• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    A method and apparatus for improving cash and liquidity management of an organization using a plurality of currency accounts is disclosed. The enhancements optimize interest income for the cash balances of each foreign currency account and minimize the expense of funding foreign currency accounts. Machine learning techniques are built in to forecast payments, receipts, interest rates, and exchange rates, then cash is transferred, borrowed, or loaned to fund payments and use available cash.
    公开号:CH716790A2
    申请号:CH00329/20
    申请日:2020-03-19
    公开日:2021-05-14
    发明作者:Joliveau Edouard
    申请人:Bottomline Tech Sarl；
    IPC主号:
    专利说明:

    Technical area
    The system, apparatuses and methods described herein generally relate to cash management software and, in particular, the use of machine learning to improve international cash management.
    Description of the related art
    The term cash management is used to describe the optimization of cash flow and the investment of excess cash. From an international point of view, cash management is very complex because the laws relating to cross-border cash transfers differ from country to country. In addition, fluctuations in exchange rates can affect the value of cross-border cash transfers. Financial software must understand the pros and cons of investing money in foreign markets so that they can make international cash management decisions to maximize business value.
    [0003] For a multinational corporation, revenues and expenses occur in multiple countries and currencies. By optimizing cash locations and borrowing or investing, significant value can be created only within the treasury function of the business. While cash management is a complex operation in a single entity operating in a single currency, in a multinational corporation with multiple subsidiaries, cash management is a very complex operation requiring complex software systems that are networked with multiple banks in multiple locations around the world, each with multiple accounts.
    [0004] International cash management software must quickly identify cash inflows and accelerate cash inflows, because the faster the inflows are received and identified, the faster they can be invested or used to pay off obligations. The currency and location of the entrances are also important.
    [0005] Payments are a second aspect of cash management software, comprising the obligations and deadlines for these payments. Again, the software must take into account provenance, country, bank, account, and currency, all of which are important factors.
    Another technique for optimizing cash movements, compensation, can be implemented by the centralized cash management software. This technique optimizes cash flow by reducing administrative and transaction costs resulting from currency conversion. First, clearing reduces the number of cross-border transactions between subsidiaries, thereby reducing the overall administrative cost of these cash transfers. Second, it reduces the need for currency conversion since transactions occur less frequently, thus reducing the transaction costs associated with currency conversion. Third, cash flow forecasting is easier because only net cash transfers are made at the end of each period, rather than individual cash transfers throughout the period. Better forecasting of cash flows can improve financing and investing decisions.
    [0007] A multilateral clearing system generally involves a more complex exchange between the parent company and several subsidiaries. For most large multinational corporations, a multilateral clearing system would be needed to effectively reduce administrative and currency conversion costs. Such a system is normally centralized in order to bring together all the necessary information. From the consolidated cash flow information, the net cash flow positions for each pair of units (subsidiaries or otherwise) are determined and the actual reconciliation at the end of each period can be requested. The centralized group can even maintain inventories of various currencies so that currency conversions for end-of-period net payments can be made without significant transaction costs.
    [0008] Laws relating to local government clearing and blocking of fund transfers also add to the complexity of international cash management.
    [0009] Many multinational corporations have at least $ 100 million in cash balances in banks in various countries. If they can find a way to earn an extra 1% on those funds, they will generate an additional $ 1 million each year on cash balances of $ 100 million. Thus, their short-term investment decision affects the amount of their cash inflows. Their excess funds can be invested in domestic or foreign short-term securities. At certain times, short-term foreign securities will have higher interest rates than domestic interest rates.
    Centralized cash management is more complicated when the company uses multiple currencies. All surplus funds could be pooled and converted into a single currency for investment purposes. However, the benefit of pooling can be outweighed by the transaction costs incurred when converting to a single currency.
    Centralized cash management is also valuable. Short-term liquidity available between subsidiaries can be pooled so that there is a separate pool for each currency. Then the excess cash in a particular currency can always be used to cover other branch shortfalls in that currency. This way, funds can be transferred from one branch to another without incurring the transaction fees that banks charge for exchanging currencies. This strategy is especially achievable when all branch funds are deposited into branches of a single bank so that funds can be easily transferred between branches.
    Another possible function of centralized cash management is to invest funds in securities denominated in foreign currencies which the subsidiaries will need in the future. Companies can use excess cash to invest cash in international money market instruments to cover all credit positions in specific foreign currencies. If they have foreign currency debts that should be revalued upwards, they can hedge those positions by creating short-term deposits in those currencies. The maturity of a deposit would ideally coincide with the date the funds are needed. By integrating with the payment software, the treasury software knows when cash outflows will occur.
    International cash management requires timely information between subsidiaries regarding the cash positions of each subsidiary in each currency, as well as information on the interest rates of each currency. A centralized cash management software system needs a continuous flow of information about currency positions in order to be able to determine whether the cash shortage of one branch can be covered by the cash surplus of another branch in this motto. Considering the major improvements in online technology in recent years, all multinational corporations can easily and efficiently create a multinational communication network between their subsidiaries to ensure that the information on cash positions is continuously updated.
    [0014] The payment and cash management (PCM) solution 110 provides business customers with a single, centralized, secure, intelligent and compliant solution for all banking relationships. PCM reduces the complexity, costs and risks involved in using multiple channels and connectivity to recover cash positions from various bank accounts.
    The modules shown in lafigure 1 effectively create a multi-bank platform 110 independent of the various banking relationships 121-126, with a configurable workflow allowing both simple processing or approval schemes, designed according to the needs of the business. The resulting payments are generated transparently behind the scenes via a list of supported payment mechanisms such as BACS, Faster Payments (Direct Corporate Access) and the SWIFT Gateway. Additional connectivity is obtained with an H2H or API connection with dedicated banks121-126.
    [0016] The PCM110 helps customers manage their short-term cash position by centralizing bank account balances and transaction data in one place, the PCM110 provides an accurate picture of cash liquidity regardless of bank, currency, geography, head office or subsidiary and also allows cash flow forecasts.
    In its most basic version, the PCM110 can be used as a multi-network payment gateway using the BT120 universal aggregator. Additionally, format validation and sanction filtering capabilities can be activated.
    [0018] In addition to the multi-network gateway, the PCM110 can activate a statement manager module112 which brings value-added functionality to the incoming workflow, taking advantage of data from banks121-126 (i.e. statement and bank confirmations).
    In addition to the multi-network gateway, the PCM110 can activate a payment manager module111 which brings value-added capabilities to the outgoing workflow. This payment manager module111 can be coupled to a sanction filtering module116. The sanction filter module 116 can use artificial intelligence to create and compare a profile of the transaction against profiles of sanctioned entities and block transactions that match a sanctioned entity.
    [0020] When it already combines the payment manager 111 and the statement 112, a cash manager 115 can be activated to provide value-added cash management functions.
    [0021] All of this analysis and planning is extremely complex and works best with a context of payment history and receipts in various currencies. To deal with the complexities, a machine learning-based solution can dramatically improve other solutions.
    Machine learning and artificial intelligence algorithms have been observed in the computer literature for half a century, with slow progress in the field of predictive analysis. Now we can take a large dataset of various features and process that training dataset through one of the many learning algorithms to create a data-driven rule set. This set of rules can reliably predict what will happen for a given event. For example, in a cash management application, with a given event (set of attributes, i.e. payments and receipts), the algorithm can determine the likely cash flow requirements.
    Machine learning is a method of analyzing information using algorithms and statistical models to find trends and patterns. In a machine learning solution, statistical models are created or trained using historical data. During this process, a sample of data is loaded into the machine learning solution. The solution then finds relationships in the training data. As a result, an algorithm that can be used to make predictions about the future is developed. Then the algorithm goes through a tuning process. The tuning process determines the behavior of an algorithm in order to provide the best possible analysis. Typically, multiple versions or iterations of a model are created in order to identify the model that provides the most accurate results.
    Typically, models are used to make predictions about the future from past data or to discover patterns in existing data. When predictions about the future are made, models are used to analyze a specific property or characteristic. In machine learning, these properties or characteristics are called features. A feature is like a column in a worksheet. When discovering profiles, a model could be used to identify data that is outside of a standard. For example, in a dataset containing payments sent by a company, a model could be used to predict future payments in different currencies.
    [0025] Once a model is trained and optimized, it is typically published or deployed in a production or quality assurance environment. In this environment, data is often sent from another application in real time to the machine learning solution. The machine learning solution then analyzes the new data, compares it to the statistical model, and makes predictions and observations. This information is then sent back to the original application. The application can use the information to perform a variety of functions, such as alerting a user to perform an action, displaying data that is outside of a standard, or prompting a user to verify that the data has been correctly characterized. The model learns from each intervention and becomes more efficient and accurate as it recognizes patterns and discovers anomalies.
    An effective machine learning engine can automate the development of machine learning models, dramatically reducing the time spent examining false positives, drawing attention to the most important things, and maximizing performance based on feedback. of the real world.
    [0027] As the treasury software has become more complex, in particular to attempt to manage the cash positions in several subsidiaries and currencies, it has become necessary to improve the cash management software with a machine learning functionality in order to that the profiles of cash inflows and outflows can be analyzed to optimize cash balances in all currencies. The present set of inventions meets this need.
    BRIEF SUMMARY OF THE INVENTION
    An improved cash management apparatus is provided which includes one or more bank rails connected to one or more banks, a special purpose server connected to one or more bank rails, and a plurality of data storage facilities connected to the special purpose server. The special purpose server is configured to retrieve a set of payment and revenue transactions from the plurality of data storage facilities for a given range of past dates, and perform ARIMA analysis in each foreign currency account on the site. 'set of payment and revenue transactions to create a revenue forecast and a payment forecast for a future period. The special-purpose server subtracts the forecast of payments from the forecast of revenue and adds a cash balance from the previous day to create a time series of forecasted cash balances for each foreign currency account. The special purpose server is configured to retrieve historical bank rate information and perform ARIMA analysis on historical bank rate information to create a time series of forecast bank rate information. The special-purpose server is configured to run an algorithm on the forecast cash balance time series for each foreign currency account and the forecast bank rate information time series to determine a set of optimal cash transfers between each foreign currency account and one or more passing accounts, then the special purpose server executes instructions for making payments and cash transfers.
    The algorithm can be a machine learning algorithm and the machine learning algorithm can be DensiCube, Random Forest or K-means (partitioning in k-means). The range of future dates and / or the given range of past dates may be configurable by the user. The payment forecasts can be modified to incorporate the actual expected payments. In addition, the revenue forecast can be modified to incorporate the actual revenue received. Execution of instructions for making payments and cash transfers can be accomplished by transferring one or more files to the plurality of bank rails. Foreign currency accounts may include the foreign currency accounts of a plurality of subsidiaries. Bank rate information can be retrieved from one or more banks via bank rails. Bank rate information may include interest rates, currency exchange rates, and money transfer fees.
    A method of managing cash flow in an organization is also described here. The method includes the steps (1) of retrieving a set of payment and revenue transactions from a plurality of data storage facilities for a given range of past dates for a plurality of currency accounts with a special purpose server which is connected to the plurality of data storage facilities; (2) performing an ARIMA analysis in each foreign currency account on all payment and revenue transactions, creating a revenue forecast and a payment forecast for a range of future dates; (3) subtracting, with the special-purpose server, the forecast of payments from the forecast of revenue and adding a cash balance from the previous day to create a time series of the forecast cash balance for each account in currencies; (4) recovery of historical bank rate information for special purposes; (5) performing the ARIMA analysis on historical bank rate information to create a time series of forecast bank rate information; (6) Running, using the special-purpose server, an algorithm on the forecast cash balance time series for each currency account and the bank rate information time series to determine a set of optimal cash transfers between each foreign currency account and one or more transit accounts; and (7) executing, by the special-purpose server, instructions for making payments and cash transfers.
    BRIEF DESCRIPTION OF THE DRAWINGS
    Figure 1 is a block diagram of a payment and cash management system.
    Figure 2 is an example of a screen showing four currency accounts and a group of payments.
    Figure 3 is a diagram of the various functions of the cash management system.
    Figure 4 is a flowchart of the periodic invoice payment process.
    Figure 5 is a diagram of a possible hardware environment for operating the results creation engine.
    DETAILED DESCRIPTION
    The present invention will now be described in detail with reference to the drawings. In the drawings, each item with a reference number is similar to other items with the same reference number, regardless of any letter designation following the reference number. In the text, a reference number with a specific letter designation following the reference number refers to the specific item with the number and letter designation, and a reference number without a specific letter designation refers to all elements with the same reference number, regardless of any letter designation following the reference number on the drawings.
    From Figure 1, corporate data is extracted from the ERP101a system (enterprise resource planning), TMS101b (transport management system) and accounting systems 101c via a secure channel 107 to the PCM110. Additionally, users using a personal computing device such as a personal computer104, smartphone105, tablet, laptop, smart watch, or similar device via secure network access106 can configure, manage, observe, and operate the PCM110. Access to the secure network106 communicates through the secure channel107 to provide computing devices104,105 access to PCM110 modules.
    The secure channel 107 interfaces with the integration layer 113 of the PCM110. This channel107 provides access between the company's facilities, via the integration layer113, to the payment manager111, the statement manager112, the audit and security module114 and the treasury manager115.
    On the other side of the PCM110, the integration layer 113 interfaces with the universal aggregator network gateway 120. The universal aggregator120 provides access to a number of banks121-126 using a number of different protocols (SWIFT, ACH, RTP, BACS, etc.) and networks (bank rails504).
    The messages between the universal aggregator 120 and the integration layer 113 are made up of bank statements 118, account balances, interest rates, bank transfer fees and exchange rates entering the PCM110. The integration layer113 sends payment messages119, as well as messages to transfer funds, loan funds, and loan funds. In addition, message acknowledgments and other internal management information are exchanged.
    The universal aggregator 120 sends and receives information from banks 121-126 through the various bank rails 504.
    The payment manager module111 offers complete visibility over the payment lifecycle from entry (import or manual creation) to confirmation by the bank. The PCM110 allows you to import transaction files from ERP101a, TMS101b, DBMS101c or any other background application via a secure channel107. The payment manager then validates the payment, and collects the approvals through an approval workflow, and prepares the payment for sending to bank rails. However, before payment is made, payments are directed to the treasury manager115 to ensure that funding is available.
    The security model 114 allows granular control of rights or access rights making it possible to control access according to functionalities and data. The user will only see the features and data to which he has been authorized to access.
    The statement manager module 112 gives customers the ability to control and monitor the correct receipt of statements and to link them to accounts and business units in static data. The PCM110 module will import all data contained in the bank statement (end of day and intraday), allowing customers to control balances and transactions. Statements from partner banks are displayed in exactly the same way, regardless of bank, country, currency or format.
    The PCM110 can interface with a sanction filtering module116. This module116 will check transactions against sanction lists maintained by various government entities that prohibit transactions with sanctioned entities. The transaction sanction filtering is performed after authorization and before the transaction is sent over a network. Likewise, the risk and fraud management module 117 verifies transactions for fraud by performing various analyzes to prevent fraudulent payments.
    The PCM110's cash manager115 integrates liquidity management301, cash flow forecasting302 and cash pooling303 functions. See figure 2 for an example of screen200 of the cash management function. The thumbnails at the bottom of the screen show pending payments. Each payment sticker has a flag indicating the country where the payment is to be made as well as an indication of the currency (USD, GBP, EURO, AUD, etc.) and the account number. The currency amount is shown on the sticker, along with the number of payments included and the date the payment is due. The payment status is also included in the sticker (APPROVED, PAID, FULL, REJECTED, ARCHIVED, CANCELED, ERROR, IN PROGRESS, etc.).
    The top of the screen displays the balances of each currency account201. Each tile displays a currency, the account balance in that currency, and a graphic showing the amount available for payment and pending payments. A warning triangle is displayed when payments exceed the balance in that currency. The user has the option of converting funds into the currency or directly funding the account.
    Figure 3 PCM110 uses current bank account balances and high value cash transaction information to facilitate cash management. The Cash Manager 115 module includes a consolidated view of accounts, short-term cash flow forecasts, and the ability to generate cash transfers between accounts. The consolidated view of all bank accounts200, the „Liquidity view“ 301, provides users with a hierarchical view of the current and available fund balances on all accounts.
    Users can configure hierarchical groups of accounts 311 for cash management purposes. A standard hierarchy consists of a main account, with subsidiary accounts in the same currency. Several account hierarchies can be defined and these hierarchies can then be used by the liquidity view301 and the cash flow forecast302. New account hierarchies can be defined for cash pooling purposes303.
    Transactions, either through the payment process or captured directly (in the case of expected revenue) from back office systems are entered in the consolidated cash ledger312. Here they are compared to confirmation messages, interim statements and bank statements to identify open “open” transactions that have not yet been reported on a bank statement.
    For accounts actively managed through the PCM user interface, daily bank statements321 are loaded automatically through the file interface with a batch. These bank statements321 are used to provide a closing position for the previous day. Additional information is derived from debit and credit confirmations322, external payments323, and adjustment transactions324. The closing balance serves as the opening balance for the current position. The bank statement is also reconciled with the “open” transactions of the day and the corresponding transactions are closed.
    By combining and matching the transactions of your organization and the transactions of your partner banks, the PCM can create an accurate representation of short-term cash flow forecasts. The onboarding, statement processing, and transaction matching processes are automated background tasks and require no user intervention. The short term cash flow forecasting module displays the current balance +/-, pending payments, transfers and expected receipts for the next few days.
    PCM302 forecasting functionalities include creation and administration of hierarchical user cash flow forecasting structures that are independent of the bank, standard cash flow forecasts based on a 5 day liquidity forecast, capacity exploration to view scheduled transactions on individual accounts, adding one-time or recurring manual adjustments to the cash flow forecast (for recurring adjustments, a background process - scheduled on the daily run list, will add automatically recurring adjustment to cash flow forecast), conversion of account balances to account hierarchy currency (uses exchange rates, stored as static benchmarks to estimate the conversion between currencies), visual indicators for each account to indicate whether the expected statements or confirmations have been received.
    The adjustment of the ledger312 can be made directly in the PCM110 to reflect the business scenarios outside the system. Standard adjustment details can be added (account, amount, date, repeat option and month start day). Users can also search for general ledger adjustments, view details of individual general ledger adjustments, and export the results in the same way as for payments.
    In conjunction with short-term liquidity forecast302, the liquidity management functionality can be enabled to provide visibility of cash positions as it displays the current and available funds balance on all accounts in the account hierarchy. selected. Liquidity management functionalities include creation and administration of user-specific hierarchical cash reporting structures that are independent of the bank, an exploration capability to view statement lines for each account as well as transactions intraday, display of closed balance and display of available funds and conversion of account balances to account hierarchy currency (uses exchange rates, stored as static master data to estimate the conversion between currencies) .
    Cash management115 can be used to implement cash pooling303 both across a group or locally as needed. Cash pool views can be created from a cash account hierarchy311. To control the sweep and fund process, the following parameters can be set for each account in the cash pool: transfer action (sweep and funds, funds only, report only), minimum transfer, rounding, minimum balance and target balance . Cash pool controls313 define the operating rules for cash pooling303.
    [0057] In addition, the cash manager 115 provides the functionality of automatically funding cash positions in various currencies and in various accounts (or it could provide suggestions to a human cash manager). This feature interprets the current cash flow forecast and configured settings for cash pool accounts and implements transactions to sweep excess cash in the group main account or fund sub accounts of the group main account. This operation is repeated for each level of the account hierarchy. Transfers will be executed automatically (or created for authorization). Cross-currency cash pools can be set up, and these will use the exchange rates, stored as static benchmarks, to estimate conversions between currencies.
    Figure 4 shows a flowchart of the machine learning required to automatically determine the appropriate funding in each currency and in each account. First, the company's historical transactions are taken from the accounting systems101c, TMS101b and ERP101a, filtering402 these transactions in order to isolate receipts and payments in a review period. This review period corresponds to the duration, for example a month, a quarter, a year or a decade. The review period should be long enough to generate a large number of receipts and payments, enough to create statistically significant data. In addition, the review period should be long enough to cover several seasonal periods. For some companies a season can be a month, for others it is a year. In some embodiments, the review period is determined by a machine learning algorithm testing the quality of the resulting data over a variety of period values and choosing the best value for the available data.
    Next, an autoregressive integrated moving average (ARIMA) 403 is run on the data. An autoregressive integrated moving average (ARIMA) model is a generalization of an autoregressive moving average (ARMA) model. These two models are fitted to the time series data either to better understand the data or to predict future points in the series (forecasts). ARIMA models are applied in certain cases where the data present evidence of non-stationarity, where an initial differentiation step (corresponding to the “integrated” part of the model) can be applied one or more times to eliminate the non-stationarity. The ARIMA model is considered to be a machine learning technique in the field of artificial intelligence technology.
    In some embodiments, the ARIMA model is run, then any outliers beyond one or two standard deviations are removed from the dataset and the data is run again. Especially with receipts and payments, sometimes a very large payment for a lot of work or a backlog, the customer all of a sudden pays. This can override the predictive values and it is best to remove them from the data, unless there is reason to believe that this is a periodic event.
    Given a time series of data Xt, where t is an integer index and the Xts are real numbers, an ARIMA model (p, d, q) is given by
    Where L is the shift operator, the ∝is the parameters of the autoregressive part of the model, the θis the parameters of the moving average part and the εts are the error terms. The error terms εts are generally assumed to be independent and identically distributed variables sampled from a normal distribution with zero mean. The parameter p is the order (number of time lags) of the autoregressive model, d is the degree of differentiation (the number of times the data has had past values subtracted), and q is the order of the moving average model.
    The ARIMA403 model is run separately on the receipts and payments for each currency and each account, each predicting the receipts and payments for each group of foreign currency accounts. The ARIMA403 model provides the Xt time series for receipts and payments for each currency account in the next 15 or 30 days. Cash inflows or outflows (loan financing or transfers or investments) are not taken into account in this calculation.
    Since payments are typically paid around 30 days after receipt and the approval process takes 7-14 days, the PCM110 may experience 15 days of actual payments (other time periods can be used for other embodiments). These actual payments404 are factored into future payment time series. In some cases, several days of receipts may also be known in advance, depending on the characteristics of the bank account and bank rails. If they are known in advance, the recipes also switch to the actual values.
    Once the Xi time series for receipts and the Xi time series for payments are determined, the current balance is calculated by taking the previous balance in the currency account and subtracting the Xi payments for the day and adding the receipts. Xi for the day in order to obtain the account balance for the end of the day (cash flow forecast 405). A fortnightly forecast should be fine, but other durations could be used without departing from these inventions. In some embodiments, this number can be configured. The results of the calculations are put together in a time series of cash flows for the Xi futures.
    [0066] Next, data associated with interest rates for loans and investments, both current and historical, for each currency and account location is collected406, along with charges for bank transfers and other forms. money transfers. The spread between the sale and purchase of different currencies is also collected, both current and historical values, to determine the exchange rate charge.
    Time series of exchange rates and interest rates are run through an ARIMA model to predict interest rates and exchange rates for the next fortnight (or whatever window is used for payments and receipts).
    Each foreign currency account is analyzed 407 to maximize the value of all the foreign currency accounts. First, critical cash needs are identified by examining each foreign currency account for negative balances expected for the current day. These accounts need funding to cover the payments due in the current day. For example, in one embodiment, based on the previous behavior of a registered user (currency exchange, trade and finance), a machine learning algorithm provides a number of operations for final validation: currency change for have an appropriate balance between the main currency registered, the funding of accounts that are overdrawn, but also the movement of cash to interest-bearing accounts. In addition, following the previous behavior of the system will affect liquidity by counterparty to mitigate risk exposure. In case of negative interest existing on accounts in a specific currency, the system will offer to empty these accounts and exchange the currencies for currency accounts generating positive interest.
    Regarding the rejection of payment, the machine learning / artificial intelligence algorithm will offer predictive and prescriptive analyzes. For example, a payment that was rejected before will be reported and data correction will be offered, a payment type channel and route will also be scanned based on time limit and cost per channel bank and channel and route alternatives will be offered in order to reduce the cost of payments.
    With regard to cash exposure and currency exposure, the system will be able to predict variations and encourage risk mitigation actions, for example, but not exhaustively, to predict that the US dollar is rising against the British pound and to offer SWAP and Forward FX operations to mitigate exposure to losses.
    The set of foreign currency accounts is optimized using multivariate forecasting techniques. In one embodiment, the algorithm for optimizing the sum of the balances of all currency accounts is a mathematical formula. For each day of the future time series, negative balances are looked for in currency accounts. For each account with a negative balance, the accounts with positive balances are searched so as to obtain the currency with the lowest interest rate and the lowest cost for a transfer in the currency with the shortfall. Transfer instructions are then recorded and if necessary additional funds are sought. Then funding for the next account with a deficit is sought until all accounts are funded. Then, accounts with a positive balance are transferred to a transit account in that currency.
    In another embodiment, a machine learning algorithm such as K-means, Random Forrest or DensiCube (see US patent 9,489,627, issued to Jerzy Bala on November 8, 2016, and the US patent application 16 / 355,985, filed by Jerzy Bala and Paul Green on March 18, 2019) is used to try out various scenarios of cash transfers, borrowing and funding in order to optimize the balance at the end of the future forecast period. The fields can be currency, cash, interest in, interest out, transfer cost, and exchange rate to different currencies. A second dimension of the fields could be each day of the forecast period. And the attributes could be the currency accounts. Each combination of cash flows is calculated to determine a sum for the scenario (over the entire forecast period and all currency accounts), and the best-case cash flows are stored until a best scenario. be found. Rather than using an F score as a measure of quality, the sum of money could be used as a measure of quality. The best case scenario at the end of the analysis is then recorded as a cash flow plan. The output of the machine learning algorithm is a set of rules that dictate the cash flow. In some embodiments, this rules engine is used as a single day cash flow plan; in other cases, the rules can be used as a cash management plan for a short period, such as a week.
    In the operations of the machine learning module over a window of a longer period (for example a week, 15 days or a month), the objective is to find opportunities to invest the money in longer term, higher interest rate bonds, such as a 15 day certificate of deposit, rather than investing only in overnight transit accounts. The same goes for borrowing money or transferring cash between accounts: looking at longer term, better rates and lower transaction costs can be achieved.
    [0074] Once the financing plan is calculated, the treasury manager moves the money according to the plan to the appropriate account, by borrowing or investing the money as specified in the plan. Once funding is in place, the day's payments are made 409 from the appropriate foreign currency account.
    [0075] Due to the complexity of machine learning algorithms, special computation may be required to build and run the machine learning model described here. FIG. 5 shows such an embodiment. The user configures the PCM110 and monitors the status of the currency account200 on a personal computing device such as a personal computer, laptop, tablet, smartphone, monitor, or the like501,104,105. The personal computing device 501,104,105 communicates via a network 502 such as the Internet, a local area network, or perhaps via a direct interface with the server 503. Special Purpose Machine Learning Server503 is a high performance multi-core computing device (which includes floating point processing capabilities) with large data storage facilities 101a, 101b, 101c, 406 (hard drives, solid state drives, optical storage, RAID disks, etc.) to store transaction data for the PCM110. Since these databases 101a, 101b, 101c, 406 are continuously updated in some embodiments, this data must be kept online and accessible in order to be able to be updated. These data analyzes require a complex calculation with large data values, requiring the special purpose high performance server503. The server 503 is a high performance computer machine electrically connected to the network 502 and to the storage facilities 101a, 101b, 101c, 406. In addition, the server 503 requires connectivity to bank rails 504 in order to have secure and high performance access to banks 121-126 where foreign currency accounts are located. The above devices and operations, including their implementation, will be familiar to and understood by those skilled in the art.
    [0076] The above description of the embodiments, alternative embodiments and the specific examples are given by way of illustration and should not be considered as limiting. Further, many changes and modifications within the scope of the present embodiments can be made without departing from their spirit, and the present invention includes such changes and modifications.
    权利要求:
    Claims (20)
    [1]
    1. Cash management apparatus comprising:one or more bank rails connected to one or more banks; a special purpose server connected to one or more bank rails; and a plurality of data storage facilities connected to the special purpose server, the special purpose server configured to retrieve a set of payment and revenue transactions from the plurality of data storage facilities for a range. data of past dates, and perform ARIMA analysis in each foreign currency account on all payment and revenue transactions to create a revenue forecast and a payment forecast for a range of future dates, where the special-purpose server subtracts the payment forecast from the revenue forecast and adds a cash balance from the previous day to create a forecast cash balance time series for each currency account with the special purpose server configured to retrieve historical bank rate information and perform ARIMA analysis on historical bank rate information to create a time series forecast bank rate information, the special purpose server being configured to run an algorithm on the forecast cash balance time series for each foreign currency account and the forecast bank rate information time series to determine a set of optimal cash transfers between each foreign currency account and one or more pass-through accounts, then the special-purpose server executes instructions to make payments and cash transfers.
    [2]
    2. A cash management apparatus according to claim 1, wherein the algorithm is a machine learning algorithm.
    [3]
    3. A cash management apparatus according to claim 2, wherein the machine learning algorithm is K-means.
    [4]
    The cash management apparatus of claim 1, wherein the range of future dates is user configurable.
    [5]
    The cash management apparatus of claim 1, wherein the payment forecast is modified to incorporate the actual expected payments.
    [6]
    The cash management apparatus of claim 1, wherein the revenue forecast is modified to incorporate the actual incoming revenue.
    [7]
    The cash management apparatus of claim 1, wherein execution of the instructions for making payments and cash transfers is performed by transferring one or more files to the plurality of bank rails.
    [8]
    A cash management apparatus according to claim 1, wherein each foreign currency account further includes foreign currency accounts for a plurality of subsidiaries.
    [9]
    The cash management apparatus of claim 1, wherein the bank rate information is retrieved from one or more banks on the bank rails.
    [10]
    The cash management apparatus of claim 1, wherein the bank rate information includes interest rates, exchange rates and money transfer charges.
    [11]
    11. A method of managing cash flow in an organization, the method comprising:extracting a set of payment and revenue transactions from a plurality of data storage facilities for a given range of past dates for a plurality of currency accounts with a special purpose server that is connected the plurality of data storage facilities;performing an ARIMA analysis in each foreign currency account on the set of payment and revenue transactions, creating a revenue forecast and a payment forecast for a range of future dates;subtracting, with the special-purpose server, the forecast of payments from the forecast of revenue and adding a cash balance from the previous day to create a time series of the forecast cash balance for each foreign currency account;extracting historical bank rate information for special purposes;performing the ARIMA analysis on historical bank rate information to create a time series of forecast bank rate information;performing, using the special purpose server, an algorithm on the forecast cash balance time series for each foreign currency account and the forecast bank rate information time series to determine a set of cash transfers optimal cash flow between each foreign currency account and one or more transit accounts;the execution by the special purpose server of instructions for making payments and cash transfers.
    [12]
    12. The method of claim 11, wherein the algorithm is a machine learning algorithm.
    [13]
    13. The method of claim 12, wherein the machine learning algorithm is DensiCube.
    [14]
    14. The method of claim 11, wherein the given range of past dates is user configurable.
    [15]
    15. The method of claim 11, further comprising modifying the payment forecast by incorporating the actual expected payments.
    [16]
    16. The method of claim 11 further comprising modifying the forecast of revenue by incorporating the actual forecasted revenue.
    [17]
    17. The method of claim 11, wherein each foreign currency account further includes foreign currency accounts for a plurality of subsidiaries.
    [18]
    18. The method of claim 11, wherein the execution of the instructions for making payments and cash transfers is performed by transferring one or more files to a plurality of bank rails.
    [19]
    19. The method of claim 18, wherein the bank rate information is retrieved from one or more banks on the bank rails.
    [20]
    20. The method of claim 11, wherein the bank rate information includes interest rates, exchange rates, and money transfer charges.
    类似技术:
    公开号 | 公开日 | 专利标题
    Lee et al.2018|Fintech: Ecosystem, business models, investment decisions, and challenges
    Beck et al.2006|The basic analytics of access to financial services
    US10991040B2|2021-04-27|System and method for optimizing collateral management
    US8762246B2|2014-06-24|System and method for optimizing collateral management
    US20170024813A1|2017-01-26|System and Method for Provisioning Financial Transaction Between a Lender and a Borrower
    US11244388B2|2022-02-08|Methods and systems for assessing performance and risk in financing supply chain
    US20100076873A1|2010-03-25|Fee refund management
    US8036987B1|2011-10-11|Method and system for accounts payable prioritization and management
    US8768809B1|2014-07-01|Methods and systems for managing financial data
    Liu2021|FinTech and its disruption to financial institutions
    Pagano et al.2019|Implementation of blockchain technology in insurance contracts against natural hazards: a methodological multi-disciplinary approach
    Flatraaker2009|Mobile, Internet and electronic payments: the key to unlocking the full potential of the internal payments market
    Memminger et al.2018|Banking RegTechs to the rescue
    CH716790A2|2021-05-14|Apparatus and method for cash management using machine learning.
    Semenyuta et al.2019|Digital technologies in lending small and medium-size enterprises in Russia
    Agarwal et al.2021|AI-powered decision making for the bank of the future
    Allen2017|Handling Risk and Uncertainty in Portfolio Production Forecasting
    US20210264522A1|2021-08-26|System and method for optimizing data processing in cloud-based, machine learning environments through the use of self organizing map
    Harder2017|The future of contracts in a digital economy: The impact of smart contracts on transaction costs in financial markets
    US20200279198A1|2020-09-03|Cash forecast system, apparatus, and method
    Peat et al.2017|Fintech: hype or reality?
    US20210406733A1|2021-12-30|System and methods for prediction communication performance in networked systems
    Mersch2016|Making Europe’s financial market infrastructure a bulwark of financial stability
    Mečionytė2022|Research on Feasibility of Implementation of Open Banking as a Product of Financial Technologies in Lithuanian Market
    WO2021198615A1|2021-10-07|Collaborative electronic platform for predicting non-payments for companies and associated method
    同族专利:
    公开号 | 公开日
    US20210142399A1|2021-05-13|
    GB2592559A|2021-09-08|
    GB201918345D0|2020-01-29|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    法律状态:
    优先权:
    申请号 | 申请日 | 专利标题
    US16/680,652|US20210142399A1|2019-11-12|2019-11-12|international cash management software using machine learning|
    [返回顶部]