Employee working on a laptop in a gift shop

How artificial intelligence will empower the tax function

Related topics

As AI is set to transform tax into an innovation hub, businesses must make sure they are ready to reap its rewards.


Three questions to ask

  • Is data being captured in a way that AI can effectively leverage?
  • What are the strategic benefits of AI in tax?
  • How can a company best begin its AI journey?

Artificial intelligence (AI) is everywhere. At any given time, it is probably just inches from our fingertips. AI is in smartphones, fitness trackers, TVs and smart speakers, and we benefit from it when shopping online or searching for new films on streaming services.

The unfolding potential of AI within the tax function is large. Among other benefits, AI can process thousands of transactions within percentage points of perfection in just seconds. This will liberate tax practitioners to focus on value-added areas such as tax approached.

“Right now, tax teams still spend way too much time on ‘copy, paste and attach,’” Daren Campbell, EY Americas Tax Innovation Leader says. To underscore that point, the EY 2020 Global Tax Technology and Transformation Survey reveals that a typical tax team spends 40-70% of its time gathering and manipulating data, when this could be done in a fraction of the time by AI.

“It won’t be long until AI frees up tax teams to focus on overall business strategy and the tax considerations of that strategy,” Campbell says. “I can also foresee a time when AI will make data-driven suggestions and recommendations showing businesses how they can achieve their strategic tax goals.”

1

Chapter 1

The barriers to AI adoption

Unstructured data is a key roadblock to AI, which needs standardized data to capitalize on its potential.

So why aren’t more tax functions already using AI? What’s holding them back?

According to Campbell, the biggest barrier to AI adoption is data. There’s enough of it – in fact, many tax functions are overwhelmed by data – but it isn’t being handled in a structured, AI-friendly way. It’s often captured inaccurately, inconsistently and in the wrong format.

This disparity in data handling is so prevalent because tax functions are made up of teams and practitioners collecting data in their own way with few standardized processes. This inconsistency is a serious problem for AI, which needs “clean” standardized data to identify data patterns and learn from them.

In fact, 40% of respondents to the survey said they still use disconnected spreadsheets, data collection packages and/or consolidation ledgers as their primary data sources, rather than an enterprise resource planning (ERP) system or data warehouse.



Disparity in data handling is so prevalent because tax functions are made up of teams and practitioners collecting data in their own way with few standardized processes.



Campbell cites an example where human-captured data was used to train an AI model to classify large volumes of sales-and-use tax data. Vendors, descriptions and accounts were coded in two completely different ways and assigned to two different categories. This totally confused the AI model and undermined the process.

“This scenario isn’t isolated to sales-and-use tax or even the tax function,” says Campbell. “It happens everywhere, and it’s a significant AI roadblock.”

The aforementioned survey shows that companies with multiple ERPs can realize 40% efficiency gains if they build a tax data warehouse as their primary data source to reduce collection, cleansing and manipulation time.

Firms that have begun harnessing their data in this way are already benefiting from non-AI tech such as data analytics dashboarding, simple rules-based robotic process automation, and real-time, data-driven insights and alerts.

The best tax AI is “explainable”

Another barrier to AI adoption is trust. Before tax practitioners utilize AI in their processes, they want to understand how the AI is arriving at predictions. The end users often don’t fully understand the math and science of the algorithms. EY approach is to pioneer so-called explainable AI, which features analytics showing how the algorithm reaches an outcome.

When classifying transactions and assigning them to tax buckets, the AI analyzes the transaction description and vendor details. In explainable AI, each word is ranked by the algorithm with the highest-ranking words used in the classification process. Word-weight graphs are shared with users so they can see how any given result was reached and take comfort that the AI is using the same thought processes they might use.

Ivan Roussev, EY Senior Manager, Tax Technology and Transformation, Ernst & Young, LLP envisions that functionality allowing tax professionals to examine how an AI algorithm reached a certain answer will soon become the industry norm.  “Today, we expect our tax professionals to stand by their decisions and to be able to show the work that supports those decisions. This helps build trust and allows teams to course-correct timely when a mistake has been made. AI should be held to the same standard.  An AI solution should not be a black box that provides an answer without any reasoning behind it.  Similar to human judgment, the decision-making of an algorithm should be subject to review, questioning and, when necessary, correction, so that it can be improved. But before we can improve AI, we need to pinpoint the area of improvement, and explainable AI allows us to do just that.”

Roussev continues by saying that he wouldn’t be surprised if AI explainability became a legal requirement.  “We saw GDPR rolled out in Europe recently, regulating how personal data is used, and similar boundaries may soon become the norm in the world of AI,” he explains, referring to the European Union’s General Data Protection Regulation. “Consequential decisions require substantial reasoning, and AI is no exemption to that standard.”

While inconsistent data and human resistance may still pose challenges to AI adoption, old barriers such as cost and lack of computing power have been swept aside by developments such as cloud computing and cost effective, more powerful on-premise servers.

2

Chapter 2

The benefits of AI in tax

Companies beginning their AI journeys should focus on document intelligence and classification.

AI has become so accessible that the benefit of early adoption has effectively disappeared. Instead, tax leaders face the risk of falling behind their competitors if they don’t invest time and resources in cleansing their data and starting their own AI programs.

The survey shows that while decades-old tech, such as business intelligence tools, have been adopted by 70% of businesses, newer end-user data-cleansing tools have already been adopted by 51%. In another EY report, more than 80% of respondents said their tax teams will pivot from a technical tax focus to an emphasis on data, process and tech skills over the next three years.

Many of the algorithms that have an immediate impact on tax have been around for a long time and do not offer a competitive advantage, says Campbell. Tax groups that have been slow to embrace AI should use these tools to tackle “low-hanging fruit”—namely, document intelligence and classification.

Every time-poor tax practitioner knows the frustration associated with laboriously picking through invoices, looking for key tax data buried within documents. Document intelligence can now complete the process automatically, accurately, at speed, and at scale, thanks to powerful machine learning algorithms and pattern recognition. Having automatically retrieved key facts, AI can then fast-track classification.

“Tax teams spend too much time simply trying to classify things and put them into tax buckets,” says Campbell. “And those buckets determine tax treatments such as VAT, sales-and-use, whether something is taxable or exempt, and much more – the list is exhaustive. This is a serious drain on time and resources, but it’s straightforward for AI.”

Human tax teams and AI working together

Campbell cites a multi-part example in which AI triggered a dramatic increase in performance. The first part was a monthly compliance exercise related to sales-and-use tax in which a tax team was asked to categorize 15,000 transactions.

The bulk of the heavy lifting was carried out by an offshore shared services center. This took 30 hours before a three-hour final review back in the US. Overall, this 33-hour manual classification process was 95-97% accurate.

“At 97%, the AI model was just as accurate,” says Campbell, “but it completed the heavy lifting in five seconds rather than 30 hours.” Manual review time was also shortened. “AI classifications meant that human reviewers could target high-dollar amounts that had a lower confidence rating, making the review more focused, more effective and faster.”

The AI was also much more consistent than a human reviewer. If it found one incorrectly classified transaction, it was likely all other similar transactions would be wrong, so it was much easier to spot errors. This helped cut the human review period by half.

How COVID-19 strengthens the case for tax AI

As governments rush through tax stimuli to kick-start COVID-19-impacted economies, changes are being enacted in weeks that would previously have kept legislators busy for years. C-suites and visionary tax leaders are rapidly recognizing that this pace of change calls for a radical rethinking of tax tools.

Nowhere is this more pertinent than in the realm of tax forecasting, much of which has traditionally been achieved manually based on gut feeling rather than data. In the wake of a crisis such as COVID-19, however, it’s very difficult for companies to adjust their models because there’s so much manual work involved.

But with clean, structured data and an end-to-end data strategy for forecasting, businesses can input new data into their AI model and produce a fresh forecast in minutes. As a crisis or other unusual situation develops, companies can quickly and continually update forecasts with new data.

Prior to the pandemic, an EY report showed 84% of firms expected an increase in workload due to new digital tax filing requirements. Post-pandemic, it’s reasonable to expect the situation will be even more onerous.

AI’s proactive tax strategy recommendations

The tax benefits of AI go far beyond document intelligence and classification. Really exciting things start to happen when teams use AI to assist with their tax planning. This involves a range of data inputs—including regulatory law, company performance data and corporate strategy—fed into an AI model so that it can proactively make tax recommendations.

For example, the AI model might highlight a regulatory change in Spain that affects telecom companies and then offer suggestions on how to best manage the tax implications. Alternatively, a company might want to alter its tax strategy to free up more cash. In this case, the AI could offer a series of recommendations to achieve that goal.

“The promise of AI is that it will take care of mundane tasks,” says Campbell. “Businesses can then focus on developing business strategies and understanding the tax implications. To be successful, businesses will need to embed AI into everything they do. I think that’s the opportunity.” 

3

Chapter 3

How to get started with AI in tax

The first steps are streamlining data processes and making sure data is AI-ready.

To build a solid data foundation and embark on an AI journey, businesses should take these four steps:

1. Streamline data processes

The first step for businesses must be to streamline data processes. “This is critical if businesses want to avoid a data disconnect,” Campbell says. For example, it’s unusual for tax teams to know where their data comes from. They usually issue a request to IT, but IT often doesn’t know why tax needs this information or how it will be used. These kinds of disconnected processes can lead to errors, confusion and incorrect data.

2. Make certain that data is AI-ready

Companies must secure a strong, standardized data pipeline that delivers clean, AI-ready data. This can be a challenge for complex organizations that struggle with convoluted IT systems forged in the aftermath of M&A activity.

In the survey, 93% of respondents reported dealing with multiple ERP systems. Those with six or more ERP systems spend six times more on data collection, cleansing and manipulation. Conversely, a strong data pipeline is often a big advantage enjoyed by fintech disruptors who don’t have to contend with legacy systems.

3. Enlighten key stakeholders

Education is another key stepping stone to AI adoption. Campbell says this doesn’t mean tax practitioners need to learn how to code, but it’s advantageous if they understand what’s possible with AI and the kinds of problems it can solve.

4. Don’t overthink, just get started

“There are so many quick wins, and the barriers to entry are now so low that it’s important not to overthink all this,” says Campbell. “The world is moving fast, so a five-year plan will be outdated very quickly. Begin using some simple AI tools to crack common problems and just get started.”

Summary

AI is a key part of building an intelligent, innovative and strategic tax function. With AI in your tax toolkit, you can streamline mundane tasks and then use the technology to make tax recommendations.

About this article

Related articles