Menu
Tax Notes logo

Hey Siri, What About Income Taxes on AI?

Posted on June 8, 2020
Michael Carey
Michael Carey
Jamie Szal
Jamie Szal
Martin I. Eisenstein
Martin I. Eisenstein

Martin I. Eisenstein is managing partner of Brann & Isaacson of Lewiston, Maine; Jamie Szal and Michael Carey are lawyers with the firm. All are members of the Brann & Isaacson group that represents the firm’s business clients — including providers of artificial intelligence and information technology services and telecommunications services, as well as a quarter of the 100 largest internet retailers — in state and local tax and unclaimed property matters.

In this installment of Eyes on E-Commerce, the authors consider two thorny state income tax issues for providers of AI: whether a provider is subject to a state’s income tax absent a physical presence in a state, in light of P.L. 86-272; and the apportionment of the revenue from AI services to states where the provider has nexus.

Copyright 2020 Martin I. Eisenstein, Jamie Szal, and Michael Carey.

All rights reserved.

We discussed in our article last month1 the challenges in determining the taxability of artificial intelligence services under state sales tax laws. Providers of AI are confronted with similarly complex questions when considering whether their companies are subject to a state’s income tax, and if so, how to apportion their income to the various states. In this Part II in our series on AI, we consider the application of state income tax laws to AI services.

In Part I of our series, we posed a set of situations to illustrate some business uses of AI that we often see. We recap those situations before going further.

In Situation 1,2 the AI provider develops a proprietary algorithm to predict the likelihood of a person contracting the COVID-19 virus. The AI provider allows users to access the software remotely through an application programming interface. A user answers a series of demographic and behavioral questions, which the program compares with data collected from other users, as well as to Centers for Disease Control and Prevention reports of COVID-19 infection and contact tracing data. The program then anonymizes the data that the user inputs. Users, who are first responders and healthcare providers working at healthcare facilities, are not able to download the software and are not able to alter or control the algorithm. After completing the questionnaire, the user receives an email with a simple prediction report.

In Situation 2, the AI provider conducts end-to-end data analysis, which includes data gathering, classification, processing, and presentation. The AI provider gathers the required input data from private and public sources. It applies its proprietary algorithm, statistical concepts, and other methods of analysis to large, complex data sets to solve difficult, non-routine problems. The AI provider produces an electronic report reflecting the insights, patterns, and trends reflected in the data sets; performs quantitative analysis to create clear, actionable narratives related to user acquisition, conversion, and retention patterns; and translates the analysis into business and marketing observations and recommendations. In effect, an AI provider can, in reliance on its model of customer behavior, report on factors that trigger and otherwise influence customer purchases. Its clients, who are retailers, rely on the insights and advice to set pricing and marketing strategies, and for product development. The agreement between the retailer and the AI provider stipulates that the retailer owns the report.

In Situation 3, the AI provider partners with its customers to identify the customer’s key metrics to design interactive, executive-facing dashboards to track the progress of the client’s highest priority initiatives in real time. The key objective for the AI provider is to generate and produce a visual storyboard to move from data to insights to decision-making immediately. The AI provider regularly updates the storyboard by email or chat with recommendations about future actions for its customers. The customers, sales and marketing professionals, rely on the real-time observations and recommendations to guide daily communications strategies.

We discussed how, in each of these situations, a state or local jurisdiction may seek to require the AI provider to collect and remit its sales tax. The jurisdictions may not stop there; they may also seek to impose income tax liability on the AI provider. In many cases, neither the AI provider nor the database or application servers are located in the jurisdiction seeking to impose tax liability. The customers may access the provider’s servers/software from the state and/or the provider may email and otherwise interact with customers or its customer’s employees in the state. In other words, the customers are the only connection to the jurisdiction.

Is an AI Provider Subject to a State’s Income Tax?

As with sales taxes, the initial income tax concern is whether the provider of AI services is subject to a state’s taxing authority in the first place. At least 12 states and two major cities have enacted economic presence laws for their income taxes, so that mere sales into the state above a specific threshold trigger nexus and require payment of the state’s income tax.3 We anticipate that other states will soon join this group.4

A state’s imposition of income tax liability based on sales into the state, without the provider engaging in any activities in the state, likely no longer violates the nexus prong of the dormant commerce clause standard.5 The real question, which we discuss in the next section, is whether an AI provider is entitled to the protection from state income taxes afforded by the federal statute, Public Law 86-272.6

All commentators agree that P.L. 86-272 protects a seller of tangible personal property from a state’s income tax if the seller does not engage in any activities in the state that exceed mere solicitation.7 Thus, the characterization of AI is an important threshold question. Although the AI provider may promote its product as a service, is it actually selling tangible personal property? Of our three hypothetical situations, Situation 1 describes a possible AI product that, as software as a service (SaaS), may be deemed the sale of tangible personal property. SaaS is taxable in many states because the underlying software that customers access is pre-written, which is deemed taxable as tangible personal property.8 The product at issue is more likely to be deemed a service, and not tangible personal property, if the primary object of the securing the AI product is not the use of SaaS, such as in situations 2 and 3.

We address in the next section whether the AI providers described in situations 2 and 3 are entitled to P.L. 86-272 protection because they sell services rather than tangible personal property. Having set the groundwork, we then address the scope of the P.L. 86-272 protection afforded to companies that operate on the internet but do not engage in any activities in the state.

P.L. 86-272 Protects AI Service Providers Just as It Does Sellers of Tangible Personal Property

The Multistate Tax Commission and most state and local tax lawyers maintain that P.L. 86-272 immunity is limited to the sellers of tangible personal property.9 Commentators routinely assert, oftentimes without citation, that P.L. 86-272 provides protection only for sellers of tangible personal property.10

We disagree with the majority opinion. Congress used clear, unequivocal language to describe the protection from state income tax: “No state, or political subdivision thereof, shall have the power to impose . . . a net income tax on the income derived within such State by any person from interstate commerce. . .”11 The protection is not limited to income derived within a state from the sale of tangible personal property; it applies broadly to any income derived from interstate commerce.12

The structure of P.L. 86-272 supports our conclusion that the law’s purpose was not limited to one category of seller. Congress excluded some companies from the protection.13 Congress did not, however, identify the types of businesses eligible for protection under P.L. 86-272, such as sellers of tangible personal property, or exclude carte blanche some businesses from immunity under P.L. 86-272, such as service providers. The absence of a specific exclusion for companies that sell services from the broad protection of the statute is another indication that P.L. 86-272 protects those companies just as it does sellers of tangible personal property.14

The legislative history of P.L. 86-272 also supports the conclusion that the law was not limited to a category of sellers but designed to restrict a state’s power to impose income taxes based on “a minimum activities approach to the problem of state taxation of income from interstate commerce.”15 As we wrote in a previous article,16 the U.S. House managers indicated that the proposed law’s purpose was “to specifically exempt from state taxation income derived from interstate commerce where the only business activity within the state by the out-of-state company was solicitation.”17 Congress implemented this purpose through two separate statutory provisions.18

The U.S. Supreme Court underscored that P.L. 86-272 was designed to immunize persons from income tax liability on income derived in interstate commerce so long as the business does not exceed the minimal levels of in-state activities specified in the law — the in-state solicitation of orders for tangible personal property by the company or representatives.19 In Heublein, the Court stated that the purpose of the statute was: “to define clearly a lower limit for the exercise of that [a state’s] power. Clarity that would remove uncertainty was Congress’ primary goal.”20

Limiting P.L. 86-272 to sellers of tangible personal property rewrites P.L. 86-272. While the law sets the solicitation of the sale of tangible personal property as the lower limit of “business activities within such State by or on behalf of” the interstate trader, this is not the same thing as limiting the protection to sellers of tangible personal property. To limit the protection of P.L. 86-272 to sellers of tangible personal property is tantamount to saying that a state is allowed to impose a tax on the net income of any company — sellers of tangible property and services — because the language of P.L. 86-272 does not state that activities conducted solely outside the state preclude taxation. But that argument proves too much. Under that logic, soliciting sales of tangible personal property from outside a state would cause a company to void its P.L. 86-272 protection even though a foreign company conducting the very same activity within the state remains protected. As the Court stated in Heublein, by establishing the limit of in-state activities consisting of solicitation of the sale of tangible personal property, “Congress did, of course, implicitly determine that the State’s interest in taxing business activities below that limit was weaker than the national interest in promoting an open economy.”21

A Provider of AI Should Be Protected by P.L. 86-272 Unless Its Employees or Representatives Enter Into the State to Provide Advice to Its Customers

We wrote previously about the efforts of an MTC Uniformity Committee working group to revise the “Statement of Information Concerning Practices of Multistate Tax Commission and Signatory States Under Public Law 86-272.”22 The latest draft (approved by the Uniformity Committee April 22, 2020) continues to assert that a business interacting with a customer via its website or app loses the protection of the federal statute if the activity is not mere solicitation, because the out-of-state company “engages in a business activity within the customer’s state.”23

Under the MTC’s theory, the providers in situations 1, 2, and 3 would be subject to the income tax of each state in which (1) a person who accesses the provider’s site is located or (2) a person to whom the provider emails or texts is located, even though neither the provider nor its representatives step foot in the state. By definition, the provider’s website in each of those situations is interacting with the persons in the other states.

We think that the MTC theory is wrong.

The appropriate starting point for analysis of whether a company engages in activities protected by P.L. 86-272 is the language of the statute itself. In the words of Justice Antonin Scalia, P.L. 86-272 “prohibits a State from taxing the income of a corporation whose only business activities within the state consist of ‘solicitation of orders’ for tangible goods, provided that the orders are sent outside the State for approval and the goods are delivered from out of state.”24 There is no ambiguity in the statute. On its face, P.L. 86-272 prohibits a state from imposing income taxes on a company whose only business activity within the state is the solicitation of the sale of tangible personal property. As Justice Scalia noted in Wrigley, section 381 “unquestionably does limit the power of States to tax companies whose only in-state activity is ‘the solicitation of orders.’”25 A company that engages in no in-state activity, but only in activities solely outside the state, should be protected by P.L. 86-272 in the same way as a company that has sales representatives soliciting the sale of tangible personal property in the state.

Nor can the MTC validly argue that Wayfair provides a basis for its interpretation of P.L. 86-272. Under the commerce clause, Congress alone has the authority “to regulate commerce . . . among the several states.”26 That authority has both an affirmative and a dormant aspect. In the affirmative aspect, Congress exercises its authority to regulate interstate commerce, either to explicitly allow state action or to prohibit it.27 In the dormant commerce clause aspect, where Congress remains silent, state regulation and taxation are allowed so long as the state law does not discriminate against or burden interstate commerce.28 Wayfair is a case analyzing and specifying the nexus standard under the dormant commerce clause. In contrast, P.L. 86-272 represents the affirmative exercise of Congress’s commerce clause powers.29 The affirmative and dormant aspects of the clause are mutually exclusive.

When Congress affirmatively exercises its commerce clause powers, the standard and analysis developed under the dormant commerce clause is irrelevant.30 The relevant inquiry, therefore, is one of statutory construction. P.L. 86-272 “makes abundantly clear [] that the Federal Government has not remained silent.”31 A state’s taxation of income derived in interstate commerce, regardless of its source, based on activities that fall below that minimum level frustrates the objectives of P.L. 86-272.32

Rather than use Wayfair as a starting point to determine what “engaged in business activities within the state” means, another Supreme Court decision, announced on the very same day as Wayfair, should be the basis for interpretation of P.L. 86-272. In Wisconsin Central,33 the Court reiterated the principle that the critical factor in statutory interpretation is the intent of the Congress at the time the statute was enacted. In that case, the Court interpreted the Railroad Retirement Tax Act, a 1937 federal statute that, in return for federal pensions, provided for the taxation of railroads based on the railroad employees’ compensation, which in turn was defined as any form of “money remuneration.” The Court determined that the term “money remuneration” did not include employee stock options, based on the meaning of the term “money remuneration” when the Railroad Retirement Tax Act was adopted in 1937. As Justice Neil M. Gorsuch stated: “[i]t is not our function ‘to rewrite a constitutionally valid statutory text’” based upon a recent IRS interpretation expanding the meaning of “money remuneration” to include stock options.34 This is so even if the more contemporary alternative interpretation makes “good practical sense.”35

When it adopted P.L. 86-272 in 1959, Congress distinguished between activities actually conducted by or on behalf of a business outside the state from those conducted within the state. Indeed, the statute provides protection even for activities occurring in the state that consisted of “the solicitation of orders by such person, or his representative, in such State for sales of tangible personal property, which orders are sent outside the State for approval or rejection, and, if approved, are filled by shipment or delivery from a point outside the State.”36 P.L. 86-272’s distinction between out-of-state and in-state activities persists, even as technology has expanded the reach of the activities conducted outside the state.

Congress envisioned the constitutional protection afforded to an interstate business would be lost only if the business or its representatives engaged in activities within the state that went beyond mere solicitation of orders within the state. Just as in Wisconsin Central, interpreting P.L. 86-272 to mean that activities conducted over the internet by a service provider from a location outside the state constitutes “business activities within the state” would rewrite this federal statute based on modern circumstances. Supporters of this expansive interpretation think that it is justified — even that the interpretation makes good practical sense — because, in their view, the statute creates “an arbitrary and destructive interstate tax shelter.”37 However, the proper way to address this alleged problem is for Congress to determine whether to amend or revise P.L. 86-272, not by reinterpreting the federal statute so narrowly that it would nullify the protections of the law.

How Is the Revenue From AI Services Sourced for Apportionment Purposes?

After determining whether an AI provider is subject to a state’s income tax laws, the next question is how to source the revenue from AI to properly apportion the provider’s income to the state. Just as in the sales tax area, where sourcing of the sale of services is not a straightforward question,38 sourcing AI revenue is complex and varies depending on the type of AI at issue and by state.

But first, a quick apportionment primer. The need for apportionment arises when a company does business in more than one state. Visually oriented readers can consider apportionment to be like slicing a pie, where the whole pie represents the total taxable income of that company. Each state, of course, wants a slice. Apportionment is the means to determine just how big a slice each state gets. This is where the apportionment formula comes in. As originally conceived, states took into account three factors — the value of the company’s property (rented, leased, and owned), the company’s payroll expense, and the company’s sales — to calculate a ratio comparing the value of those factors within the state to the aggregate value of those factors everywhere.39 Within the last decade, the standard formula changed in two ways:

(1) states either placed greater weight on the sales factor by giving the factor disproportionate weight (giving the factor double, triple, or quadruple weight), or by eliminating the property and payroll factors and consolidating the three factors into a single sales factor; and

(2) many states no longer source sales of services and other intangibles by cost of performance (that is, the value of the inputs needed to perform the service), but to the market, in effect attempting to distribute the value of services among the company’s customer base.40

Market-based sourcing is equally, if not more, difficult for an AI provider, as for other sellers of services. How does a business calculate and geographically measure its market when it provides something that is not physical?41 More than 25 states have adopted market-based sourcing rules, with different definitions and rules on what constitutes the market.42 The variations among the states fall along a spectrum. Depending on the state, the market for purposes of sourcing may be based on the state or states in which the benefit is received, the service is delivered, the service is received, or the customer is located. In the remainder of this section, we consider the application to an AI services provider of the market-sourcing laws of California and New York and the MTC Uniform Division of Income for Tax Purposes Act rules.43

California

On one end of the spectrum we have California. The state does not have a specific rule for sourcing digital services, but instead relies on rules written for services generally. These sourcing rules are based on where the purchaser receives the benefit of the service.44 The regulation spells out the rules for determining where the benefit is received45 based on the general principle that the benefit is received where the customer has directly or indirectly received value from the service.46 The regulation provides a cascade of four rules, where the provider proceeds to the next level for sourcing only if a higher level rule is not satisfied.47 The starting presumption is that income from services will be sourced to the state as reflected in either the service contract or the service provider’s books and records as the place where the benefit is received.48 If that location cannot be determined, or if the presumption is overcome, then the AI provider shall reasonably approximate the location where the benefit is received.49 The next levels, in the event neither tier 1 nor tier 2 is satisfied, are the state from which the customer placed the order,50 followed by the customer’s billing address.51

On the other hand, the rules governing the sale or license of intangibles provide that sourcing is based on where the intangible property is used.52 The determination of the location of use is also based on a cascade, adopting the same tiers 1 and 2 as for sourcing services, but eliminating the tier that sources to the location where the order was placed and proceeding directly to the billing address of the customer.53 When the seller is providing a license that is related to the marketing of goods by the purchaser, however, the regulation provides for a determination of use based on the location of the sale of the customer’s product.54

Although the California sourcing rules appear relatively straightforward, there remain at least three problems with this sourcing regime. First, there is no clear test to determine whether the seller is selling a service or intangible personal property. As we detailed in Part I of our series, this is a gating issue as AI services can fall under several categories. This is particularly problematic when the provider develops custom software to provide information to facilitate its customer’s marketing efforts. Our situations 2 and 3 are good examples of the difficulty that the service provider faces when determining how to characterize the product it sells. Should we treat the provider’s work as the sale of intangible personal property? After all, the AI provider in both situations crafts software tailored to its customer. If so, should the software be deemed to relate to the sale of the customer’s products, making the place of use of the software to be the place where the customer sells its products? Alternatively, we believe that the primary purpose of the relationship is to secure insights and recommendations generated by the AI provider, and therefore is arguably a service. As a service, the income from AI should be sourced to the location where the AI service provider’s customer received the benefit. And where, exactly, is that benefit received? Will the contract with the customer spell that out? Will the Franchise Tax Board argue that the seller should reasonably approximate sales of its customer’s goods based on population statistics, irrespective of the parties’ relationship?

The second and third problems relate to the ambiguity of the standards. There is no specification as to what the FTB or taxpayer needs to demonstrate to overcome the presumption for level 1. In turn, if the presumption is overcome, or if neither the contract nor the provider’s books and records specify the place where the benefit is received or the location of use, then there is no clear test to determine the second level — under the “reasonably approximated” standard.

New York

New York lies on the other end of the spectrum. Like California, New York sources services based on where the benefit of the service is received.55 Here the similarities end. Unlike California, New York has adopted specific rules for digital products which are separate and apart from its sourcing rules for services or other intangible property generally.56 Digital products, defined broadly to include SaaS and some other digital services, are sourced based on a hierarchy which sources receipts first to the customer’s primary use location and then to the location where the customer receives the digital product.57 For services other than digital products, receipts are sourced based on a cascade that starts with (1) the state or states of receipt of benefit, before moving to (2) the delivery destination, and, as necessary, (3) the apportionment method used during the preceding tax year.58

The New York sourcing law provides significant challenges for an AI provider. As a threshold question, the AI provider must determine whether it provides a digital product or a more general service.59 New York law does not specify a test to use in making that determination. Should the provider use the primary function test, as adopted for sales tax purposes?60 In Situation 1 described above, one could argue that the provider’s primary function is providing SaaS, an information service, or a data processing service. If characterized as SaaS, then the sales would be considered digital products and the provider would be required to determine the location of primary use. But what if the user’s customers are located in several states? Is New York the primary use location if it has the largest number of users? Is the primary use location New York solely because the customer’s corporate headquarters is in Manhattan? If New York is the primary use location, then it would seem that all revenue from that customer for that service would be sourced to New York, even though another state such as California might claim that a portion of the revenue should be sourced there because users are in that state.

If, however, the service in Situation 1 is characterized as an information service or a data processing service, then the income from that service would be sourced according to the rules for general services. As a general service, only the percentage of the revenue attributed to New York users of the service would be sourced to New York. Situations 2 and 3 present similar challenges. Does the fact that the AI provider makes the software available electronically mean that the provider is selling a digital product? We think not, if the transaction is thoroughly analyzed, since the provision of software is only a small part of the work that the provider performs.

The concerns that arise from applying the New York rules do not end there. Critically, there is no definition in the statute (or any regulations) of where the benefit is received. If the primary purpose of the service is to provide analysis to guide decision-making for the business activities of different departments, such as marketing and product development, as is the case for Situation 2, should the service be sourced to where those departments are located? Does the analysis change depending on the department’s function? Would services be sourced differently if they are used by customer support rather than the internal auditing department? What should an AI service provider do if the departments of its customer are in offices that are in more than one state? If it is impossible to accurately determine where the benefit is received, are the locations for delivery of the reports an appropriate basis to source the revenue from the services discussed in Situation 2? And if so, what does the term “delivered” mean in the AI provider’s digital world where the reports are transmitted electronically?

Attempting to source the services provided in Situation 3 raise similar concerns. Providing and updating dashboards for marketing and sales professionals arguably provides a benefit to those professionals where they are located. But the benefit may be to the sales for the overall organization, so are sales by state of the customer an appropriate way to source the AI service described in Situation 3? If so, how can the AI service provider be expected to know the locations of its customer’s customers?

In short, the open-endedness of the New York statute raises several questions for an AI provider.

UDITPA Rules

The new 61 rules, adopted by the MTC Uniformity Committee in 2017, may prove a more workable middle-ground solution for AI providers than the California and New York market sourcing laws.62 At first blush, however, the UDITPA rules raise some questions for AI providers.

The UDITPA rules place sales (now called receipts) into four categories, two of which — sale of services and sale or licensing of intangible personal property — are implicated when it comes to sourcing AI.63 The new rules source services to where they are delivered.64 Likewise, the rules source intangible property to the state in which it is used.65 So, the initial question is whether AI is a service or the provision of intangible property. Complicating this question is the need to determine the meaning undefined terms: “delivery” in the case of services, and “used” in the case of intangible property. Does either term mean “where the customer realizes the benefit of the service,” which is the basis for sourcing services under the California and New York market-based rules and, in many states, for sourcing sales tax receipts of data processing and information services? Similarly problematic is the sourcing of receipts if the AI provider’s customer has locations in many states from which the customer’s employees may access the AI product. More complicated still is the sourcing of receipts if the AI provider’s customer accesses the AI services on mobile devices from a constant variety of locations.

We first address how services are sourced under the UDITPA rules, which resolve some (but not all) of these issues. If the AI provided is deemed a service, the UDITPA rules distinguish between professional services, such as consulting services and data processing services,66 and other services delivered by physical means67 or electronic transmission.68

If the services are categorized as professional services provided to a business customer, the rules generally source the services to a single state, as reasonably determined by the provider based on a cascade. The first tier is based on where the contract of sale is principally managed by the customer; followed by the customer’s place of order; and then the customer’s billing address.69

For services other than professional services, the rules distinguish between those delivered by physical means and those delivered electronically. The rules provide an example: custom software delivered on a physical medium to a particular customer and installed on the customer’s server is sourced only to the state where the custom software is delivered.70 But that is not a difficult scenario. What is more complex are the rules for the sourcing of other services delivered electronically, which include yet another cascade of sourcing levels, similar to the rules that apply to sourcing of consulting services, and which apply only if the provider cannot reasonably determine the location of the end users of the service.71

The benefit of the MTC rules is that they provide a relatively straightforward method to apply based on the provider’s reasonable efforts to determine the location where the service is delivered — though, only after the provider determines what is being provided.

The most problematic aspect of the rules for an AI provider is the lack of a clear test to determine whether its service is a professional or nonprofessional service.72 As discussed above, the service is sourced based on the sourcing cascade for professional services if the provider furnishes data processing or consulting services. But sourcing is based on the provision of nonprofessional services if the service is characterized as a digital good or other service. Further, the rules do not spell out whether information services are deemed a professional service, such as data processing, or another service, such as custom software development.

Another drawback of the rules is that when it comes to AI, the line between services and intangible property can be quite blurry; there is no test under the UDITPA rules to dictate whether AI services trigger the rules regarding the sale/license of intangible property rather than those for the sale of services.

With that in mind, we now address how the license or sale of intangible property is sourced under the rules. The rules provide that the license or sale of intangible property is based on where the intangible property is used,73 which is different from where the service is delivered, in the case of services. The intangible property rules are geared toward franchise agreements, canned software downloads, and trademark licensing, and were not drafted with AI in mind. They provide for the sourcing of the sale or license of pre-written software delivered on a physical medium based on the principles applicable to sourcing of the sale of tangible personal property,74 while the sourcing of the sale or license of digital goods, including software, is based on the rules applicable to the electronic sourcing of services other than professional services.75 But, unlike the New York law, the rules do not define digital goods. Are SaaS or other AI services deemed the sale or license of a digital good? If the AI software is not pre-written software delivered on a physical medium or deemed a digital product, the applicable sourcing rules for software are unclear. The provider is confronted with five potential rules for sourcing without bright-line tests to determine which rule applies.76

In short, the MTC rules go a long way toward providing relatively straightforward rules for sourcing of revenue for service providers. The problems with the rules, for an AI provider, are similar to those that it faces when determining whether the services it provides are subject to sales tax, as shown by each of the three situations outlined in this article.

In Situation 1, the AI service provider generates revenue from healthcare providers who pay to secure the diagnostic report prepared by the AI provider. The users are likely located in different states. If the service in Situation 1 is characterized as SaaS, then the sourcing remains unclear for the reasons described above. But if Situation 1 is deemed a data processing service to a business customer, it would be sourced based on the rules’ sourcing cascade for professional services.

Situation 2 blurs the lines between pre-written software and digital services. One can argue that AI products that consume large volumes of data to identify trends or other patterns but lack more sophistication, such as those in Situation 2, more closely resemble intangible property than services. Because the UDITPA intangible property rules are geared toward franchise agreements, canned software downloads, and trademark licensing, it is not clear whether the AI services in Situation 2 would be considered digital services. The services we describe in Situation 2 more closely resemble the category of services referred to as data processing or information services and would most likely be sourced under the rules for services delivered by electronic transmission to business customers.

Situation 3 adds further complexity. The AI provider generates revenue by selling its services to organizations primarily for their sales and marketing personnel. The dynamic, real-time dashboard and reporting services are designed for and predominantly used by the sales reps and marketers, a highly mobile group. Dynamic AI algorithms more closely resemble services than they do intangible property. Are the services deemed consulting services and, therefore, sourced as professional services? Or are the services deemed the development of custom software, in which case different sourcing rules apply? And how does one determine the location of a highly mobile user if the applicable rules are those for sourcing services delivered by electronic transmission to business customers, which look first to the location where the AI provider’s customer receives (that is, directly uses) the service?

Conclusion

AI providers face an array of thorny issues in the application of the income tax laws to their services. Different states have adopted different approaches to sourcing AI products, adding to the confusion. States have made the transition of their sourcing methods from cost of performance sourcing to market sourcing in an ad hoc and divergent manner.77 Providers of services that are subject to state income taxes face a confusing array of state laws and the real risk that the same receipts could be sourced to multiple jurisdictions and, as a result, be subject to double taxation.78 That risk is even greater for AI, in large part because of the difficulty of characterizing AI services and pigeonholing them into a sourcing category, as was the case for determining taxability under the sales tax laws. It is particularly problematic in states such as New York, where the standards used are ambiguous at best. This leaves AI providers and their tax advisers to think creatively about how to apply state income tax laws to situations that may have seemed like science fiction to legislators at the time that the laws were written. Speaking of which:

“Hey, Siri, what about income taxes on AI?”

FOOTNOTES

1 Martin I. Eisenstein, Michael Carey, and Jamie Szal, “Alexa, Is AI Taxable?” Tax Notes State, May 11, 2020, p. 719.

2 If you thought Situation 1 farfetched, Apple announced in March that it updated Siri with a questionnaire intended as a guide to those concerned that they may be experiencing symptoms of the COVID-19 virus. See Eli Blumenthal, “Worried About Coronavirus? Siri Can Help You Check Symptoms,” CNET Health and Wellness, Mar. 22, 2020.

3 Among the states that have adopted economic presence income tax laws are: Alabama, Ala. Code section 40-18-31.2; California, Cal. Rev & Tax Code section 23101(b) and (d)); Colorado, Colo. Code regs. section 39-22-301.1; Connecticut, Conn. Gen. Stat. section 12-216a; Hawaii, Haw. S.B. No. 495 (codified at Haw. Rev. Stat. section 235) (effective for tax years beginning on or after Dec. 31, 2019); Indiana, Ind. S.B. 563 (codified at Ind. Code section 6-3-2-2 (effective Jan. 1, 2019); Massachusetts, 830 CMR 63.39.1(3)(d) (effective for tax years ending on or after Oct. 1, 2019); Michigan, Mich. Comp. Laws section 206.621(1); New York, N.Y. Tax Law section 209.1; Tennessee, Tenn. Code section 67-4-702; and Virginia, Va. Code section 48.1-400. Both Philadelphia and San Francisco have also adopted economic nexus, effective Jan. 1, 2019. See Philadelphia BIRT Regulation section 103; and San Francisco Bus. & Tax Reg. Code art. 6, section 6.2-12, and art. 12-A-1, section 953. In addition to income taxes, all the major states that levy gross receipts taxes have adopted economic nexus standards for these taxes, including: Nevada, Nev. Rev. & Tax. Code section 363C.200(1); Ohio, Ohio Rev. Code Ann. section 5751.01(H)-(I); Oregon, Ore. Rev. Stat. ch. 317A, et seq. (effective Jan. 1, 2020); Texas, Tex. Tax Rule section 3.586; Tennessee, Tenn. Code section 67-4-701, et seq.; and Washington, Wash. Rev. Code section 82.01.010, et seq., and S.B. 5581 (new lowered threshold of $100,000 effective Jan. 1, 2020).

4 The Pennsylvania Department of Revenue sent shock waves last year when it announced in a tax bulletin, without the support of specific or enabling legislation, that it would enforce economic nexus for those companies generating $500,000 or more gross receipts in Pennsylvania effective Jan. 1, 2020. Pa. DOR Tax Bulletin 2019-04.

5 The Court’s holding in South Dakota v. Wayfair that economic nexus is sufficient to establish nexus for sales tax collection purposes under the dormant commerce clause should apply to the dormant commerce clause standard for state income taxes as well. See 138 S. Ct. 2080 (2018).

6 15 U.S.C. section 381.

7 See Multistate Tax Commission, “Statement of Information Concerning Practices of the Multistate Tax Commission and Signatory States under Public Law 86-272” (as revised July 27, 2001); see also Richard Pomp, “Wayfair: Its Implications and Missed Opportunities,” Tax Notes State, Dec. 23, 2019, p. 1035.

8 The following states treat SaaS as taxable because it is a sale or lease of tangible personal property: Indiana, Louisiana, Massachusetts, New York, Pennsylvania, Rhode Island, Tennessee, and West Virginia. See Part I of our series on AI, supra note 1.

9 MTC, “Statement of Information,” supra note 7, and draft fourth revision (approved by Uniformity Committee Apr. 21, 2020); Jamie Yesnowitz, Chuck Jones, and Sonia Shaikh, “The Catch-22 of Public Law 86-272,” Tax Notes State, May 18, 2020, p. 845.

10 For example, Darien Shanske and David Gamage recently stated without citation that “P.L. 86-272 protects taxpayers from state income taxation if certain criteria are met. Those criteria include. . . only selling tangible personal property.” Shanske and Gamage, “The Ordinary Diet of the Law: How to Interpret Public Law 86-272,” Tax Notes State, Apr. 13, 2020, p. 161.

11 15 U.S.C. section 381(a) (emphasis added).

12 15 U.S.C. sections 381(a) and 382(a). See Disney Enterprises Inc. v. Tax Appeals Tribunal, 888 N.E.2d 1029, 1036 (N.Y. App. 2008) (the statute “creates a tax exemption for a ‘person’ whose instate activities do not exceed solicitation”).

13 See 15 U.S.C. section 381(b), which states that the provisions of 15 U.S.C. section 381(a) do not apply in any state to corporations incorporated under the laws of the state or to individuals domiciled or residing in the state.

14 The presumption that “when a statute designates certain persons, things, or manners of operation, all omissions should be understood as exclusions” is recognized under the maxim expressio unius est exclusio alterius. Copeland v. Ryan, 852 F.3d 900, 906 (9th Cir. 2017). The specific enumeration, in 15 U.S.C. section 381(b), of the categories of taxpayers who are excluded from the protections of P.L. 86-272 must be understood as meaning that only corporations incorporated in the state and domestic residents and citizens of the state are excluded from P.L. 86-272’s protection, and all others are included.

15 See Statement of the Managers on the part of the House, Conference Report No. 86-1103 (Sept. 1, 1959).

16 See Eisenstein and Nathaniel A. Bessey, “Wayfair and P.L. 86-272 in a Services Economy,” State Tax Notes, Nov. 5, 2018, p. 501.

17 Statement of the Managers, supra note 15, and Congressional Record, Vol. 105, Part 13, at 16470.

18 See 15 U.S.C. section 381(a) (“No State, or political subdivision thereof, shall have power to impose . . . a net income tax on the income derived within such State by any person from interstate commerce.”); and 15 U.S.C. section 382(a) (“No State, or political subdivision thereof, shall have power to assess . . . any net income tax which was imposed by such State or political subdivision . . . on the income derived within such State by any person from interstate commerce, if the imposition of such tax for a taxable year ending after such date is prohibited by section 381 of this title.”).

19 Heublein Inc. v. South Carolina Tax Commission, 409 U.S. 275 (1972).

20 Id. at 280 (emphasis added).

21 Id.

22 David Bertoni, David Swetnam-Burland, and Szal, “Crossfire Hurricane: Perils in a Post-Wayfair World,” Tax Notes State, Mar. 16, 2020, p. 937.

23 MTC, Discussion Draft 7.2 of “Statement of Information Concerning Practices of Multistate Tax Commission and Signatory States Under Public Law 86-272” (Feb. 14, 2020).

24 Wisconsin Department of Revenue v. William Wrigley Jr. Co., 505 U.S. 214, 216 (1992) (emphasis added).

25 Id. at 224 (emphasis in the original).

26 U.S. Const. Art. I, section 8, cl. 3.

27 Wardair Canada Inc. v. Florida Department of Revenue, 477 U.S. 1, 9, 12 (1986). The supremacy clause of the Constitution governs once Congress has legislated in an area. U.S. Const. Art. VI, cl. 2.

28 See Oklahoma Tax Commission v. Jefferson Lines Inc., 514 U.S. 175, 179 (1995) (“We have consistently held this language to contain a further, negative command, known as the dormant Commerce Clause, prohibiting certain state taxation even when Congress has failed to legislate on the subject. Quill Corp. v. North Dakota, 504 U.S. 298, 309 (1992); Northwestern States Portland Cement Co. v. Minnesota, 358 U.S. 450, 458 (1959); H.P. Hood & Sons Inc. v. Du Mond, 336 U.S. 525, 534-535 (1949).”). (Citations from other reporters omitted.)

29 Cf. Shanske and Gamage, “The Ordinary Diet of the Law,” supra note 10. Shanske and Gamage err in assuming that Congress created some narrow add-on to the commerce clause rule, when, in fact, Congress displaced the dormant commerce clause standard with a new legislative standard, just as it is empowered to do.

30 Wardair, 477 U.S. at 9. “For the dormant Commerce Clause, in both its interstate and foreign incarnations, only operates where the Federal Government has not spoken.” Id. at 12.

31 Id. This admonition is even clearer for P.L. 86-272 than in Wardair, in which the congressional action was based on a pattern of international treaties.

32 See, e.g., Wrigley, 505 U.S. at 223 (“[s]ection 381(a)(1) confers immunity from state income taxes on any company whose ‘only business activities’ in that State consist of ‘solicitation of orders’ for interstate sales”); Edgar v. Mite Corp., 457 U.S. 624, 635 (1982) (law frustrated the provisions of the federal Williams Act because it created consequences which Congress in the Williams Act sought to avoid); and Jones v. Rath Packing Co., 430 U.S. 519, 540-541 (1977).

33 Wisconsin Central, 138 S. Ct. 2067.

34 Id. at 2074.

35 Id.

36 15 U.S.C. section 381(a)(1).

38 See, e.g., Eisenstein and Carey, “Where’s Waldo: Sourcing IT and Cloud Services,” State Tax Notes, Aug. 8, 2016, p. 429.

39 MTC Model Compact Art. IV, para. 9 (originally adopted July 11, 1986). This is commonly referred to as UDITPA.

40 For purposes of this article, we refer to the MTC’s UDITPA Article IV, section 17 (as adopted and revised effective Feb. 24, 2017). Before the adoption of these amendments, services were sourced based on the cost of performance, as is still the case in a minority of states.

41 Setting aside software for the moment, geographic sourcing of real property and tangible personal property is not as difficult as sourcing services or intangibles. Real and tangible personal property are physical, after all, and can exist in only one place at one time.

42 Although 25 states have adopted some form of market-based sourcing, the rules are not uniform among these states. See, e.g., Sheelagh Beaulieu, Maria Eberle, and Liz Jankowski, “Living With Market-Based Sourcing — Calculating Today’s Sales Factor,” Florida Bar Tax Section 2020 National Multistate Tax Symposium (Jan. 29-31, 2020); and Shirley Sicilian, John Biek, and Hayes Holderness, “What’s Left of State Tax Planning?” American Bar Association State and Local Tax Committee meeting (May 10, 2019).

43 See MTC, Model General Allocation and Apportionment Regulations (Feb. 24, 2017). The MTC market sourcing regulations are very similar to the applicable Massachusetts regulation. 830 CMR 63.38.1(9)(d)7.b.

44 Cal. Rev. & Tax Code section 25136.

45 Cal. Code Regs. tit. 18, section 25136-2(c).

46 Id. at (b)(1).

47 Id. at (c)(2).

48 Id. at (c)(2)(A).

49 Id. at (c)(2)(B). The “presumption may be overcome by the taxpayer or the Franchise Tax Board by showing, based on a preponderance of the evidence, that the location (or locations) indicated by the contract or the taxpayer’s books and records was not the actual location where the benefit of the service was received.” Id. at (c)(2)(A).

50 Id. at (c)(2)(C).

51 Id. at (c)(2)(D).

52 Id. at (d).

53 Id. at (d)(2)(B).

54 Id.

55 N.Y. Tax Law sec. 210-A(10)(b)(1).

56 N.Y. Tax Law sec. 210-A.

57 N.Y. Tax Law sec. 210-A(4)(c).

58 Id. (10)(b).

59 To complicate the analysis, while the apportionment statute defines digital products broadly to include any digital property or service, the same statute also indicates that digital products do not include analytical or consulting services. N.Y. Tax Law sec. 210-A(4)(a).

60 See, e.g., New York State Department of Taxation and Finance, Advisory Opinion TSB-M-10(7)S (July 19, 2010) (“The Tax Department will determine a service’s primary function based on an examination of the nature of the service being sold and what is being paid for by the purchaser.”).

61 UDITPA Art. IV, section 17(a) (2017). The rules as proposed and later adopted can be found online.

62 The following states have, in whole or in part, adopted the amended model regulations: Montana, Mont. Admin. reg. 42.26.245; Tennessee, Tenn. Comp. R. & regs. 1320-06-01.-42 (adopted only in part); Oregon, Rule 150-314-0435; and Rhode Island, R.I. Admin. Code section 60-1-194:8.

63 UDITPA Art. IV, section 17(a) (2017).

64 Id. at (a)(3).

65 Id. at (a)(4).

66 UDITPA reg. IV.17.(d)(4).

67 UDITPA reg. IV.17.(d)(2).

68 UDITPA reg. IV.17.(d)(3).

69 UDITPA reg. IV.17.(d)(4)(C)(1)(b).

70 UDITPA reg. IV.17.(d)(3)(B)(1)(c)(v).

71 UDITPA reg. IV.17.(d)(3)(B)(2)(b).

72 For example, the rules could specify the use of a true object or primary purpose test, as in the case of the sales tax. For a more general discussion of the ways in which the amended regulation could be improved, see Jeffrey Friedman, Nicholas Kump, and Robert Merten, “Implications of the MTC’s Market-Based Sourcing Model Regulations,” State Tax Notes, Mar. 26, 2018, p. 1201.

73 UDITPA reg. IV.17. (e) and (f).

74 Id. at (g)(1).

75 Id. at (g)(2). As part of its 2017 amendments, the MTC borrowed heavily from the then-new Massachusetts market-sourcing regulation and incorporated a special rule directed at the sale or license of digital goods or services. See, e.g., UDITPA reg. IV.17.(g)(2); and 830 CMR 63.38.1(9)(d)7.b.

76 Id. at (g)(1) provides as follows:

In all other cases, the receipts from a license or sale of software are to be assigned to [state] as determined otherwise under Reg. IV.17. (e.g., depending on the facts, as the development and sale of custom software, see Reg. IV.17.(d).(3), as a license of a marketing intangible, see Reg. IV.17.(e).(2), as a license of a production intangible, see Reg. IV.17.(e).(3), as a license of intangible property where the substance of the transaction resembles a sale of goods or services, see Reg. IV.17.(e).(5), or as a sale of intangible property, see Reg. IV.17.(f)).

77 Letter from George Howell, ABA Section of Taxation chair, to Brian Hamer of the MTC, at 1 (Mar. 1, 2016).

78 As a result, at the behest of the ABA tax section, the MTC adopted a provision calling for mediation in the event that a taxpayer is subject to different sourcing methods regarding intangibles or services. UDITPA reg. IV.17.(h). The ABA urged the MTC to do so in light of the growing confusion caused by the states’ ad hoc popcorn approach.

END FOOTNOTES

Copy RID