Interviews conducted by Forrester suggest that artificial intelligence (AI) has not yet helped transform finance, yet it is advancing rapidly in some areas.
Procure-to-pay (P2P), for example, uses natural language processing (NLP) and machine learning (ML) and has shown immediate returns, while order-to-cash and audit analytics show near-term benefits from AI. Furthermore, predictive analytics can augment basic business intelligence (BI) reporting for financial planning.
Four ways AI is empowering finance and accounting
Audit analytics, procure to pay, order to cash and financial planning are four finance and accounting (F&A) processes where the AI technology required to elevate the process already exists. There is also an active community of technology providers and customer references indicate strong progress. Forrester gives these four use cases strong scores for adoption, such as a manageable skills gap, stable data and clear-cut business outcomes.
Starting with audit analytics, auditors tend to spend too much time buried in compliance checklists and creating reports that few people read, with little time to seek anomalies in every transaction. Rather than manually sampling data points, Forrester says machine learning is being used for risk assessment of transactions.
The member-based industry association American Institute of Certified Public Accountants (AICPA) is developing guidance for ML in the audit function. Mature audit support providers such as Thomson Reuters and Wolters Kluwer, as well as emerging companies like Caseworks Cloud and MindBridge, are embedding AI into their audit platforms.
Technology readiness is high, with mature ML, while NLP extraction brings unstructured content such as email into play. The adoption profile is strong as well, with few governance issues, high business value and strong disruptive potential. However, training auditors in ML aspects exposes a current skills gap.
Procure to pay
Looking at the procure-to-pay (P2P) process, Forrester found that P2P can take advantage of ML to standardise and analyse spend, contract, market and supplier data. Augmented BI can isolate payments that once led to late-payment penalties and can surface invoice exceptions, classify spending into categories for follow-up, onboard new suppliers faster and automatically detect fraud.
A prominent area is invoice processing, where level 1 capture, optical character recognition (OCR) and workflow automation patterns have applied for decades. Early solutions were template-based, where extraction rules aligned with a specific invoice or a purchase order template. New approaches use NLP to provide template-free and zone-free extraction. To ensure digitisation quality, each extraction can be stamped with a certainty level.
ML can handle complex document structures more easily, without preconfiguration. NLP and traditional ML are mature, providing a strong technology readiness score. The adoption profile is also strong due to a high potential for disruption and a mix of stable data from semi-structured forms and data.
Order to cash
Order to cash is another untapped candidate for AI-driven automation. Cash is the lifeblood of most enterprises, yet it remains underserved by the latest automation practices, particularly when compared with P2P.
In most cases, accounts receivable (AR) invoice automation software generates the customer invoice in formats like CXML (commerce XML), ebXML (electronic business XML) and Edifact, and tracks the status, while F&A handles the cash. Modern order-to-cash solutions elevate the role of the AR professional, as many tasks move to AI-based bots that can take over email communications or build a collection letter based on auto-classification, core system data and dispute stage.
Analytics will control cash applications. AI will drive automated payment lifecycles, credit management and predictive remittance forecasting. The adoption profile is strong due to a clear business outcome, such as improving cash performance. Rules-based workflow and decisions are starting to give way to AI-based ones, but the business value is now moderate. The technology is ready today, with ML, robotic process automation (RPA) and text analytics ready to help.
A fourth use case, financial planning and analysis, is starting to move beyond Excel. Financial analytics has strong potential for AI support, yet most finance departments depend on Excel or basic reporting from specialist supplier platforms.
However, future budget planning and forecasting will use simulation, optimisation and ML-based statistical modelling that link corporate strategy to execution. One example is Vena Solutions, which offers a Microsoft-oriented F&A product with Power BI embedded to provide an easy path to predictive analytics and machine learning (PAML).
Four areas where AI in finance and accounting needs further development
Contract analytics has wide potential across a number of use cases. Contract analytics isn’t a core function of finance and accounting, but it is of increasing interest to chief financial officers (CFOs) and their staff.
The primary use of AI is to automate the importing and metadata tagging of legacy and third-party contracts. Platforms such as ContractPodAI and Icertis and specialist AI providers like Corticol.io are embedding AI functions in contract lifecycle management (CLM).
ML can help assess risks and anomalies within the overall contract portfolio, find contracts with wording related to a new issue or topic such as Brexit or new tax laws, and feed CLM or other workflow automation platforms to support service-level agreements (SLAs) and other terms and conditions. Forrester is seeing early progress in providing chatbots to help assemble a contract draft.
The primary AI building block is text analytics, providing a moderate to strong technology readiness score. Forrester’s adoption profile score shows a high business value but less-than-clear outcomes, and the variety of document formats makes data less than stable.
AI can also be used in accounts reconciliation to solve data issues. Many tasks in finance and accounting require two or even three sets of records to agree, particularly when money leaves a bank account. Prepaid expenses, bad debts, fixed assets, cash accounts, and general ledger and sub-ledger tasks are typical targets to reconcile. Missing or lost transactions, unreconciled accounts, or improper use of roll-forwards are typical.
ML can handle a wide variety of structured data sources in many formats (CSV, XML, SQL or NoSQL) where it can “learn” the data sources and patterns, with data control rules in a central location. Most reconciliations deal with only two data sources, but AI can expand this to multiple sources. RPA bots help extract data, provide data entry support and run an approval process.
In accounts reconciliation, Forrester scores technology readiness as high, but the adoption profile is average due to poor data stability and moderate business and disruption value. Reconciliation is often a necessary subset function of the closing process.
Monthly and quarterly close automation relies on basic workflow. Automating the closing process is top of mind for many F&A departments. A well-run close is a sign of a well-run company. Transparency, speed, accuracy and meeting reporting deadlines are top concerns. Close automation must integrate with enterprise apps, spreadsheets and various accounting systems to document relevant data and identify inconsistencies.
AI has the potential to gather data from different sources, collate and merge it, accelerate the monthly process and be more accurate. Automation centres on checklists and individual tasks, tracking the close process, timelines and approvals.
Forrester rates closing the books with an average technology readiness score but a high adoption profile, driven by high business value and disruptive potential – although acquisitions and core system conversions make the data less stable.
Another area where AI and automation can be used is in expense management. Forrester hasn’t seen a strong push to use advanced forms of AI in this area. Intelligent automation centred on RPA has been the primary enhancement. For example, one federal agency is using RPA bots to audit line-item detail, which humans formerly did. When travel ramps up again, this back-office function will once more be a staple of any well-run enterprise. Major market players such as SAP Concur haven’t pushed AI, nor have CFOs looking for efficiency or anomaly detection.
Forrester scores technology readiness for the use of AI in expense management as high due to dependence on RPA bots and traditional ML. However, despite clear outcomes, the adoption profile is lower due to low perceived business value, minimal disruptive potential and less-than-stable data.
Summing up, Forrester notes that many finance and accounting processes are fraught with unnecessary variation. AI works best against standard and repeatable actions. But extra process steps, offline behaviour, rogue spreadsheets and personal shortcuts are common. This lack of standardisation of tasks across firms prevents software providers from building targeted and easy-to-implement AI for finance and accounting processes.
This article is based on an excerpt of Forrester’s “AI in finance and accounting” report. Craig Le Clair is a vice-president and principal analyst at Forrester.