Artificial intelligence for management control: necessity or utopia?
Author: David Hatchuel
Artificial intelligence (AI) is now at the heart of organizational strategy. Controllers have the opportunity to create value by making the most of the operational and financial data available in all the company's information systems. How can we take advantage of this technological revolution and put AI at the service of "augmented" management control?
Data producer or Business Partner?
The role of management controllers is to help managers better understand and manage company performance. In practice, they are often faced with the tedious task of producing a large number of reports. The tools used communicate with each other with difficulty, due to a lack of data quality or unsynchronized repositories. The management controller is often forced to multiply reconciliation work and consistency checks, with no apparent added value. As a result, controllers generally spend 80% of their time producing figures and 20% on analysis. Too often overwhelmed by the analysis of past performance, they lack the time to build and refine financial forecasts.
AI precursors already widely adopted in business
Marketing and sales functions already embraced AI a few years ago. Relationship marketing relies on the use ofdeep learning technologies to optimize the understanding of publications on social networks. Banks and insurance companies are deploying chatbots (conversational agents) and using AI to detect fraud risks.
"The wave of AI appropriation has largely passed: the era of industrialization has already begun!"
Jean-Philippe Desbiolles - Vice-President - Data, Cognitive & AI - Financial Services at IBM
In accounting functions, there are already many use cases for robots and AI:
- Accounting robotization: Accounts payable processing continues to be automated using RPA(Robotic Process Automation) software, which can perform many of the tasks involved in recording an invoice: connecting to the e-mail system, opening the e-mail containing an invoice, reading the information, connecting to the ERP system and posting. Other time-consuming accounting processes, such as managing expense claims and processing pay slips, also use RPA and optical recognition processes. Consolidation departments are also deploying robots to industrialize data collection and the feeding of reporting packages. These solutions enable accounting departments to focus their teams on higher value-added tasks.
"Artificial intelligence, a historic opportunity to de-automate human work: it enables us to develop truly human capabilities".
Cédric Villani
- Data analytics: companies' accounting and financial data are being analyzed in greater depth. Major audit firms are now using big data and data analytics tools to review FECs (accounting entry files), which often comprise several million lines. Exploiting data from the FEC is a major challenge in moving from an audit based on sampling to an audit based on exhaustive data. For example, for sales audits, substantive tests were traditionally carried out on the basis of a sample of a few hundred sales invoices. Thanks to new technologies, auditors can now analyze 100% of sales transactions and perform substantive tests only on anomalies or inconsistencies identified.
- Process mining: process mining is used within accounting departments to "x-ray" cross-functional processes, by analyzing data traced in the event logs of the various information systems. For example, when analyzing the "Purchasing" process (from purchase request to supplier invoice settlement), process mining enables us to map the various transactions recorded by operational staff and accountants in the ERP system, and to highlight the hidden costs generated by non-optimized processes. One example is the Celonis platform, which uses an artificial intelligence engine to improve all company management processes.
Controlling can benefit from artificial intelligence
Management control must be able to model and project the company's future results. It is precisely in the field of predictive analysis that AI can make a major breakthrough.
Predictive analysis, the new playground for management controllers
The main management control solutions deployed today are :
- EPM(Enterprise Performance Management) tools: reporting on actual results, budget preparation, financial forecasting and variance analysis.
- BI(Business Intelligence) tools: data extraction and visualization, and construction of dynamic dashboards
The most advanced EPM / BI solutions already offer advanced functionalities that enable :
- Enhancing the quality of data and repositories
- Model and simulate revenues and costs using automatically detected drivers: number of production runs, number of orders placed, number of deliveries, number of customers, etc.
- The implementation of automated controls to guarantee the consistency of the forecast model
Solutions already featuring Machine Learning and AI modules include Tagetik, SAP Analytics Cloud, Anaplan, Onestream, IBM Planning Analytics, Pyramid Analytics, Oracle FCCS and PBCS...
These predictive analysis tools will enable management controllers to fully play their role as "Business Partners". In concrete terms, the AI will propose its statistical projections, and the controller will be able to draw inspiration from them to adjust his own forecasts, carried out in a more conventional way.
By integrating Big Data, AI can facilitate the processing of structured data (financial data in a report) and unstructured data (photos, likes on social networks, weather trends...) and uncover correlations previously unsuspected in conventional financial analyses. How can we legitimize the decisions proposed by the machine? A real challenge for the "augmented" management controller!
In controlling, as in many other fields, AI must be seen as an auxiliary, not as a system that can replace human beings. As Jean-Philippe Desbiolles points out in his book L'AI sera ce que tu en feras: "an AI knows what we don't know, anticipates events we hadn't foreseen, and proposes options we hadn't thought of".
What complexity factors need to be addressed?
- Feeding AI with vast amounts of data
Machine learning requires a considerable amount of data, as AI is based on a "data-to-model" approach. From a large volume of data, AI attempts to establish models. For example, a machine will learn what a cat is by analyzing thousands of cat images, whereas a human can learn by observing three cats. The challenge for management control is to rely on reliable corporate data that incorporates five to ten years of historical data to feed predictive models and facilitate the detection of seasonality and long-term trends. In practice, collecting management data to feed AI means replaying and aligning long- and medium-term plans, budgets and reforecasts made over several years in numerous Excel files. It will also be necessary to deal with the impact of changes in the company's scope of consolidation (acquisitions, disposals, mergers), and changes in finance department processes and tools.
- Ensuring data quality / quantifying uncertainty
The machine is not capable of judging the quality of the data. To deploy effective AI, reliable data must be collected upstream. This reliability can be achieved by classifying or labeling data. For example, to enable the machine to recognize an image, we need to indicate its content using labels. This stage is often entirely manual, and human errors or interpretations are possible.
- Control algorithmic bias
Algorithmic biases are "parasites" that can disrupt the AI's reasoning. For example, in visual recognition, these "noises" are nuisances related to changes in object position in the image, changes in point of view, lighting problems or changes in facial expression. For example, if an autonomous car is confronted with a stop sign and a sticker is stuck on it (a bias), the machine may interpret it as another road sign and adopt an inappropriate trajectory. AI for management control will not be immune to these algorithmic biases, which teams will need to be able to detect and correct.
Anticipating the shift to AI!
Deploying AI within a finance department requires a structured project management approach: scoping requirements, selecting the most suitable platform, gathering data, implementing and deploying the solution, training and change management...
"If you want to build a ship, don't just bring together men, wood and iron, but tell them about the seas that the ship will sail until they dream of it...". - Antoine de Saint-Exupéry
An "AI champion" will need to be appointed during the project implementation period. This person doesn't necessarily have to be an AI expert, because today, knowing how to handle data and having a grounding in programming are all it takes to quickly create AI models.
Taking an interest in and preparing for technological change is a priority for management control. The first projects on the horizon will help define the best practices and most appropriate approaches for exploring the new frontiers of management control...
ABOUT THE AUTHORS
David Hatchuel is a partner at Emerson Audit & Conseil. A graduate of the Dauphine Master's in Banking and Finance and a chartered accountant, he supports the digital transformation of finance departments in their projects to upgrade information systems, improve the efficiency of finance functions and manage performance.
ABOUT EMERSON AUDIT & CONSULTING
With a team of 120 consultants, Emerson audit & conseil covers the fields of public accounting, auditing and consulting. The fast-growing firm supports Finance Departments of all sizes and in a wide variety of sectors, through functional assistance and project management for the transformation and improvement of organizations and information systems.
Author: David Hatchuel
Artificial intelligence (AI) is now at the heart of organizational strategy. Controllers have the opportunity to create value by making the most of the operational and financial data available in all the company's information systems. How can we take advantage of this technological revolution and put AI at the service of "augmented" management control?
Data producer or Business Partner?
The role of management controllers is to help managers better understand and manage company performance. In practice, they are often faced with the tedious task of producing a large number of reports. The tools used communicate with each other with difficulty, due to a lack of data quality or unsynchronized repositories. The management controller is often forced to multiply reconciliation work and consistency checks, with no apparent added value. As a result, controllers generally spend 80% of their time producing figures and 20% on analysis. Too often overwhelmed by the analysis of past performance, they lack the time to build and refine financial forecasts.
AI precursors already widely adopted in business
Marketing and sales functions already embraced AI a few years ago. Relationship marketing relies on the use ofdeep learning technologies to optimize the understanding of publications on social networks. Banks and insurance companies are deploying chatbots (conversational agents) and using AI to detect fraud risks.
"The wave of AI appropriation has largely passed: the era of industrialization has already begun!"
Jean-Philippe Desbiolles - Vice-President - Data, Cognitive & AI - Financial Services at IBM
In accounting functions, there are already many use cases for robots and AI:
- Accounting robotization: Accounts payable processing continues to be automated using RPA(Robotic Process Automation) software, which can perform many of the tasks involved in recording an invoice: connecting to the e-mail system, opening the e-mail containing an invoice, reading the information, connecting to the ERP system and posting. Other time-consuming accounting processes, such as managing expense claims and processing pay slips, also use RPA and optical recognition processes. Consolidation departments are also deploying robots to industrialize data collection and the feeding of reporting packages. These solutions enable accounting departments to focus their teams on higher value-added tasks.
"Artificial intelligence, a historic opportunity to de-automate human work: it enables us to develop truly human capabilities".
Cédric Villani
- Data analytics: companies' accounting and financial data are being analyzed in greater depth. Major audit firms are now using big data and data analytics tools to review FECs (accounting entry files), which often comprise several million lines. Exploiting data from the FEC is a major challenge in moving from an audit based on sampling to an audit based on exhaustive data. For example, for sales audits, substantive tests were traditionally carried out on the basis of a sample of a few hundred sales invoices. Thanks to new technologies, auditors can now analyze 100% of sales transactions and perform substantive tests only on anomalies or inconsistencies identified.
- Process mining: process mining is used within accounting departments to "x-ray" cross-functional processes, by analyzing data traced in the event logs of the various information systems. For example, when analyzing the "Purchasing" process (from purchase request to supplier invoice settlement), process mining enables us to map the various transactions recorded by operational staff and accountants in the ERP system, and to highlight the hidden costs generated by non-optimized processes. One example is the Celonis platform, which uses an artificial intelligence engine to improve all company management processes.
Controlling can benefit from artificial intelligence
Management control must be able to model and project the company's future results. It is precisely in the field of predictive analysis that AI can make a major breakthrough.
Predictive analysis, the new playground for management controllers
The main management control solutions deployed today are :
- EPM(Enterprise Performance Management) tools: reporting on actual results, budget preparation, financial forecasting and variance analysis.
- BI(Business Intelligence) tools: data extraction and visualization, and construction of dynamic dashboards
The most advanced EPM / BI solutions already offer advanced functionalities that enable :
- Enhancing the quality of data and repositories
- Model and simulate revenues and costs using automatically detected drivers: number of production runs, number of orders placed, number of deliveries, number of customers, etc.
- The implementation of automated controls to guarantee the consistency of the forecast model
Solutions already featuring Machine Learning and AI modules include Tagetik, SAP Analytics Cloud, Anaplan, Onestream, IBM Planning Analytics, Pyramid Analytics, Oracle FCCS and PBCS...
These predictive analysis tools will enable management controllers to fully play their role as "Business Partners". In concrete terms, the AI will propose its statistical projections, and the controller will be able to draw inspiration from them to adjust his own forecasts, carried out in a more conventional way.
By integrating Big Data, AI can facilitate the processing of structured data (financial data in a report) and unstructured data (photos, likes on social networks, weather trends...) and uncover correlations previously unsuspected in conventional financial analyses. How can we legitimize the decisions proposed by the machine? A real challenge for the "augmented" management controller!
In controlling, as in many other fields, AI must be seen as an auxiliary, not as a system that can replace human beings. As Jean-Philippe Desbiolles points out in his book L'AI sera ce que tu en feras: "an AI knows what we don't know, anticipates events we hadn't foreseen, and proposes options we hadn't thought of".
What complexity factors need to be addressed?
- Feeding AI with vast amounts of data
Machine learning requires a considerable amount of data, as AI is based on a "data-to-model" approach. From a large volume of data, AI attempts to establish models. For example, a machine will learn what a cat is by analyzing thousands of cat images, whereas a human can learn by observing three cats. The challenge for management control is to rely on reliable corporate data that incorporates five to ten years of historical data to feed predictive models and facilitate the detection of seasonality and long-term trends. In practice, collecting management data to feed AI means replaying and aligning long- and medium-term plans, budgets and reforecasts made over several years in numerous Excel files. It will also be necessary to deal with the impact of changes in the company's scope of consolidation (acquisitions, disposals, mergers), and changes in finance department processes and tools.
- Ensuring data quality / quantifying uncertainty
The machine is not capable of judging the quality of the data. To deploy effective AI, reliable data must be collected upstream. This reliability can be achieved by classifying or labeling data. For example, to enable the machine to recognize an image, we need to indicate its content using labels. This stage is often entirely manual, and human errors or interpretations are possible.
- Control algorithmic bias
Algorithmic biases are "parasites" that can disrupt the AI's reasoning. For example, in visual recognition, these "noises" are nuisances related to changes in object position in the image, changes in point of view, lighting problems or changes in facial expression. For example, if an autonomous car is confronted with a stop sign and a sticker is stuck on it (a bias), the machine may interpret it as another road sign and adopt an inappropriate trajectory. AI for management control will not be immune to these algorithmic biases, which teams will need to be able to detect and correct.
Anticipating the shift to AI!
Deploying AI within a finance department requires a structured project management approach: scoping requirements, selecting the most suitable platform, gathering data, implementing and deploying the solution, training and change management...
"If you want to build a ship, don't just bring together men, wood and iron, but tell them about the seas that the ship will sail until they dream of it...". - Antoine de Saint-Exupéry
An "AI champion" will need to be appointed during the project implementation period. This person doesn't necessarily have to be an AI expert, because today, knowing how to handle data and having a grounding in programming are all it takes to quickly create AI models.
Taking an interest in and preparing for technological change is a priority for management control. The first projects on the horizon will help define the best practices and most appropriate approaches for exploring the new frontiers of management control...
ABOUT THE AUTHORS
David Hatchuel is a partner at Emerson Audit & Conseil. A graduate of the Dauphine Master's in Banking and Finance and a chartered accountant, he supports the digital transformation of finance departments in their projects to upgrade information systems, improve the efficiency of finance functions and manage performance.
ABOUT EMERSON AUDIT & CONSULTING
With a team of 120 consultants, Emerson audit & conseil covers the fields of public accounting, auditing and consulting. The fast-growing firm supports Finance Departments of all sizes and in a wide variety of sectors, through functional assistance and project management for the transformation and improvement of organizations and information systems.
Author: David Hatchuel
Artificial intelligence (AI) is now at the heart of organizational strategy. Controllers have the opportunity to create value by making the most of the operational and financial data available in all the company's information systems. How can we take advantage of this technological revolution and put AI at the service of "augmented" management control?
Data producer or Business Partner?
The role of management controllers is to help managers better understand and manage company performance. In practice, they are often faced with the tedious task of producing a large number of reports. The tools used communicate with each other with difficulty, due to a lack of data quality or unsynchronized repositories. The management controller is often forced to multiply reconciliation work and consistency checks, with no apparent added value. As a result, controllers generally spend 80% of their time producing figures and 20% on analysis. Too often overwhelmed by the analysis of past performance, they lack the time to build and refine financial forecasts.
AI precursors already widely adopted in business
Marketing and sales functions already embraced AI a few years ago. Relationship marketing relies on the use ofdeep learning technologies to optimize the understanding of publications on social networks. Banks and insurance companies are deploying chatbots (conversational agents) and using AI to detect fraud risks.
"The wave of AI appropriation has largely passed: the era of industrialization has already begun!"
Jean-Philippe Desbiolles - Vice-President - Data, Cognitive & AI - Financial Services at IBM
In accounting functions, there are already many use cases for robots and AI:
- Accounting robotization: Accounts payable processing continues to be automated using RPA(Robotic Process Automation) software, which can perform many of the tasks involved in recording an invoice: connecting to the e-mail system, opening the e-mail containing an invoice, reading the information, connecting to the ERP system and posting. Other time-consuming accounting processes, such as managing expense claims and processing pay slips, also use RPA and optical recognition processes. Consolidation departments are also deploying robots to industrialize data collection and the feeding of reporting packages. These solutions enable accounting departments to focus their teams on higher value-added tasks.
"Artificial intelligence, a historic opportunity to de-automate human work: it enables us to develop truly human capabilities".
Cédric Villani
- Data analytics: companies' accounting and financial data are being analyzed in greater depth. Major audit firms are now using big data and data analytics tools to review FECs (accounting entry files), which often comprise several million lines. Exploiting data from the FEC is a major challenge in moving from an audit based on sampling to an audit based on exhaustive data. For example, for sales audits, substantive tests were traditionally carried out on the basis of a sample of a few hundred sales invoices. Thanks to new technologies, auditors can now analyze 100% of sales transactions and perform substantive tests only on anomalies or inconsistencies identified.
- Process mining: process mining is used within accounting departments to "x-ray" cross-functional processes, by analyzing data traced in the event logs of the various information systems. For example, when analyzing the "Purchasing" process (from purchase request to supplier invoice settlement), process mining enables us to map the various transactions recorded by operational staff and accountants in the ERP system, and to highlight the hidden costs generated by non-optimized processes. One example is the Celonis platform, which uses an artificial intelligence engine to improve all company management processes.
Controlling can benefit from artificial intelligence
Management control must be able to model and project the company's future results. It is precisely in the field of predictive analysis that AI can make a major breakthrough.
Predictive analysis, the new playground for management controllers
The main management control solutions deployed today are :
- EPM(Enterprise Performance Management) tools: reporting on actual results, budget preparation, financial forecasting and variance analysis.
- BI(Business Intelligence) tools: data extraction and visualization, and construction of dynamic dashboards
The most advanced EPM / BI solutions already offer advanced functionalities that enable :
- Enhancing the quality of data and repositories
- Model and simulate revenues and costs using automatically detected drivers: number of production runs, number of orders placed, number of deliveries, number of customers, etc.
- The implementation of automated controls to guarantee the consistency of the forecast model
Solutions already featuring Machine Learning and AI modules include Tagetik, SAP Analytics Cloud, Anaplan, Onestream, IBM Planning Analytics, Pyramid Analytics, Oracle FCCS and PBCS...
These predictive analysis tools will enable management controllers to fully play their role as "Business Partners". In concrete terms, the AI will propose its statistical projections, and the controller will be able to draw inspiration from them to adjust his own forecasts, carried out in a more conventional way.
By integrating Big Data, AI can facilitate the processing of structured data (financial data in a report) and unstructured data (photos, likes on social networks, weather trends...) and uncover correlations previously unsuspected in conventional financial analyses. How can we legitimize the decisions proposed by the machine? A real challenge for the "augmented" management controller!
In controlling, as in many other fields, AI must be seen as an auxiliary, not as a system that can replace human beings. As Jean-Philippe Desbiolles points out in his book L'AI sera ce que tu en feras: "an AI knows what we don't know, anticipates events we hadn't foreseen, and proposes options we hadn't thought of".
What complexity factors need to be addressed?
- Feeding AI with vast amounts of data
Machine learning requires a considerable amount of data, as AI is based on a "data-to-model" approach. From a large volume of data, AI attempts to establish models. For example, a machine will learn what a cat is by analyzing thousands of cat images, whereas a human can learn by observing three cats. The challenge for management control is to rely on reliable corporate data that incorporates five to ten years of historical data to feed predictive models and facilitate the detection of seasonality and long-term trends. In practice, collecting management data to feed AI means replaying and aligning long- and medium-term plans, budgets and reforecasts made over several years in numerous Excel files. It will also be necessary to deal with the impact of changes in the company's scope of consolidation (acquisitions, disposals, mergers), and changes in finance department processes and tools.
- Ensuring data quality / quantifying uncertainty
The machine is not capable of judging the quality of the data. To deploy effective AI, reliable data must be collected upstream. This reliability can be achieved by classifying or labeling data. For example, to enable the machine to recognize an image, we need to indicate its content using labels. This stage is often entirely manual, and human errors or interpretations are possible.
- Control algorithmic bias
Algorithmic biases are "parasites" that can disrupt the AI's reasoning. For example, in visual recognition, these "noises" are nuisances related to changes in object position in the image, changes in point of view, lighting problems or changes in facial expression. For example, if an autonomous car is confronted with a stop sign and a sticker is stuck on it (a bias), the machine may interpret it as another road sign and adopt an inappropriate trajectory. AI for management control will not be immune to these algorithmic biases, which teams will need to be able to detect and correct.
Anticipating the shift to AI!
Deploying AI within a finance department requires a structured project management approach: scoping requirements, selecting the most suitable platform, gathering data, implementing and deploying the solution, training and change management...
"If you want to build a ship, don't just bring together men, wood and iron, but tell them about the seas that the ship will sail until they dream of it...". - Antoine de Saint-Exupéry
An "AI champion" will need to be appointed during the project implementation period. This person doesn't necessarily have to be an AI expert, because today, knowing how to handle data and having a grounding in programming are all it takes to quickly create AI models.
Taking an interest in and preparing for technological change is a priority for management control. The first projects on the horizon will help define the best practices and most appropriate approaches for exploring the new frontiers of management control...
ABOUT THE AUTHORS
David Hatchuel is a partner at Emerson Audit & Conseil. A graduate of the Dauphine Master's in Banking and Finance and a chartered accountant, he supports the digital transformation of finance departments in their projects to upgrade information systems, improve the efficiency of finance functions and manage performance.
ABOUT EMERSON AUDIT & CONSULTING
With a team of 120 consultants, Emerson audit & conseil covers the fields of public accounting, auditing and consulting. The fast-growing firm supports Finance Departments of all sizes and in a wide variety of sectors, through functional assistance and project management for the transformation and improvement of organizations and information systems.