Identifying the Optimal Location for a New Business

MIDTERM PRESENTATION

November 13, 2017

W. Dai, A. Srivastava, R. Castellanes, D. First

Columbia University: Y. Garg, A. Mueller

Synergic Partners: G. Ribeiro

Agenda

• The Problem and Our Approach
• Literature Review
• Data Sources
• Initial Data Exploration
• Goals and Next Steps

THE PROBLEM AND OUR APPROACH

• Problem Statement
• Our Approach

The objective of this project is to identify the optimal location to open up a new business.

ASSUMPTIONS

• Begin with Chinese Restaurants and then expand
• Optimize towards maximizing profit
• Limited to Manhattan, New York
• We want the user to be able to decide for herself what factors to prioritize

PROBLEM STATEMENT

METHODOLOGY

Approach 1: Acquire a dataset of NYC Chinese restaurants and with their profitability, then understand drivers of profitability and model it

Problem: Dataset?

Approach 2: Focus on business considerations that are drivers of profitability: revenue, cost, competition, and closeness to transportation

Example

We do not know the costs of real estate for restaurants, so we will use publicly available real estate prices instead​

Recommending a location based on user preference

• The user will rank how much they care about each of these factors
• For each NYC location area, we derive a score to each restaurant for four categories:
  • Expected popularity
  • Cost
  • Competition
  • Distance to subways
• A location is then selected that scores highest on these factors

1

2

3

Profit Score

Recommending a location based on user preference

• The user will rank how much they care about each of these factors
• For each restaurant, we derive assign a static score to each restaurant for four categories:
  • Profit and Popularity
  • Cost
  • Competition
  • Distance to subways
• A location is then selected that scores highest on these factors

Example:

1

2

3

Profit Score

1

2

3

"I want a low-cost restaurant in a popular area, somewhat close to subways. I don't care about competition, because I'll differentiate."

1

.5

0

.5

Location 1

Location 2

Location 3

Location 4

1

0

.5

.75

.25

.30

.60

.15

.80

.20

.20

.15

.40

.60

.80

.90

Total Score

1.325

.45

1.2

.506

Recommend Location 1

Cost

Popularity

Comp.

Transportation

Literature Review

Unsupervised Learning: Collaborative Filtering

In Eravci et al., the authors divided up NYC into neighborhoods. Then they used collaborative filtering to recommend businesses to new neighborhoods

Recommendations based on Linear combination of Drivers of Profitability

Two approaches have been taken in the literature

In Khateryna et al., the authors generate a location recommendation for Ukranian businesses based on combining the estimated profits and costs for different locations.

Our Approach

Problem: We want to give more granular-level recommendations
We want to allow the user to prioritize cost vs. popularity, and the like

We will integrate various

publicly-available datasets

In the CF approach, which is not in our model, the authors split up NYC into neighborhoods based on an unsupervised clustering method

EXAMPLE: CLUSTERING

DISTANCE

http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=7836791

PROBABLISTIC NEIGHBORHOOD SELECTION

COLLABORATIVE FILTERING

DATA SOURCES

• Yelp
• Foursquare
• NYC Open Data
• Loopnet

Example:

1

2

3

"I want a low-cost restaurant in a popular area, somewhat close to subways. I don't care about competition, because I'll differentiate."

1

.5

0

.5

Location 1

Location 2

Location 3

Location 4

1

0

.5

.75

.25

.30

.60

.15

.80

.20

.20

.15

.40

.60

.80

.90

Total Score

1.325

.45

1.2

.506

Recommend Location 1

Cost

Popularity

Comp.

Transportation

We integrated various data sources in order to score each location on four metrics

Expected Popularity

Distance to Subways

Cost of Real Estate

Competition

Dataset

LoopNet

Yelp

Foursquare

Demographic Data (NYC Open Data)

Yelp

Foursquare

NYC Open Data

Metric

Yelp

Dataset: ~1,000 Chinese Restaurants in NYC

Fields Collected:

• Name
• Latitude
• Longitude
• Price
• Rating
• Number of Reviews

Foursquare

Dataset: ~600 Chinese Restaurants in NYC

Fields Collected:

• Name
• Latitude
• Longitude
• Check-ins
• Visitors

NYC OPEN DATA

Datasets Collected:

• Population by Zip Code (cut by Age and Sex)
• Subway Latitude and Longitude
• Bus Stop Latitude and Longitude

LOOPNET

Background: Loopnet lists out commercial real estate listings, including retail spaces

​

Parameters: We limited our search to NYC listings of locations <2000 SF, ground level

• Vacant Address
• Price per Sq. Ft. per Rent
• Derived latitude and longitude

INITIAL DATA EXPLORATION

• Price per Square Feet
• Density of Chinese Restaurants
• Number of Reviews
• Ratings
• Initial Profit Score

PRICEST ARE HIGHEST IN marquee RETAIL NEIGHBORHOODS

Tribeca and Midtown Manhattan, expensive retail neighborhoods, show the highest price per square foot per year

As expected, chinatown has the highest density of chinese restaurants

Other neighborhoods have a significantly lower and consistent density of Chinese Restaurants

Downtown Manhattan's Chinese restuarants seem to be the most popular

Both no. of reviews (Yelp) and checkins (Foursquare) seem to be highest for lower Manhattan

RATINGS ARE INCONSISTENT ACROSS NEIGHBORHOODS

In general, downtown generated higher ratings than uptown

OUR INITIAL PROFIT SCORE PROJECTS HIGHEST PROFIT ON THE UPPER WEST SIDE

Areas with high density of Chinese Restaurants and fewer reviews show lower profit scores at this point

For our initial model, we weighed each factor equally: expected popularity, expected cost, competition, and distance to subways: Coef = [1,1,1,1,]

GOALS AND NEXT STEPS

• Define Location Areas
• Model Target Scores
• Create Web App

DEFINE LOCATION AREAS

In this, we will cluster locations based on price per square foot per year and distance

Once we see clusters that make sense, we will create their respective shape files that will serve as location areas.

FINE TUNE TARGET SCORE

We will fine-tune a meaningful target score that well-represents estimated profit at the locations we have defined.

Example:

1

2

3

"I want a low-cost restaurant in a popular area, somewhat close to subways. I don't care about competition, because I'll differentiate."

?

?

?

?

Location 1

Location 2

Location 3

Location 4

1

0

.5

.75

.25

.30

.60

.15

.80

.20

.20

.15

.40

.60

.80

.90

Total Score

?

?

?

?

Recommend ?

Cost

Popularity

Comp.

Transportation

Final output - WEB APP OF RECOMMENDED LOCATION AREAS BASED ON USER INPUT
- INPUT PAGE 1

Final output - WEB APP OF RECOMMENDED LOCATION AREAS BASED ON USER INPUT
- INPUT PAGE 2

Final output - WEB APP OF RECOMMENDED LOCATION AREAS BASED ON USER INPUT - RESULTS PAGE

APPENDIX

Estimating Revenue

# of Customers

$$ per Order

=

x

• Our target variable

• Can be assumed constant
  (e.g. Chinese Restaurants can expect $20 per order)

• We will need to estimate this with data we have collected

Estimating COSTS

Fixed Costs

Variable Costs

=

+

• Our target variable

• The main variable cost is retail space rental, which varies by location

• A lot can be assumed to be fixed. For Chinese Restaurants:
  • Overhead
  • Maintenance
  • Size of Restaurant
  • Staff

MODELING APPROACH

We will look into a sample of following features that could potentially impact overall profit:

• Distance from nearby subways and bus stops
• Populations by zip code by age and gender
  • Population density
• Number of competitor businesses in the vicinity
• Estimates of popularity
  • Yelp rating
  • Foursquare checkins
• ... And of course, price per square foot per year

Zillow

Dataset: All neighborhoods and zip codes in NYC

Fields Collected:

• Neighborhood
• Zip Score
• Zestimate rank (gives an idea of home price)