiments)
In certain cases a single good or bad segment can skew the average enough to invalidate the overall results. This happened in one Microsoft experiment in which one segment, Internet Explorer 7 users, couldn’t click on the results of Bing searches because of a JavaScript bug, and the overall results, which were otherwise positive, turned negative. An experimentation platform should detect such unusual segments; if it doesn’t, experimenters looking at an average effect may dismiss a good idea as a bad one
188
Explain carryover effects
(Kohavi & Thomke, 2017: Online experiments)
Results may also be biased if companies reuse control and treatment populations from one experiment to another. People’s experience in an experiment alters their future behaviour.
To avoid this phenomenon, companies should “shuffle” users between experiments.
189
Explain why correlation is not causation and exemplify
(Kohavi & Thomke, 2017: Online experiments)
The following two examples illustrate why, and also highlight the shortcomings of experiments that lack control groups.
1. The first concerns two teams that conducted separate observational studies of two advanced features for Microsoft Office. Each concluded that the new feature it was assessing reduced attrition. In fact, almost any advanced feature will show such a correlation, because people who will try an advanced feature tend to be heavy users, and they tend to have lower attrition. So while a new advanced feature might be correlated with lower attrition, it doesn’t necessarily cause it. Office users who get error messages also have lower attrition, because they too tend to be heavy users. But does that mean that showing users more error messages will reduce attrition? Hardly.
2. The second example concerns a study Yahoo did to assess whether display ads for a brand, shown on Yahoo sites, could increase searches for the brand or related keywords. The observational part of the study estimated that the ads increased the number of searches by 871% to 1,198%. But when Yahoo ran a controlled experiment, the increase was only 5.4%. If not for the control, the company might have concluded that the ads had a huge impact and wouldn’t have realized that the increase in searches was due to other variables that changed during the observation period.
3. Including too many variables in tests also makes it hard to learn about causality. With such tests it’s difficult to disentangle results and interpret them. Ideally, an experiment should be simple enough that cause-and-effect relationships can be easily understood. Another downside of complex designs is that they make experiments much more vulnerable to bugs.
Clearly, **observational studies cannot establish causality.**
