Decoding The Success Factors for Co-working Space
Co-working, a modern concept of work space leasing business, has been over the top in the past few years. The business model is rather simple: the operator takes long term lease from the landlord — partially or the entire building, renovates the space and subleases to new tenants.
The public attention for the fancy concept went peak in 2019 and cooled down soon afterwards. While the hype faded, investors backed out, few companies failed, the remaining players struggled to sustain their business.
The challenge: from expansion to optimization
A co-working space operator in China, like many of its competitors, readjusted its pace from rapid expansion to a more steady business growth. The very goal now is to make the current business sustainable — a healthy P&L, improved product-market-fit, and smarter portfolio management.
CVI was engaged to look at this operator’s proprietary data as well as external data sources to provide a data-backed strategy for improving their business.
Knowing which factors matter the most
Before we can make any business suggestions, we need to understand why some of the spaces sell better than others. We concluded all the possible factors we could think of and built up a “store profile” comprising two types of factors: operational and location factors.
Operational factors are things that the operator directly owns or controls, such as pricing per unit, size of the workspace, tenant types, etc. These factors have direct impact to store performance, but not sufficient enough to explain how performance differs from one place to another. Hence, location factors.
Location factors are based on a strong hypothesis: location matters. It’s especially true for offline businesses. These factors are things that reflect the surrounding environment such as functions mix, number of amenities, foot traffic of the neighborhood, as well as accessibility, competitors, regional market price, etc. In this study, we used a 15-minute walking distance for every store as geo-boundaries (a.k.a. isochrone) to calculate the variables within.
After gathering all the factors we considered relevant, we then use a machine learning approach to build a “model” to help us identify the most relevant ones and explain how they impact the performance.
Typically, a multiple variable model looks like this:
In this equation, Y ( dependent variable) means the value we want to explain or predict, and in this exercise, the actual leased area of a store; and Xi ( independent variables) means the factor that influences Y.
After a series of model fine-tuning processes, we came down to the most important variables. Among all the 30+ factors we collected, the combination of nearby prime office counts, nearby company counts, nearby company size, area foot traffic and pricing has the best explainability for store performance. The formula goes like this:
Notation: The detailed process of modeling and variable selection is omitted intentionally in this article because it was lengthy and less fun. But one key principle we’d like to share is that for the purpose of business insight and implementation, we tend to appreciate a model that could be understood intuitively rather than a model that has high accuracy but low explainability. It’s often a trade-off between precision and explainability in practice. And it’s more of an art rather than hard science.
knowing which stores to improve and how
Once we have the model ready, we can use it to forecast each store’s future performance. In the following graph, each store gets three bars: gray bar means the total leasable area of a store, black bar means the actual leased area, and yellow bar means the predicted leased area. Comparing the three, we can apply different strategy accordingly:
1. Doubling down on the location
Total area (gray) ≌ Leased area (black) < Predicted area (yellow).
This means the current occupancy is great, but predicted performance is even greater than maximum capacity. The strategy would be taking more space from the landlord and harness the unsatisfied local demand.
2. Downsizing
Total area (gray) > Leased area (black) > Predicted area (yellow).
This means current performance is bad, and predicted performance is even worse or not significantly better. The strategy would be to stop bleeding immediately.The action could be decreasing the tenancy, or even closing the spot.
3. Improve sales strategies
Total area (gray) > Leased area (black) < Predicted area (yellow).
In this case, current performance is bad, but predicted performance is significantly better, this means something went wrong at operational level. Most likely, the sales may not be doing their best at the moment. The strategy would be improving sales strategy, and make sure the spot is well-perceived among community members and potential customers.
Uncovering the hidden meaning from the model
Data modeling is often used for prediction (revenue, sales, etc), and the accuracy of the model is valued over explainability. But for a company to plan for its future strategy, people expect the model to be understandable (to know not only what to do, but why doing it). The modeling approach can also be used to explain why things are the way they are and uncover insights that couldn’t be acquired otherwise.
Let’s look at our formula again:
Note that each variable alone can cause an increase or decrease of performance (coefficient positive or negative). But in reality, factors are often interrelated and we should interpret them in bundles.
Insights 1: current consumers are quality seekers
According to the equation, as nearby prime offices increase, the operator tends to lose business; and stores with higher prices tend to have better business.
This means the current customers are not very sensitive to price (or else they wouldn’t choose the brand in the first place), and they are quality seekers. The operator is essentially competing with premium offices that offer better experience, as opposed to other lower-tier co-working brands competing on sheer price.
Insights 2: diverse and matured business areas are better fits
According to the equation, not only we want the neighborhood to have more companies, but we also want the size of them to be bigger (more employees).
This means the operator should not just go for the busiest downtown — usually old lane houses with large amounts but tiny one-man companies; nor the designated industrial parks — usually remote, filled with big corporates, large building footprints but less quantity; but somewhere in between- areas that house the most mid-sized companies. One can think of those matured mix-use office clusters in the city, where living amenities are also well-provided, including retail, transportation, food and beverage, entertainments, etc.
Conclusion
Use the model to understand the past and prepare for the future
The case demonstrated here was about workplace leasing, but the business logic applies to many other industries as well, such as predicting and explaining retail sales, restaurant visitors, public amenity usage, etc. Data modeling is not only useful for forecasting and preparing for the future, it also helps us to examine our past and learn from it. It makes the most impact on an organization when it is internalized and applied across functions.
Engage domain knowledge for iteration and interpretation
The limitation of using data intelligence is that data tells us the ‘what’s and ‘how’s, but not the ‘why’s. This is why the explanatory modeling method is sometimes more critical than sheer prediction when it comes to business improvement and disruptive innovation. Choosing and interpreting different factors can be tiring and tedious, and it requires a deep understanding of the business logic. One should avoid miss-interpretation by engaging domain experts early in the process, and all the way to business suggestions.
This article was produced by CVI, a data analytics company with a mission to help retail businesses make better decisions. At CVI, we take a citizen-centered and data-driven approach to build technology tools and to formulate strategies that empower city-shapers. If you are a food service operator, a retailer, a city planner or an urban design enthusiast and are interested in harnessing the power of location intelligence, give us a holler at info@cvi-tech.com.
