ATM cash management solution for a bank

Client Issue

A large bank wanted to know, whether there is an opportunity to optimize costs related to the ATM network management. Bank provided required data to the experts of Deloitte Analytics Institute (namely costs related to the cost of capital, current encashment costs, ATM's daily status, daily ATM's residuals, withdrawals etc.). Project has been done for a big branch in one of the Russian cities with a population above 1mln people and amount of ATM's above 250 items.

DAI Solution

Deloitte Analytics Institute developed sophisticated system that utilizes wide domain banking expertise and machine learning algorithms that took into account data on considered ATMs performance. DAI applied a complex approach including the following steps:

  1. Problem framing.

    Deloitte collected all costs related to ATMs and other important data related to cash transactions (both cash in and cash out).
  2. Data collection and preprocessing.

    Deloitte used data collected from many different sources and merged it with banks data. Examples of data used includes:

    • ATM’s summed daily transactions
    • Calendar (weekends, public holidays etc.)
    • Weather
    • Area surroundings
    • Macroeconomics drivers (i.e., currency rates, industrial indexes, price indexes etc.)
    This step included data consistency check between different sources and inconsistencies treatment, handling with duplicates and null values, dictionary forming and transformation, generation of new features, statistics calculation.
  3. Data processing and algorithms engineering.

    Deloitte AI built 3 stages model that allows to establish connection between changes in macroeconomics and certain bank financial performance:
    • 1st stage: Forecasting macroeconomics indicators using recurrent neural network – macroeconomics in this model is assumed as whole, highly connected and complex system.
    • 2nd stage: Using macroeconomics indicators and ATMs data to predict certain ATM performance. The model is based on linear models with regularization and allows to capture specific dependencies for each ATM.
    • 3rd stage: Solving an optimisation problem between cost of capital, cost of missed opportunities of using cash and encashment cost (incl. travelling cost).
    Outcome of this three stages model is a precise list of ATM’s with a schedule of servicing them and amounts to be serviced.


  • Machine learning
  • Neural networks
  • Time series analysis
  • Econometrics
  • Optimization algorithms


  1. A solution with a frontend and backend parts has been developed to capture the ATM’s behavior and optimize costs of encashment taking into account cost of capital, cost of transportation and cost of processing.
  2. The solution has been tested for the results with real world data.

How our solution affects the Shareholder Value

Shareholder Value

  1. Revenue Growth
    1. Volume
      1. Acquire New Customers
      2. Retain & Grow Current Customers
      3. Leverage Income Generating Assets
    2. Price Realization
      1. Strengthen Pricing
  2. Operating Margin (after taxes)
    1. Selling General & Administrative (SG&A)
      1. Improve Customer Interaction Efficiency
      2. Improve Corporate/ Shared Service Efficiency
    2. Cost of Goods Sold (COGS)
      1. Improve Development & Production Efficiency
      2. Improve Logistics & Service Provision Efficiency
    3. Income Taxes
      1. Improve Income Tax Efficiency
  3. Asset Efficiency
    1. Property, Plant & Equipment (PP&E)
      1. Improve PP&E Efficiency
    2. Inventory
      1. Improve Inventory Efficiency
    3. Receivables & Payables
      1. Improve Receivalables & Payables Efficiency
  4. Expectations
    1. Company Strenghs
      1. Improve Managerial & Governance Effectiveness
      2. Improve Execution Capabilities
    2. External Factors