Adding Data to a Process Model

You can use your real world business data to power your process model and create a digital twin of your business process.

Loading Data

HASH supports CSV and JSON datasets. Upload your dataset by clicking the 'new dataset' option in the editor (in the top left of the screen, over the file picker) or by finding/reusing an existing dataset from the HASH index and adding it to the simulation.

The dataset is stored in the 'data/' folder, and the data will be available in the simulation through context.data()

const behavior = (state, context) => {
  let cs = context.data()["CustomerServiceEmployeesSheet1.csv"];
};

Customizing Parameters

With the data loaded in the simulation we can use it to power the process model. There are many different ways to use data + process models, but the most common are:

• Set the properties of the process model. For example, if a pizza delivery service model has a certain number of drivers available to deliver pizzas, they could use a dataset to set that number.
• Create distributions for modeling uncertainty. Now, if the pizza delivery service wants to estimate the number of orders they'll get in a day, they could take a dataset that captures last weeks orders of pizzas, choose a distribution that accurately models the data, and then use that to simulate new orders.

Example: Customer Service Data

A common business process simulation is the ticket support queue. Tickets from users come in and arrive in a queue. An available customer service agent takes a ticket, solves it, and then picks up the next ticket.

This process would have three elements:

• At the beginning, a source block that generates tickets
• At the end, a sink block that collects finished tickets
• In the middle, a service block where customer service workers take tickets and work on them. When a CS employee is working a ticket, they can't take a different ticket until they finish.

Using the Process Model Builder, we can set properties for each of the blocks and update my simulation with it. The resulting code will look like:

const agent = {
  agent_name: "process_model",
  behaviors: [
    "@hash/age/age.rs",
    "@hash/process/source.js",
    "@hash/process/service.js",
    "@hash/process/sink.js",
  ],
  process_labels: ["", "start", "work_on_tickets", "solved_tickets"],
  process_parameters: {
    start: {
      template: {},
      frequency: 1,
      rate: null,
      next_block: "work_on_tickets",
    },
    work_on_tickets: {
      time: 1,
      resource: "cs_employees",
      next_block: "solved_tickets",
    },
    solved_tickets: {},
  },
  position: [0, 0, 0],
};

Now we can take a dataset defining different customer service employees, download it as a CSV, load it into the simulation, and use the data to set the number of CS employees we have on hand.

// create_process.js

 "start": {
        "template": {},
        "frequency": 1,
        "rate": null,
        "next_block": "work_on_tickets"
      },
      "work_on_tickets": {
        "time": 1,
        "resource": "cs_employees",
        "next_block": "solved_tickets"
      },
      "solved_tickets": {}
    },
    //added a line for the cs_employees resource
  "cs_employees": context.data()["ExampleCustomerServiceEmployeesSheet1.csv"].length
}

This is a good start, but to make the model more realistic we can add a dataset representing ticket volumes over the past five days. I'll take the mean and standard deviation of the data, and set a triangular distribution for the number of tickets to generate every time step.

// create_process.js

const { triangular } = hstd.stats;
let tickets = context.data()["Tickets.csv"];
//Get min and max number of tickets received per day
let min = tickets.reduce(
  (min, t) => (t.count < min ? t.count : min),
  data[0].count,
);
let max = tickets.reduce(
  (max, t) => (t.count > max ? t.count : max),
  data[0].count,
);
//Get the mode of the array and average with min and max to find the peak
let peak = (min + max + mode(tickets)) / 3;

// in the process parameters ...
{
 "process_parameters": {
      "start": {
        "template": {},
        "frequency": 1,
        "rate": triangular.sample(min, max, peak),
        "next_block": "work_on_tickets"
      },
      "work_on_tickets": {
        "time": 1,
        "resource": "cs_employees",
        "next_block": "solved_tickets"
      },
      "solved_tickets": {}
    }
}

Now when we we run the simulation, we have a good estimate of what my customer service process will look like using real world data.