Pitch: Point-to-Point

A smart marketplace for smarter travelers


According to Bond Brand Loyalty, there is $100 billion worth of unclaimed loyalty points. According to another loyalty research firm, Colloquy, about $48 billion worth of loyalty points and rewards are earned each year, of which approximately 33% will never be redeemed.

Many customers are unaware of their points balance or what their points can purchase. Hence points accumulate and expire, and customers lose incentives to stay loyal to brands. For companies, even though unredeemed points are a large source of frequent flyer program profits, research shows that customers who redeem points are more likely to display greater engagement and work harder to earn the next reward. A positive redemption experience can drive members to continue to spend with a brand.

While some customers might be interested in transferring their points or redeeming them on behalf of others, there is currently an institutional void borne out of what economists refer to as the hold-up problem and a lack of trust between any buyers and sellers.



To fix the problem of unused points balances, we create a two-sided marketplace where “sellers” can post how many points they have for which program, and “buyers” can bid on the points they want. Once the auction closes to the highest bidder, funds will be held by the platform in escrow, and the seller will then utilize the points according to the buyer’s wishes (ex: United flight from NY to SF). Upon the buyer’s receipt, funds will flow to the seller.

In this first iteration, we enable sellers to book travel for buyers. Most current loyalty programs do not allow for free transfer of points between individuals; thus, until this program changes, buyers and sellers will need to communicate to ensure that sellers book itineraries correctly. Although there is nothing to prevent customers from doing this already, our platform provides value both by improving discoverability of these opportunities and solving the problem of buyer and seller trust by holding payment in escrow until delivery is verified.

Over time as more loyalty programs enable free transfer of points, our marketplace can accommodate more situations. For example, for sellers with too few points to be of much use, we can allow them to auction their points directly. Buyers can accumulate points from multiple sellers and earn otherwise inaccessible rewards. Eventually, with enough liquidity, our platform can add further value by matching blocks of points to satisfy arbitrary trades, in the same way securities exchanges match bids and asks.

As Point-to-Point grows, it will not only benefit from cross-side network effects, but also open up additional product and business opportunities. As the platform collects data from transactions, we can build a recommendation engine to highlight deals for specific customers. We will also utilize algorithms to educate sellers on points required to cross more valuable thresholds. For example, if a seller had 1489 points with Chase, the algorithm could list shops that the seller would be interested in that could also yield a high return point value, enabling that seller to cross the threshold and sell their 1500+ points for a higher amount. Our platform will also collect relevant data outside the marketplace (ex: new credit card bonuses, 5x points on certain categories, etc.) and distribute this information to members so that they can have a one-stop shop for all loyalty point programs. This will induce stores to participate in our program, as our algorithms can enable higher spending and more efficient advertising.

We intend to monetize through a commission-based revenue model. For trades in which points are paid for in cash, Point-to-Point would collect a 3-5% fee on the transaction amount. For trades in which points from one program provider is exchanged for those of another, Point-to-Point will not collect any commission. This incentivizes more buyers (and therefore, sellers) to use the platform, creating benefit for all.



Our initial pilot sets up partnerships with loyalty programs that have a wide network of points and already have transfer systems in place, such as Starwood Preferred Guest, Chase Ultimate Rewards, or American Express Membership Rewards. These points can already be redeemed in a variety of settings (hotels, flights, and online retailers) and can sometimes be shared between household members.

We can market Point-to-Point to the popular points forums, including The Points Guy and Reddit, who in general capture a market that is more likely to be motivated by points programs. In line with what Eisenmann, Parker, and Van Alstyne recommend in their article “Strategies for Two-Sided Markets,” we would subsidize early adoption and adoption by price-sensitive users, and work to secure “marquee” users’ exclusive participation with loyalty programs. To incentivize early adopters, we can subsidize transactions for buyers and provide bonuses to sellers.

Furthermore, we will induce further loyalty through discount referrals. To ensure that new entrants do not steal market share, we would form temporary exclusive contracts with participating programs as well as offer our own form of points (earn points to redeem for percent discounts on trades). Over time, we will expand to more loyalty programs, offering an enormous selection for users.



To ensure our solution meets its objective of creating a liquid market for rewards points while benefiting both points programs and users, we would focus on empirical measures of market liquidity, such as those proposed by Gabrielsen, Marzo, and Zagaglia like liquidity ratio, turnover ratio, and variance ratios. We could also measure the following product success metrics:

  • Number of Sign-Ups
  • Retention Rates (e.g., 7-day and 28-day return rates, DAU, WAU, MAU)
  • Usage Per Member (e.g., # of trades and volume executed)
  • # of Referrals
  • Increase in Engagement with Loyalty Programs



Although loyalty programs have become a differentiating feature, allowing companies to win repeat customers and gain market share, the concept of a secondary market where buyers and sellers are matched based on factors, such as  willingness to pay, is still relatively new. Currently, Giift is a loyalty programs and gift card network, allowing consumers to track and exchange points from program issuers (airlines, hotels, retailers, etc.) with one another. However, the “exchange rate” is dictated by Giift (not what the market will bear) and given that only points are exchanging hands, not cash, this limits what a consumer would be able to receive and thus, overall trading activity.

The only other company taking a similar approach of creating a two-sided market platform is Digital Bits. By tokenizing points accumulated with an airline and allowing customers to redeem them for cash or rewards at another rewards provider of their choice, Digital Bits is creating a blockchain-based market and along with it, more flexibility for users. We feel that our solution can successfully compete with Digital Bits because it facilitates communication between buyer and seller without requiring integration with a brand’s existing app or interface. We are also skeptical of the incremental value a blockchain framework provides relative to the additional usage and development complexity it requires. As such, we feel our first-mover advantage and initial user base can perpetuate adoption on both sides of the market.










Eisenmann, Thomas, Geoffrey Parker, and Marshall W. Van Alstyne. “Strategies for two-sided markets.” Harvard Business Review 84, no. 10 (2006): 92.

Gabrielsen, Alexandros, Massimiliano Marzo, and Paolo Zagaglia. “Measuring market liquidity: An introductory survey.” (2011).



Siddhant Dube

Eileen Feng

Nathan Stornetta

Tiffany Ho

Christina Xiong

Hidalgo: Buy Better, Buy Smarter


Our software application will provide prospective car buyers with a better informed and more cost efficient method of selecting the perfect vehicle for them. Hidalgo is a customer profiling platform that partners with insurance companies in order to determine what specific vehicle a consumer should seek to purchase by taking into account tracked data points that insurance companies collect on their insured customers. Given annual vehicle sales of $17M in 2016, there is a huge opportunity for Hidalgo to come in and remove some of the friction involved in the current sales process. Many individuals do not have a strong foundation of knowledge with which to purchase vehicles. As such, many individuals may be buying cars that do not fit their driving style or make emotion based purchase decisions. By removing some of the friction of uninformed purchasing, we hope to boost sales for manufacturers, save costs for buyers, and reduce claims for insurance companies.


By taking the data points such as average driving speed, make and model of car, and braking behavior, we can build a large database of driver types and vehicles. For example, if an individual fits the category of a hard braker who likes to drive fast but never has an accident, Hidalgo would recommend a high performance car that fits this specific profile. It would not recommend a slow Volvo. This match process also allows the consumer to save on insurance when they purchase a car that the  insurance company has recommended, a car that matches the individuals driving style leading to safer driving. The more a car fits an individual’s driving style, the less surprises there are while driving resulting in less accidents. Less accidents means less claims processed by insurance companies, allowing them to save substantially on claims processing. The increased saving is then passed partially on to the consumer reducing their car insurance payments.

Effectiveness, Commercial Promise, and Competition:

For individuals who are price sensitive, the promise of lower insurance payments on their car can greatly impact their car purchasing decisions. Additional savings will result from reduced repair and maintenance costs as individuals’ driving style will better match the vehicles physical components. Individuals will also be able to create a profile and add in their vehicle and style preferences allowing them to choose without feeling like they’re being forced to buy a car.

There is quite a bit of commercial promise due to the large size of  the car buying and insurance industry. Insurance companies can maximize their returns by having to spend less on claims processing, which frees up capital to be deployed elsewhere. Car manufacturers benefit by producing cars that better match the buyers in their target markets and segments and boost sales. Hidalgo will earn revenue by taking a portion of claims processing savings as well as a portion of referral fees from a manufacturer when an individual buys a car through our platform.

The platform’s main competitor will be in house data teams of each insurance company. We believe we can avoid competition through the advantage of data collection across different insurance companies and standardization of data within the Hidalgo platform. This allows insurance companies to reduce their own tech development costs while having access to a larger set of data, making the recommendation software smarter and more accurate.

Suggestions / Improvements:

  • Improved accuracy of recommendations over time as data set grows
  • Improved connectivity so user accounts can be linked via bluetooth to onboard computers within any car
  • Partner with car producers to create more sensor points on a car that can be used to inform buyers’ decisions







Team Members:

Pavlina Plasilova, Kelly de Klerk, Yuxiao Zhang, Aziz Munir, Megan McDonald



Travel and tourism is a huge sector, with direct contributions of $2.57 trillion to the global economy in 2017 and total contributions of $8.27 trillion [1]. Travel agencies (including companies such as Expedia and Priceline) generated $40.5 billion dollars in 2017, with annual growth of 5.3% from 2012 to 2017 [2]. Given the size of this industry, even small improvements in helping consumers decide on and book travel arrangements would have a huge impact. Many consumers are looking for more tailored and personalized recommendations. Yet as the amount of options and information on travel increases, it becomes ever more time consuming to sift through all of the possibilities to find the unique and perfect fit for any given individual. This creates a large opportunity to go beyond the traditional and old-fashioned ranking systems.



Duffle is a response to the increasing demand for unique experiences. It facilitates travel discovery and booking by analyzing personal preferences (stated and based on past reviews), airline, restaurant, car rental, tourist activities, weather, and social media data. AI-powered chatbots consider budget, frequent flyer and other reward membership information, location, and availability. Users are able to swipe left or right trip and activities recommendations. The algorithm learns over time from the user response to recommendations and can improve them when users input personal information such as pet or kids joining for the trip. Fellow travelers can also be added to a trip so Duffle and not the main user is the Q&A point for all people on the trip. For solo travelers, Duffle can connect users to other travelers considering the same destinations and dates.


Once a trip is booked, Duffle compiles a list of recommended attractions or activities and historical information about sites the user can visit. The service includes automatic check-in, 24/7 travel assistance app, and on the ground guide.


Duffle assists users past the initial booking by providing packing list recommendations, local sim card options, or advice on recommended vaccinations. It can also alert to security concerns or unfavorable weather changes and offer quick assistance with rebooking. The platform benefits from network effects on the client side since the algorithm learns from the likes and dislikes of the specific user and those of other similar users and can connect travelers with similar interests, budget, and desired travel dates.



Our pilot would require the initial aggregation of source data needed for the curation algorithm. Thus, we would begin to create the network effects within the university format at the University of Chicago or University of Illinois at Chicago (or any comparable university) that have a population of active social media participants that engage with rich multimedia content which can provide travel preference data. We would initially pool this data to create curated content which our data scientists and travel agent partners would initially evaluate for relevance and feasibility. After 2-3 months of the aggregation process, we would look to merge the data and algorithmic learnings to create a machine learning software that can input new data sets, curate existing ones, and accept continuous inputs that will allow it to learn continuously.


Post this initial data collection and curation period, we will engage our university volunteers in person and present them our recommendations. We will conduct online surveys and in-person interviews to gauge their WTP for the trips recommended to them and their relative interest levels. Additionally, we will compare our recommendations vs those of our competitor set (see below), using the data we collect on the student’s interests, geography, and finances. We would also have them rate their interest levels relative to our recommendations and those of competitors on a blind basis.


While the student population is ideal to begin with, they have limited income levels and travel abilities. We would lastly roll out this final pilot stage to all members of the university (faculty, staff, etc.) in order to maintain a comprehensive set of age and income ranges for our algorithm that encompass all market segments. We will look for our pilot to confirm/negate our hypothesis that our recommendation platform with be 1) more robust and comprehensive than competitors, 2) more thorough in its presentation, 3) provide more value for each budget, 4) have a higher probability of booking than competing platforms.



Currently, there are various companies offering services that Duffle will offer, but none that have successfully combined the data that we plan to combine, and utilized artificial intelligence to optimize the service. Hopper, a discount flight platform, has launched an “AI driven travel agent” that allows users to input preferences about their ideal trip (i.e. duration, time of year) and get recommended flight deals. The platform learns from which deals the user does or does not accept. We plan to utilize AI in a similar way, but through connecting a user’s social media and allowing for more preference information, can offer a more customized result, beyond just flight deals. Mezi, a corporate travel website acquired by American Express, uses AI to learn about a traveller with every additional trip – when they book, what type of ticket they prefer, etc, and provides a platform for booking and managing work travel. Our goal is to provide a similar platform for booking and managing consumer travel and combining it with recommendations for activities, hotels, etc while on the vacation.  


While feasibility of the product is reasonable given we have seen companies utilize different aspects of the technology, and are just looking to aggregate these features, there are risks associated with Duffle. Travel is a popular industry with many companies looking to disrupt or own the booking and planning process. Given that, we will need to rely on network effects and push marketing while continuing to innovate and improve the product.



[1] World Travel and Tourism Council. Total Contribution of Travel and Tourism to the Global Economy from 2006 to 2017. Reported through Statista.


[2] IBIS World Industry Report. Travel Agencies in the US. September 2017.



Team Members:

Sam Steiny

Rosie Newman

Gergana Kostadinova

Javier Rodriguez

Personalized travel app

New Opportunity:

International spending by Chinese tourists has skyrocketed in the past 8 years, showing increases of over 150% from $38B to $261B with the growth of the Chinese middle class.  The tourism market is currently very fragmented, as would be expected due to the number of countries and people involved, there is an opportunity to integrate. Traditionally, Chinese tourists travel with their families or through travel agencies that offer group tours. Due to the data capabilities and market share of major Chinese companies such as Tencent and Baidu, we believe there is a chance for them to quickly disrupt the tourism market by offering personalised recommendations and trip planners using prediction models as well as matching algorithms.  


AI Solution:

AI and machine learning techniques will be used to create a powerful trip planner and recommendation tool that is personalized to individuals. Building off the kind of matching techniques that are used by platforms such as say, Netflix, machine learning can play a large part in the recommendations section. The AI solutions will be powered by continuous real-time data through surveys, reviews and location tracking. Therefore, the application will identify travel-companions and restaurant recommendations by overlaying these techniques on top of traditional filtration tools specified by the users. The solution required must be able to combine different kinds of data types, for example users may have an image of a tourist spot with a comment about how much they loved it, therefore the company would need to apply image recognition techniques. This could be combined with multiple searches in the week prior to travelling on the Baidu search engine, for best restaurants in a certain district of a city. Thus, combining and profiling using such data would create an optimal route and itinerary for the traveller.


Design of Demonstration:


To understand the commercial promise, we must focus on the pain points of customers and see how the product addresses these. One of them is logistics and excessive paperwork to track all kinds of reservations, compounded for families. With the centralized dashboard (pictured below), these can be compiled in one place, and this can be done through automatically pulling data from e-mail and other communications. A proof-of-concept has already been offered by Google on its Trips app.

Another pain point can often be the recommendation system. Younger travellers often want to skip the touristy spots and go somewhere authentic. This is often not possible due to the language barrier, but by utilizing crowdsourced recommendations (already a part of these companies’ infrastructure) and filters, it will be easy to save time and money finding suitable places.


We have talked about many features, so a pilot would focus on rolling out some core features in a few cities. Say, we pick Paris, a popular tourist destination with a lot to see and a language barrier for most. We would beta test with young travellers going to Paris, and initially propose itineraries and recommendations based heavily on crowd-sourced data and location tracking. In this service, the main user is not the main source of revenue, and thus we would have to focus on user experience. As more users are attracted to the platform, there would be various ways to monetize, the most popular of which we foresee as being a B2B model, where businesses are charged for giving deals to travellers and advertising on the platform. Moreover, as travellers take multiple trips, machine learning techniques can be employed more heavily to solve prediction problems related to planning itineraries and suggesting establishments.




10 Most Popular Social Media Sites in China (2018 Updated)



Google launches a personalized travel planner, Google Trips


Team Members:

Mohammed Alrabiah

Tuneer De

Mikhail Uvarov

Colin Ambler

Lindsay Hanson

Profile on the National Residency Match Program


In 1952, the National Residency Program (NRMP) was established as a non-profit organization to match medical students and residency programs. However, by the 1990s, the system was under serious strain. With thousands of students attempting to match across dozens of specialties and thousands of residency programs, there was concern that hospital preference was favored over that of applicants’. Further, more applicants were trying to match as a couple, something the current system could not handle, causing students to opt out of the process. The NRMP hired Al Roth to design an algorithm that would lead to more desirable outcomes and encourage applicants to opt into the process.


Today, the NRMP uses Roth’s adaptation of the Gale-Shapley algorithm to produce “stable” matches, that ultimately favor applicants. It considers their preferences first, then the hospitals’. It also can do joint matching for couples. Beyond residency programs, this algorithm has applications from matching students to public high schools in NYC and to organ donor matching. In 2012, Alvin Roth and Lloyd Shapely won a Nobel Prize in economics for the algorithm’s application across markets that require choice from both sides of the market, and where price is not a factor.

The algorithm works by having both students and hospitals rank their preferences after an interview process. Both applicants and hospitals simultaneously submit their preferences in rank order. Then, the algorithm considers the first student and determines if that student has ranked a given hospital, if not, it moves on. If she has ranked the hospital, then the algorithm determines if the hospital has ranked the student. If so, then they are “tentatively matched”. The algorithm goes to the second student, looking at the same hospital. If it is a mutual match, then it determines 1) if the hospital has a spot remaining and 2) if the ranking is higher or lower than the previous student already “tentatively matched”. If higher than the previous match, then the previous match is demoted. If not, and all spots are filled, then the algorithm considers the student’s second choice. This process occurs across dozens of specialties, thousands of global applicants and thousands of residency positions.


In 2017 Match, 30,478 students matched, or ~71% of registrants, an all-time high. Unfilled positions were placed in the Match Week Supplemental Offer and Acceptance Program, ultimately leading to a 99.4% fill rate. Couples are increasingly satisfied with the new algorithm – 95.4% matched in 2017. The previous methodology would not have been able to handle this volume and level of complexity at this speed, about 17 seconds. Literature exists to refine the algorithm, but there are no commercially viable alternatives as most agree with the “stable marriage” approach to this process. Considering how adaptable the algorithm is, it is unlikely any competing approach will exist in the short-term.


The process is imperfect. Both students and hospitals attempt to influence the other party’s rankings by over-embellishing their true preferences. Trust in the process and concerns about whether applicant preferences are taken into account remain.

To maintain trust, the NRPM is not currently changing the algorithm. However, there are other methods of engendering trust – such as continuing data transparency. In not adjusting the algorithm, the NRMP loses any opportunity to improve upon it. Significant time and money is also spent in the interview process itself.

This algorithm could be expanded to other industries that have many companies with similar offerings targeting a specific pool of candidates. Example industries are investment banking, consulting, and law. This process could also expand globally, for example NRMP has licensed the algorithm to the Canadian Resident Matching Service. It also could be used for other professions such as matching law students to firms, though Dr. Roth indicated this only would work if there was a demand for a market to be created.

This type of algorithm could be used in any two-sided market with uncomplicated preferences (firms or people or a combination of both) and where money is not a deciding factor. Evolving the algorithm to handle complicated preferences is necessary to further commercialize to other markets where price is not a reasonable arbiter to set supply and demand.












Derivative Risk Control – Augmented Judgment


Financial Broker-Dealers such as Goldman Sachs, JP Morgan, and Morgan Stanley are constantly monitoring their trading activity for errors, particularly in the Over-the-Counter or OTC market, where trades are done off of exchanges and are bilateral agreements between parties with bespoke terms.

Simple, “vanilla” products such as Calls and Puts may have as many as 15 relevant fields, but structured products and exotic derivatives may have upwards of 100. Each field is negotiated and confirmed with counterparties on the “street” in a multi-level proofreading and verification process that can take anywhere from three days to two weeks or more. Errors on the direction of a trade, the underlying asset a derivative references, the size, strike, or expiration date are all relevant economic factors that can significantly damage the profit margin, reputation, and volume capacity of a given securities business.

It is in the interest of Investment Banks to mitigate the risk they incur within this OTC market by intelligently sensing where potential risk may lie and devoting resources accordingly. However, the current model involves visual, manual reconciliations performed by specialists on an ad hoc, continual basis with each trade, and field, reviewed as it comes.

 Process for Confirming and Controlling OTC Trades Currently:

Generic Sample OTC Option Ticket:


Commercial promise and challenges:

According to the Bank of International Settlements (BIS) the outstanding notional amount of OTC derivatives exceeded $416 Trillion globally, with a gross market value of $13 Trillion.

Utilizing these assumptions as well as error rates and sizes provided by conversations with industry experts we create an estimated total addressable market and assign a value to the potential industry savings annually per incremental improvement in error reduction:

Number of Major Broker-Dealers 25
Average Transactions 100
Market Transactions / Day: 2,500.00
Yearly Transactions: 625,000.00
Gross Market Value: $   13,000,000,000,000.00
Error Rate: 3.0%
Average Value of Error: $                              5,000.00
Annual Error Cost: $                   93,750,000.00

Hence, a reduction in the error rate by 1% and error size by $1,000 would yield annual savings of approximately ~$44mm, whereas a 2% reduction and $3,000 reduction saves ~$80mm. Thus, even a small market penetration rate with moderate success could be significant.

The principal challenge would be in getting Investment Banks, which closely guard their internal affairs, to be willing to integrate an outside vendor’s technology in their operations without having direct control over that entity and in addition having that entity service other banks. Furthermore, the usage of this technology to reduce error rates in an opaque and poorly-understood market, which is the reason for the opportunity in the first place, does not lend itself well to clear-cut spending decisions. Therefore getting buy-in for COO’s and CTO’s will be a lengthy process and require imaginative data visualizations and proof of data security.


We face competition on two fronts, internal to Broker-Dealers and externally in the form of “RegTech” companies seeking to provide tailored AI Solutions. Internally, there are significant resources available to banks within their IT departments, however, the deployment of those resources to such targeted AI solutions is limited at current due to the intense product-specific knowledge requirements that are often siloed. Externally, there are numerous RegTech companies which primarily focus on compliance rather than financial risk. AQMetrics and Ancoa, which received the 2016 Operational Risk Award, both are focused on automated auditing of transactional data and real-time trader behavior surveillance. These companies thus pose the greatest direct threat.

Proposed alteration:

Model Design:

  1. Data from email, chat, firm booking systems, issue trackers, integrated in a pipeline to create a dataset for analysis. Historical testing and scenario analysis used to create weightings of risk for fields/attributes.
  2. Using Supervised Learning we will use cross-validation and other techniques to tune hyperparameters on predictive indicators of the risk a particular trade has
  3. We believe this should be constructed as follows:

Risk (1-100) = F(Trader Behavior, Objective Risk)

Trader Behavior = F(Historical Error Rate, Estimated Error Rate)

Estimated Error Rate = F(Trade Complexity, Number of Trades, Time of Day, Experience Level, etc)

  1. Issues will be prioritized based on the risk construct and highlighted for further investigation


The model will create informative risk parameters for each trade that will be incorporated into the prioritization of the item either in terms of rigor of checks (can be automated via email/UI interface) or legacy human-performed checks. Large-significance errors will be disproportionately targeted by the model’s tuning. Long-term, the risk identification process will precede trade booking and help establish a more rigorous booking process at the origin. This will skew control processes to reduce error size and frequency measurably.









Mean Error Size and Error rate estimated based on conversations with Operations Professionals from Goldman Sachs and UBS Securities

Feedback App

Feedback logo



Feedback is a company based in Toronto, CA, that aims to reduce food waste and convert sunk costs to incremental profits for restaurants.  Restaurants have always struggled with minimizing their food waste. Although many have employed different front-end planning methodologies to try to accurately forecast their demand and keep waste low, many still struggle to responsibly dispose of unused cooked food or soon-to-expire ingredients at the end of the day.  Though many choose to donate their some of their remaining food to the homeless and hungry, most non-bread food products are still wasted. The Feedback app gives restaurants the opportunity to sell cooked food that would otherwise be wasted at discounted prices during off-peak hours.

Feedback App



The Feedback app is a platform for restaurants to offer customers a selection from their menu at discounted prices during hours of their choice.  Although this is primarily intended for restaurants to sell food that would be otherwise wasted at off-peak hours (i.e. between lunch and dinner or between during late night), restaurant users can choose to offer discounted food at any time (if they wanted to use it as a marketing tool for example).  Hungry customers are able to select from a variety of menu options in their area that fit their preferences for cuisine, price and time frame.

Feedback’s technology does more than pair up restaurants with restaurant-goers.  The application uses algorithms to intelligently determine pricing of the menu products depending on time of day, location, the product itself, the quality of the restaurant, the shelf-life of the product and the parameters set by the restaurant itself.  Although restaurants have the ultimate say in what the pricing actually is, Feedback utilizes the data of all the restaurants in its user base plus additional data from restaurants in the area though scraping menus on websites to provide an accurate pricing recommendation.  As the best pricing off-peak pricing data comes from other users of the platform, Feedback is strongly subjected to network effects.

Effectiveness, Commercial Promise, and Competition:

Overall, we are not sure about their effectiveness as they are still limited to only the Toronto area.  They have signed up over 7,500 users in Toronto alone with over 1,000 meals sold within its first month of operation (October 2017). [1] While this opportunity does help make additional revenue on food that would have gone to waste, restaurant margins are quite low (~ 6% [2]).  So, there is not a lot of margin to be captured by a technology company like Feedback.

There are several companies trying to reduce food surplus through technology, either by providing discounts direct to consumers or by sending surplus direct to organization in need, mostly nonprofits.  For example, Gebni, a NY-based company that has raised $250,000 in funding [3], provides automatic smart menu pricing for its users to reflect real-time demand for that specific item.

Copia, another competitor, allows restaurants to schedule pickups for surplus food.  They then match the surplus, based on the amount and food type, to a local nonprofit.  They also provide opportunities to make better buying decisions and help organization get proper tax deductions.

Another competitive force is better trained staff and improved internal processes to help restaurants to better predict supply and demand for their menu items. That said, there is likely to be some element of unpredictability of supply and demand irrespective of these kinds of improvements.


Suggestions / Improvements / Risks:

To better position Feedback for commercial success, the company should consider implementing the following:

  • Delivery integration to reach more customers.
  • Consider changing brand name to something catchier that’s not a generic word to avoid crowding search results and to make branding easier.
  • Because this business model already has small profit margins, consider limiting or discontinuing donations to non-profit food pantries as the company scales.    
  • Offer promotions to certain restaurants (e.g., new partners) to display their menus more prominently in app
  • Clarify in app how fresh the food item is, to mitigate consumer concerns about freshness

Feedback should also improve its algorithm to

  • Better match discounted pricing with maximum, real-time consumer willingness to pay during off-peak hours in order to extract most value for restaurant partners
  • Identify better restaurant options based on location, consumer preferences, consumer purchase history
  • Narrow pick-up window based on type of food (e.g., pizza can be picked up within 30-60 mins, dozen bagels with 2-4 hours)


  • Depending on the target market of the restaurant, the Feedback could open its clients to the risk of cannibalization.  Customers who would normally eat during peak hours at full price could be converted to off-peak users.



[1] https://www.feedbackapp.ca/single-post/2017/11/01/What-A-Month-It-Has-Been

[2] http://smallbusiness.chron.com/average-profit-margin-restaurant-13477.html  

[3] https://www.crunchbase.com/organization/gebni#section-overview






As of 2017, indoor rock climbing in the US was a $402.9MM industry by revenue, with an annual growth rate of 3.9%. This is an industry that sees growth with the growth of disposable income, and sees its growth driven primarily by younger consumers. Consumers age 35 and below comprise 83.6% of the market, with 47% of the market aged 25 to 34.

Despite this growth, safety concerns often dissuade new climbers. Additionally, high costs in the form of the climbing gym needing liability insurance and knowledgeable staff to train and keep an eye on the climbers, increases the barriers to entry in the market.

Further, to keep climbers continuously engaged, climbing gyms employ route setters – people who redesign the patterns of the climbing holds every few weeks. To communicate the level of difficulty of a route, routes are graded. While most indoor gyms are two standard grading scales – V Scale and Font Scale – grading is very subjective and there can be a high level of variance of how routes are graded across gyms. Further, individual attributes of the climber – such as height and weight distribution (men and women for instance, may have different weight distributions which will cause them to approach a climb differently) can make routes more difficult for some people than others. While there are six main forms of rock climbing holds, there can be a high level of variation on the main forms.


We believe an algorithm can be developed, factoring in types of holds, height and pitch of the climb, attributes of the climber, such as height and weight, to create new rock climbing routes with personalized ratings. Using artificial judgment in this way will create a more objective standard of route rating that will allow climbers to practice safely. Gyms across the country could then communicate a standardized definition of route difficulty, accessing a database of infinite climbing routes. We shall imbed the algorithm into a mobile app we named PocketClimber, both after the name of a rock climbing hold (Pocket) and the ease of using the app for setting new climbing routes.

Our solution reduces the time and cost for route setting. It will reduce the dead time between climbers as well as likelihood of injuries due to route-setting judgement error. Additionally, having routes analyzed by computers according to a person’s fitness and physical characteristics (such as height and weight) may make people feel more safe, thereby attracting new people to the sport.

Commercial value and promise

Our solution targets a rapidly growing international market and is scalable, as it can be used by individual climbers as well as large gym chains, such as Gold Gyms. PocketClimber can establish the standard for climbing and be adopted by all accredited gyms. We will design a prototype app to demonstrate the potential capabilities of the PocketClimber, pitching it to investors such as GoPro founder Nick Woodman who have a vested interest in adventure and sports. Live demonstrations will prove the viability of our product. Currently, no customized solution exists in the market. The closest is the app Vertical Life, which provides standard guides to climbing.


We will start our pilot program with the demonstration of a prototype app to climbing gyms across Chicago and get feedback from climbers from our network and from the gyms. Additionally, we plan on hosting events around the app – such as human vs AI setters, and different climbing competitions sparking an interest in the app and in the concept. Climbers may be intrigued by the unusual routes designed by AI, so there could be an event-based opportunity here. After we collect enough data, both from the human setters and the climbers we will be able to refine our algorithm and demonstrate to investors the advantages of using PocketClimber with optimized routes and less accidents for beginner climbers.



  •      Wall scanner, some features of a climbing wall are not changeable. The app uses the phone camera to assess inherent difficulty of the wall
  •      Hold Inventory:  Route setters can input the types of holds they have and how many holds
  •      RouteBuilder: You can specify the level of difficulty you want, and RouteBuilder will make a route for you. Additionally, could provide further insight such as range of difficulty climbers will face, so that routesetters are taking diverse body types into account when setting  
  •      RouteFinder: On the other side of the market, consumers could use the app as well to input personal metrics such as height in order to get personalized gym and route recommendations




Fun Run


The global athletic footwear market size was valued at $64.3 billion in 2017 and the use of fitness apps has grown by 330% in the last 3 years. With fitness becoming an increasing part of a young professional’s life, finding ways to improve the efficiency of a workout or prevent injury are key to increased fitness success.


Fun Run uses machine learning and sensors embedded in running shoes to track particular metrics and improve a runner/walker’s stride. In addition to tracking steps, heart rate (in the foot) and typical fitness tracking metrics that are currently available in the marketplace, Fun Run uses weight distribution to measure a user’s posture and stride. It also scans the running surface and analyzes data on both the surface type (soil, sand, concrete, etc.) and condition (dry, wet, icy, etc.). By continuously tracking and learning performance, the app can make recommendations based on a user’s goals. For example, if a user is concerned about hip pain, the app can let the user know if he or she is putting too much pressure on one side, then recommend relevant stretches/therapies and adjustments to the way they run in order to correct it. The algorithm can also use predictive analytics and alert people of the potential for injury or excessive soreness before people have experienced any pain.

Feasibility and Commercial Promise:

Athletes are increasingly interested in tracking their fitness. Of people who exercise at least monthly, 30% have a wearable fitness tracker and 29% use a mobile app to track fitness stats. Another ~25% plan to use these features in the future [1]. Despite significant commercial promise, the activewear market is saturating. 37% of fitness junkies state that brand is important, and 90% opt for high-end activewear brands [2]. As such, branding and high-performance will be integral to the success of Fun Run. Some of the features we plan to utilize may be expensive in the short term, such as sensors to analyze surface type and condition. The cost could be brought down by pulling in data from other sources (e.g., weather websites) or by relying on manual entry of some data.


A pilot would be rolled out at university track and field programs. We would work closely with a team of physicians and sports science staff at the University of Chicago (or comparable University), to certify and monitor all data and recommendations. We would seek to have each tailored recommendation and pain management solution be created and reviewed by our physician team to ensure proper legal protocol and medical viability.

Our pilot would seek out volunteer athletes who can provide large usage data sets (e.g., long distance runners) that would allow us to collect data on running strides, foot placement, weight pressure, shoe type and other relevant pieces of information. Additionally, we would bring the volunteers in for an initial screening process to create a baseline model of their postures, existing pain or medical issues, bone structure, etc. which would allow us to make better assessments from the running data we collect.

We would analyze the recommendations from our analytics software with our physician team to gauge relevance, accuracy, and benefit to the athlete above the normal utility of simple pain-based care.


A number of devices in the market measure biometrics such as heart rate and/or activity indicators such as steps and speed. They mainly allow people to track progress. Our product will be a leader in using AI to generate predictions based on such data combined with weather, running surface, and more granular biometrics information such as weight distribution. A competitive product on the market is Lifebeam VI, the self-proclaimed “first true AI personal trainer.” The product is a voice-activated Bluetooth biosensing headset with AI personal trainer. VI is focused on making you a proficient runner and is priced at $249.99. VI doesn’t analyze posture and weight distribution and doesn’t seem to be able to warn about potential injuries or excessive stress on your body. Other potential competitors are Google and Apple who can build on their personal assistants to also function as personal trainers. Existing user base and data will be a big advantage for them.


[1] Mintel Report. Exercise Trends. US, October 2016.

[2] Mintel Report. Activewear. US, October 2016.



Team Members:

Sam Steiny

Rosie Newman

Gergana Kostadinova

Javier Rodriguez

Augmented Judgment – Autonomous Vehicle


Autonomous vehicle industry and the problem it solves

The global autonomous vehicle market is estimated to be $42B by 2025. This is segmented by Partially Autonomous Vehicles and Fully Autonomous Vehicles, representing $36B and $6B, respectively. The main benefit from autonomous vehicles is the expected increase in safety. Thousands of people die in car accidents and autonomous vehicles are expected to be less error-prone than humans. Autonomous vehicles also have additional benefits such as helping those with physical limitations to mobilize easier. Other benefits include: reducing the number of vehicles on the road, lowering the amount of traffic violations, and providing a more comfortable and efficient way of transportation. These solutions can then be expanded to optimize ride-sharing services and reduce traffic congestion and lower fuel consumption.

Optimus Ride is at the forefront of creating an entirely autonomous vehicle. It leverages a system of hardware products (and machine learning algorithms) in concert with human drivers to develop semi-autonomous vehicles. Initial use cases of the vehicles include shuttle services in communities, commercial developments, airports, college campuses, amusement parks, and other relatively low traffic areas.

The solution is two-fold – hardware augmenting people becomes software augmenting people.

Vehicles come with two lightweight Velodyne lidar sensors, eight cameras, GPUs and motion sensors, and a proprietary switchboard that translates the sensor data into mechanical responses from the vehicle. The system uses cameras and lidar sensors, but dependence on lidar will decrease as Optimus’ accumulates data, which will train computer vision algorithms. The system will eventually shift reliance to computer vision, which will use less expensive hardware and is more scalable.

Path to autonomous vehicle

Time to prepare, calibrate, test, and deploy a vehicle currently takes several weeks. That timeframe is expected to decrease as the company solidifies formal production partnerships with OEMs. Discussions are ongoing for several pilots, including some in Massachusetts and Florida. The product is currently comprised of a full-stack autonomous solution encompassing lidar-based perception (front and rear), vision-based perception (via front and rear cameras), motion planning (via wheel encoders), computer integration (via the NvidiaDrive PX platform), and drive-by-wire control. Lidar is the industry standard, but Optimus is working towards an advanced computer vision-based autonomous solution through a multi-layered rendering from three distinct visual input techniques: visual slam, deep learning, and stereo vision. The resultant camera-focused autonomous system, complete with sensor fusion, will represent the core IP and proprietary software. The unique solution will have the capability to detect objects and obstacles in the vehicle’s path and determine the location of the vehicle in proximity to its surroundings with centimeter-level accuracy. With this scalable turnkey solution available, Optimus could deploy full fleets of autonomous vehicles controlled by client platforms.

Roadmap to vision-based autonomy                Vehicle use cases


The fully autonomous vehicle landscape is highly competitive:

Robust proprietary data is a key competitive differentiator in the autonomous vehicle space. Driving data will train machine learning algorithms, underpinning self-driving technology. To scale, video or image-based data will be the most valuable because it can lessen dependence on expensive lidar technology by shifting reliance to computer vision and software. Incumbent players have accumulated road-mileage but capturing vision-based data remains to be an arms race.

Competitive Landscape

  • Automakers (Ford, GM, Tesla) have been actively establishing partnerships with technology startups and making strategic acquisitions and investments in the autonomous vehicle space.
  • Ride Hailing Companies (Uber, Lyft) are well-aware of the transition to self-driving cars and are developing in-house, through partnerships or via acquisitions.
  • Technology Companies (Google, Apple)–Alphabet leverages GPS, Waze, and Google Maps to generate routes for autonomous vehicles and is developing autonomous offerings via Waymo.
  • Autonomous Software Startups (Nexar, Nauto, Drive.ai, nuTonomy, Varden Labs, Aurora, NextEV)–have received significant funding to power the autonomous shift

But Optimus Ride benefits from a proprietary dataset and pilot partnerships:

Over time, Optimus Ride can leverage its multifunctional ride complex and license to operate in Boston’s self-driving vehicle zone, along with partnerships it has signed with private developers and community and transit authorities to use controlled city zones to accumulate driving data. One of the most obvious constraining market factors is the limited pool of talent, which is a crucial factor driving the numerous strategic acquisitions by incumbents that have defined the autonomous vehicle market in the last year. The technical expertise of the founding team provides a strong competitive advantage.

Team Members: Brentt Baltimore, Moises Numa, Corey Ritter, and Mitchell Stubbs