You may also be interested in:

Articles

Why Are Artificial Intelligence and Machine Learning Important for FP&A?

  • By Cameron Koo
  • Published: 7/11/2018

As FP&A professionals, we are constantly diving into data to develop insights. We need to be aware of current and emerging tools as they get more sophisticated and powerful, so we can become more efficient in our work. Artificial intelligence (AI) is really a set of various tools, and their applicability to a specific task relates to the type of data you want to analyze.

The data world is divided into two camps—structured data, which is stored in structured and ordered databases (think rows and columns), and unstructured data, such as texts, emails, videos and websites. Different data calls for different AI tools. In recent years, machine learning (ML) and deep learning (DL) have gathered significant coverage, especially for finance, because of the applicability to create predictive models based on structured data. ML and DL use statistical science and probability on historical data. They’re great for processing structured datasets to find trends and create forecasts.

Autonomous learning complements ML and DL in analyzing dynamic unstructured text. It uses symbolic logic to analyze textual data and discover relevant subjects and relationships based on context. The use cases for finance to consider autonomous learning may not have been fully explored.

ML and DL

Organizations use solutions built on ML and DL to drive automation, lower costs, increase efficiency and optimize outcomes. ML and DL apply statistical science and probability on large datasets to find hidden insights using curve-fitting algorithms, regression and conditional probability. AI solutions built on ML and DL are highly effective with structured datasets (e.g., transactions, traffic patterns, etc.). Some common applications of ML and DL include SPAM email filtering, product recommendation engines, supply chain management and sales forecasting.

ML and DL, however, are not perfect solutions. ML and DL models are costly to build and maintain. They require large amounts of historical data on a narrow problem domain and pre-defined variables. Models are not interchangeable. A model built for one domain cannot be reused for another without retraining. Moreover, ML and DL does not work when presented with new and unknown information. For this reason, ML and DL produce limited results with dynamic unstructured datasets like natural language communications (e.g. email, IM chat logs).

The need for an unstructured data AI

Today, over 80 percent of an organization’s data assets consist of unstructured textual data and it is projected to grow exponentially as a result of digital transformation. Text drives nearly every business function in an organization. It connects business processes, enables team collaboration and exists as documents that articulate opinions, customer feedback, industry news and trends, competitive intelligence, risk, policy, regulation, etc. Organizations need a method to automate reading and comprehension so that business teams can get timely and precise information.

Autonomous learning automates the process of reading and comprehension of textual data. It enables humans to interact with machines to get the relevant information in a fraction of the time. Autonomous learning uses symbolic logic to discover and map subjects, concepts, relationships and context on-the-fly. The acquired knowledge is continuously accumulated into an ever-growing knowledgebase for reference and inference. It is scalable and lightweight and has no dependency on big data or historical data.

By automating reading and comprehension of unstructured text, autonomous learning enables an organization to achieve end-to-end visibility into its operations. It allows the finance department to complement quantitative forecasting with qualitative analysis and insight into underlying context behind the number. Finance can draw upon the collective knowledge embedded in textual data with the same efficacy as quantitative analysis.

Here are some use cases in which the basic research tool can be applied:  

Forecasting and variance from forecasts: Discover and qualify market signals that relate to business drivers

  • A consumer goods company wants to know whether the recent “slime” craze among children is before or after its peak in order to determine whether to make more glue (a key ingredient).
  • How will recent changes in the tariff outlook impact interest rates, and therefore net interest rates for banks? Scan the proposed NAFTA legislation, trade-deals, and FOMC minutes.
  • If Medicare and Social Security run out of funds earlier than anticipated, what is the impact on life expectancy, housing defaults and therefore insurance claims?
  • Link internal operations to external events to help explain variance.

Risk and opportunity analysis: Qualify signals and measure their progression over time across different information mediums

  • News reports of a geopolitical event where your company is considering a planned expansion.
  • A new regulator is being considered for a key government post that oversees your industry; what is her track record on matters related to your company? Scan government and committee records.
  • Small startups are entering your market space.
  • Develop scenarios to stress test your business.

Peer monitoring: Analyze information streams of competing firms for insights into activity and changes to strategy

  • Major competitors are getting a lift from a new area of growth or increasing/decreasing investments in geographies. Scan 10Ks, press releases, analyst reports.
  • Peers are expecting to be impacted by market forces and lowering income forecast.

Check out AFP's interactive Roadmap to The Future of Finance to learn how you can master the new capabilities and models impacting finance.

Cameron Koo is co-founder, head of business development and partnerships for SiteFocus Inc. Reach him at kkoo@sitefocus.com.



Top Line Modeling
On October 18 & 25, discover key techniques and practices for building revenue specific financial models that account for seasonality and uncertainty. Through hands-on examples, build financial models that produce baseline and dynamic outputs in response to real-world scenarios.
Learn More

Copyright © 2018 Association for Financial Professionals, Inc.
All rights reserved.