Operations research at Tokyo Gas

From demand forecasting to emergency response, Japan’s largest city gas supplier counts on O.R. to advise the company’s decision-making process.

By Kosuke Shaku and Toshinori Sasaya

Figure 1: The business areas of Tokyo Gas.

Figure 1: The business areas of Tokyo Gas.

Tokyo Gas, which is Japan’s largest city gas supplier catering to more than 11 million customers in the Tokyo Metropolitan area, has been conducting operations research (O.R.) for more than 40 years, mainly in the following areas:

  • Marketing: How to increase the sales of gas, electricity and appliances (e.g., demand forecasting of gas appliances, gas/electricity pricing)
  • Emergency response: How to respond swiftly to accidents (e.g., gas leaks) and emergencies (e.g., earthquakes), and optimizing the dispatch of employees
  • Customer satisfaction: How to enhance customer service efficiency (e.g. human resource planning in call centers)
  • Liquefied natural gas procurement planning: How to make decisions pertaining to the procurement of liquefied natural gas (LNG) (e.g., optimizing the balance between long-term LNG contracts and spot LNG contracts

This article introduces two typical instances that describe how O.R. is applied to our business.

Marketing:
Demand Forecasting of Gas Appliances

Tokyo Gas has more than 70 service shops that sell gas appliances (e.g., hot-water heaters), residential fuel cells (also known as ENEFARM) and floor heating systems, among others. In addition to increasing appliance sales, these shops help increase the volume of gas sales. Therefore, the sale of more gas-consuming appliances is promoted over that of less gas-consuming ones. For example, selling a residential fuel cell that generates both hot water and electricity is given higher priority than a conventional hot-water heater.

Figure 2: The number of stocked appliances (Step 1). Here, we aggregate the data on customer ownership of appliances based on the years elapsed and on the type of gas equipment (e.g., more gas-consuming water heater: Type A; less gas-consuming water heater: Type B).

Figure 2: The number of stocked appliances (Step 1). Here, we aggregate the data on customer ownership of appliances based on the years elapsed and on the type of gas equipment (e.g., more gas-consuming water heater: Type A; less gas-consuming water heater: Type B).

In order to devise a reasonable strategy to raise the sales of such gas-consuming appliances, we forecast the demand for such appliances by utilizing the customer relationship management (CRM) data from our CRM systems. These systems record the history of the company’s relationship with every customer, such as the records of customers owning gas appliances. Related data from the system are pulled up in the following manner in order to estimate the replacement demand.

  • Step 1: Count the number of stocked appliances.
  • Step 2: Calculate the probability of replacement by conducting a survival analysis of the gas appliances.
  • Step 3: Estimate the transition probability of the types of appliances that need to be replaced.

By multiplying these three elements, we are able to forecast the potential demand for each type of appliance.

The results of this analysis are utilized in the following manner:

Setting fair sales goals: As mentioned earlier, Tokyo Gas has more than 70 service shops. Given that these shops compete with one another with regard to the achievement rate of their sales goals, it is very important to set a fair sales goal that corresponds to the demand of each area to ensure proper sales management and efficient sales operations. However, determining “fair” sales goals is very difficult because the demand for gas appliances depends on the wealth or the condition of equipment in different areas. Therefore, we consider sales goals to be proportional to the scientifically estimated demand in areas consisting of individual service shops.

Reducing the cost of the wholesale purchase of gas appliances: To achieve incentives from bulk orders, Tokyo Gas purchases all the gas appliances that would be sold by all the 70 service shops instead of purchasing them individually. To reduce the purchase cost and avoid over-ordering, the precise forecasting of appliance demand is essential. Thus, we forecast the appliance demand for several upcoming years.

Figure 2: The number of stocked appliances (Step 1). Here, we aggregate the data on customer ownership of appliances based on the years elapsed and on the type of gas equipment (e.g., more gas-consuming water heater: Type A; less gas-consuming water heater: Type B).

Figure 3: The probability of replacement (Step 2). We calculate the probability of replacement by using survival analysis by considering the so-called right censoring. Based on the years elapsed and the type of gas equipment, we aggregate the number of replacements as the numerator and the number of stocks as the denominator from the CRM system.

Area marketing (effective marketing that reflects area characteristics): The characteristics of each area served by our shops are distinctive. For instance, one service shop is located in an urban area with mostly apartment houses, whereas another is located in the suburbs with mostly stand-alone houses. As a result, the proportions of the types of gas appliances stocked for each area differ significantly. Thus, in order to implement effective sales promotion measures, we must analyze the characteristics of and respective demand of each area. In fact, we are initializing a trial in which a service shop uses the estimated potential and cluster data on sales promotions. For example, we send direct messages to customers residing in an area with a high demand forecast in order to drive sales.

Figure 4: The transition probability of the types of appliances that need replacing (Step 3): the probability of change in the type of appliance from type-i to type-j.

Figure 4: The transition probability of the types of appliances that need replacing (Step 3): the probability of change in the type of appliance from type-i to type-j.

Emergency Response:
Dispatching Employees in Case of an Earthquake

When large earthquakes occur, Tokyo Gas employees are dispatched to predesignated locations, even when they are off duty. Transportation systems often become non-functional, thereby forcing employees to walk long distances, consequently prolonging response times. In order to get employees together quickly, we optimize the assignment of each employees in the following manner.

  • Step 1. Simulation of employee travel: Search for the shortest path from the departure location to the predesignated destination by using Dijkstra’s algorithm while considering road widths and the estimated delay coefficient.
  • Step 2. Dispatch optimization of employees: The average travel time required to reach each predesignated location is minimized under the constraint of staff requirements. We subsequently redesignate the target locations by optimizing employee dispatch to minimize travel time.

As a result, we allocated 1,300 employees to 13 offices. Our results showed that travel time can be significantly improved by this optimization; we thus changed the allocation rule successfully.

Figure 5: A conceptual image of area marketing.

Figure 5: A conceptual image of area marketing.

Figure 6: A conceptual simulated image of employee travel.

Figure 6: A conceptual simulated image of employee travel.

Figure 6: A conceptual simulated image of employee travel.

Figure 7: A conceptual image of dispatch optimization.

Figure 6: A conceptual simulated image of employee travel.

Figure 8: The effects of the optimization.

Future Work:
Electricity and Gas Market to be Freed

In Japan, electricity and city gas sales have long been restricted to only one company in one district. However, the longstanding rules of electricity sales will be abandoned this April, and those of gas sales the following year.

In this scenario, Tokyo Gas is determined to begin the sales of low-voltage electricity and aims to become a “total energy company.” We are currently conducting operations research in several relevant fields.

In order to win electricity accounts and defend gas accounts effectively, we are utilizing CRM data and other available data (e.g., national population census data) for segmentation, targeting and other marketing measures. In addition, we are optimizing our electricity/gas tariffs in order to realize strategic pricing.
To maximize profits in power procurement, we are optimizing the operations of our own power plants and amount of trading simultaneously (asset optimization and trading).

The fields in which operations research is essential for our business are increasingly expanding.

Kosuke Shaku is a data analyst at Tokyo Gas, a Japanese vertically integrated gas supplying company, and specializes in marketing research and business efficiency.

Toshinori Sasaya is a senior data analyst at Tokyo Gas and specializes in marketing research and business efficiency.