Benjamin Alarie is the Osler Chair in Business Law at the University of Toronto and the CEO of Blue J Legal Inc., Susan Massey is a director at Blue J Legal, and Christopher Yan is a senior legal research associate at Blue J Legal.
In this article, the authors use the Blue J machine-learning algorithm to examine Podlucky, an innocent spouse relief case recently decided by the Tax Court, demonstrating how useful artificial intelligence tools can be in calibrating the strengths and weaknesses of clients’ circumstances and predicting likely outcomes in similar cases.
Copyright 2023 Benjamin Alarie, Susan Massey, and Christopher Yan.
All rights reserved.
Claims for innocent spouse relief often arise and turn on time and resource-intensive facts and circumstances analyses. Perhaps surprisingly, these kinds of cases are well suited to machine-learning analysis. Practitioners now have access to artificial intelligence tools to assist them in rapidly analyzing likely outcomes for potential claims for innocent spouse relief. These tools can facilitate the careful weighing and calibration of the strengths and weaknesses of the unique circumstances of clients, leading to fully optimized advice and confident settlements. In short, you can be sure that you have truly explored all aspects of your client’s position in a potential settlement when you understand the most likely outcome in court.
To demonstrate, in this article, we examine the facts and circumstances of Podlucky,1 a recent Tax Court case concerning the denial of innocent spouse relief. The pro se taxpayers’ appeal to the Ninth Circuit was dismissed for lack of jurisdiction. We use the facts of this case to illustrate how Blue J’s machine-learning technology could have been used to assess the likelihood of success for an innocent spouse claim. More specifically, we identify several ways in which machine-learning analysis could have led to sharper arguments and better positioning for the taxpayer, based on a deeper exploration of key facts and circumstances. Using machine learning, we determine that the IRS had an overwhelmingly high chance of success had the appeal of the Tax Court’s decision to deny innocent spouse relief not been dismissed, but that there could have been an avenue for the taxpayer to secure at least partial relief with some modest variations of the facts.
Podlucky involves significant underreporting of income and the correlative underpayment of tax on a joint federal income tax return. The IRS claimed that Gregory and Karla Podlucky engaged in fraud and received constructive distributions from a corporate bank account. Karla sought innocent spouse and equitable relief under section 6015(b) and (f), respectively. Blue J’s machine-learning technology predicted with greater than 95 percent confidence that the facts adopted by the Tax Court would not warrant innocent spouse relief. However, if we make favorable inferences about certain underlying facts while applying the apportionment of relief provision set forth in section 6015(b)(2), machine learning indicates that Karla could have increased her chances of receiving partial relief to 59 percent.
We proceed with some background on the facts of the case, the applicable law, and the decision of the Tax Court, before turning to the conclusions we draw from the machine-learning analysis.
A. The Podluckys
The Podluckys are a married couple who filed joint federal income tax returns for 2003-2006. Greg is a CPA and ran a bottling business, Le-Nature’s Inc. (LNI). Although records indicate that the business enjoyed increasing profits during its early years, the income was apparently insufficient to satisfy the couple’s material desires.
LNI had two minority shareholders, both private equity funds, which had invested (in aggregate) over $15 million in LNI. In 2005 these minority shareholders sought to sell their stock in LNI, anticipating large gains on their investments because of the sharp increase in reported revenue. Although several purchasers expressed an interest, no sale was completed. Apparently, Greg refused to provide the potential buyers with access to the financial books. This prompted the minority shareholders to sue LNI in Delaware Chancery Court for obstructing the sale. During this court proceeding, one of LNI’s lenders raised a concern that some of the documents provided by LNI to secure loans might have been fabricated. This evidence led the court to remove Greg as CEO. The court appointed a financial restructuring specialist as custodian of LNI.
Upon investigation, it was discovered that Greg, with the assistance of an employee, had maintained two sets of books. The official set of financial accounting records documented many fictitious transactions that inflated the company’s actual earnings by nearly tenfold. These fraudulent financial records had induced LNI’s creditors to provide hundreds of millions of dollars in financing to the company. Greg and his accountant kept a second, secret, set of books that tracked the actual transactions of LNI.
Further investigation revealed that Greg had created two shell companies that he used to funnel millions of dollars from LNI to himself and Karla during the relevant tax years, fraudulently labeling many of the transfers as payments to suppliers. Karla and Greg each had access to the bank accounts of these shell companies, from which they wrote checks for various expenditures, including luxury jewelry valued at over $22 million, $10,772 in tuition and fees to their son’s college, and over $1 million worth of toy trains. A substantial portion of the jewelry had been custom made for Karla.2 Also, Greg caused the corporation to purchase land adjacent to his personal home and to build a mansion on it valued at $11 million. He later transferred the property to himself for $1, using forged approval from LNI’s board of directors. During each of these years, the Podluckys reported income between $350,000 and $600,000 on their joint return.
As a result of these fraudulent transactions and following the criminal convictions of the Podluckys for money laundering, the IRS examined the Podluckys’ tax returns and determined that the couple had underreported their income by nearly $35 million during the years at issue. This resulted in a $4.8 million underpayment of federal income tax for which the IRS assessed $3.6 million in fraud-related penalties. The Tax Court sustained the IRS’s deficiency determination.3 As part of the proceeding, Karla sought and was denied relief as an innocent spouse under section 6015.
B. Relief Under Section 6015
Section 6013(a) permits married taxpayers to file a joint federal income tax return. In filing jointly, under section 6013(d)(3), spouses submit to joint and several liability for all taxes due on that return. However, the code provides relief from that joint and several tax liability under the innocent spouse provisions in section 6015(b), or the provisions for equitable relief in section 6015(f).
To qualify for innocent spouse relief under section 6015(b)(1), a taxpayer must satisfy five requirements outlined in the statute. Failure to satisfy any of these requirements will render the requesting spouse ineligible for relief. These requirements are that:
the spouses filed a joint return;
the return contains an understatement of tax that is attributable to erroneous items of the non-requesting spouse;
the requesting spouse establishes that she neither knew, nor had reason to know, of the understatement;
it would be inequitable to hold the requesting spouse liable for the deficiency arising from the understatement; and
the requesting spouse timely filed a request for relief.
The key considerations in determining eligibility for innocent spouse relief are generally (2), (3), and (4). Thus, to prevail on a claim for innocent spouse relief in this case, Karla would need to demonstrate that the income was not attributable to her, she had no reason to know of the understatement, and it would be inequitable to hold her liable for the deficiency arising from the understatement. The first of these requirements considers whether the understatement of tax is attributable to erroneous items of the non-requesting spouse. When the items resulting in the underpayment are even partially attributable to the requesting spouse, she will not be eligible for innocent spouse relief.4
The next consideration is whether the requesting spouse knew or had reason to know of the underpayment.5 She will be considered to have reason to know of the underpayment if a reasonable person in similar circumstances would have known.6 This is a facts and circumstances determination, and the regulations set forth elements that will be taken into consideration when weighing whether a requesting spouse should be treated as having constructive knowledge, including the extent to which she participated in the activity that resulted in the understatement and whether she failed to inquire about the items on the return.7
Finally, the requesting spouse must prove that the imposition of joint and several tax liability for the underpayment would be inequitable.8 This is also based on a facts and circumstances analysis.9 An important factor in this analysis is whether the requesting spouse received a significant benefit from the understatement reported on the joint return.
C. Availability of Partial Relief
Section 6015(b)(2) provides for partial relief for a requesting spouse who would qualify for relief but for her knowledge or reason to know of the understatement on the joint return. Under this provision, the requesting spouse will be relieved of liability to the extent that the liability is attributable to an understatement amount of which she had no reason to know.
This issue was addressed directly in Freman,10 in which Kari Jane Freman admitted to knowing about a $90,000 understatement in the joint federal income tax return she filed with her husband but alleged that she was unaware that the full amount of the understatement was $172,342. The Tax Court denied her request for partial relief under section 6015(b)(2), reasoning that even though she did not have actual knowledge of the full extent of the understatement, she had reason to know the full amount because it had been deposited into the couple’s joint bank account.11
A taxpayer who does not qualify for innocent spouse relief under section 6015(b) may seek equitable relief from joint and several tax liability as set forth in section 6015(f). Equitable relief is available when it is shown, based on a facts and circumstances analysis, that it would be inequitable to hold the requesting spouse liable.
Rev. Proc. 2013-34, 2013-43 IRB 397, sets forth a nonexclusive list of considerations in determining whether it would be inequitable to hold a requesting spouse liable for understatements on a jointly filed tax return. The criteria for relief set forth in section 6015(f) are generally the same equitable factors considered for purposes of section 6015(b)(1)(D). Consequently, if a requesting spouse cannot demonstrate that it would be inequitable to hold her responsible for joint and several tax liability in accordance with the requirement for innocent spouse relief, she will also be unable to obtain equitable relief.
D. Tax Court Decision
Podlucky involved tax years 2003-2006. The Tax Court denied Karla’s claim for innocent spouse relief under section 6015(b). Based on a facts and circumstances analysis, the court determined that the fraudulently obtained funds were attributable to her, she either knew or had reason to know of the underpayments considering the couple’s lavish lifestyle in comparison with the income they reported on their joint return, and it would not be inequitable to hold her liable.
The Tax Court determined that the understatement was also attributable to Karla, noting that she had played a “crucial role” in Greg’s fraudulent actions.12 The court found that she had directly benefited from the fraud, as indicated by her involvement in the construction of the mansion and the purchase of luxury jewelry.13 Notably, Karla had signature authority over the bank accounts of two shell companies that had been created by her husband to carry out the fraud. During the relevant years, she used that authority to issue over 100 checks worth more than $6 million, for purchases that benefited her personally.14 Buttressing Karla’s claimed belief that some of the jewelry was investment property of LNI, Greg said the jewelry, some of which was customized to fit Karla, had been purchased to enable LNI to barter for tea with “Tibetan monks.”15 The Tax Court labeled this assertion “utterly implausible.”16 The court further noted that Karla had personally participated in jewelry measurements and provided detailed written instructions to the jewelers regarding her personal design preferences.17 Also, she conceded that much of the jewelry was her own property.18
The Tax Court found that Karla knew or had reason to know of the underpayment of taxes. This conclusion relied on the fact that her expenditures far exceeded the reported income on their joint tax returns during the relevant years, which ranged from $350,000 to $600,000.19 Karla made significant expenditures on jewelry, for which she signed checks for millions of dollars. Karla also wrote checks drawn on the bank accounts of the shell companies to architects and contractors for the construction of the mansion she shared with Greg. This further highlights the extent of her involvement in the fraudulent financial scheme. The court concluded that a reasonable person in Karla’s position would have questioned the family’s ability to afford their lifestyle.
E. Appellate History
The Podluckys appealed the adverse Tax Court decision to the Ninth Circuit, which dismissed the case for lack of jurisdiction. Based on the facts and circumstances found by the Tax Court, had the taxpayers been able to proceed with an appeal, our machine-learning technology predicts that any appeal regarding innocent spouse relief would have failed.
We now turn to consider how Blue J’s machine-learning technology would have predicted an unfavorable outcome for Karla with high confidence and delve into which factors were most influential in driving the predicted result. However, we also explore how she might best have positioned her case based on the known evidence. Our analysis suggests that she might have been able to obtain partial relief if she had convinced the fact-finder that it was plausible that she was unaware of the full extent of the underpayment and had maintained separate bank accounts from her husband.
III. Machine-Learning Analysis
A. Blue J’s Prediction
Blue J uses machine learning to assess and model a data set of approximately 350 cases that have determined whether innocent spouse or equitable relief is available under section 6015. Blue J’s predictions consider and weigh the factors that judges refer to and rely on in their decisions in a way that best fits the entirety of the accumulated case law.
The predictive technology works as follows. Users are first prompted to enter all the relevant facts of their case. Blue J’s model then takes these facts and, based on an analysis of the prior case law on which Blue J has been trained, predicts how likely it is that a court would grant innocent spouse or equitable relief based on those stipulated facts and circumstances. As part of its output, Blue J also attaches the degree of confidence it has in that prediction. In almost two-thirds of all the relevant cases in our system, the court declined to grant any relief to the requesting spouse.
Based on the facts found by the Tax Court in Karla’s case, Blue J predicts with greater than 95 percent confidence that neither innocent spouse nor equitable relief would have been granted on appeal. Even when tweaking the Tax Court’s characterization of the facts, Blue J’s analysis suggests that Karla’s case had little hope of success; her facts would have had to be different for her request to be given a reasonable chance of a different outcome. This conclusion holds even when characterizing elements not explicitly mentioned in the decision that were in Karla’s favor, such as her lack of work experience or relevant education, lack of involvement in tax return preparation, absence of oversight over the household budget, and lack of involvement in the financial affairs of the business.
The innocent spouse and equitable relief analyses necessarily require courts to assess the evidence, make findings of fact based on that evidence, and weigh those facts against each other to conclude whether relief is appropriate. Our predictor considers all these factors and provides professionals with an expected outcome and the most relevant research results. Here, Blue J’s algorithm considered several significant factors addressed by the courts, including (1) the fact that the Podluckys incurred extraordinary personal expenses, (2) the fact that Karla had knowledge of and access to the bank accounts at issue, and (3) the extent to which Karla was involved in managing household finances in the tax years at issue.
The total value of the checks Karla signed from 2003 to 2006 exceeded approximately $6.6 million,20 whereas the IRS determined the amount of underreported income to be nearly $35 million.21 Karla personally paid for approximately $2 million worth of jewelry,22 while jewelry valued at $22 million was found in the secret room at LNI.23 Also, while Karla should have been aware that her family could not afford the construction of their mansion with their reported income, there is no evidence that she knew the full extent of its $11 million value.24 Further, Greg’s toy trains were stored in the LNI warehouse, and the facts do not indicate that Karla knew of the expenditures for them at all.
B. Reframing the Understatement of Taxes
Section 6015(b)(2) provides that a taxpayer who fails to qualify for innocent spouse relief solely because she knew or had reason to know of the understatement may be partially relieved of liability if she did not know, or have reason to know, the extent of the understatement of income. In other words, the taxpayer may be relieved of liability for any portion of the understatement for which she did not know and had no reason to know, to the extent the underpayment is attributable to those items.
Had Karla been able to establish that she was unaware of approximately $28 million of the $35 million in income, the analysis regarding her request for innocent spouse relief would require an inquiry into the amounts Greg transferred into the bank accounts of the shell corporations. It is not clear from the facts whether the additional unreported income was transferred to a bank account to which Karla had access.
Although Karla, as a pro se litigant, attempted to articulate this principle in her defense without apparent reference to this provision, the Tax Court’s analysis evaluated only whether Karla ought to have been aware of the understatement of taxes.25 The Tax Court did not examine each component of the understated amount to determine whether there was any portion of it that Karla did not know about and had no reason to know about. Arguably, if a portion of the understatement was derived from items that were paid for from a bank account to which Karla had no access, Karla may have had no reason to know about the extent of the understatement if she had no knowledge of the expenditures.
Items Contributing to Understatement
Extent of Karla’s Knowledge
Jewelry at residence
Jewelry at LNI
Probable knowledge/lack of knowledge
As can be seen, although Karla ought to have been aware that the tax returns understated their tax liability, she arguably was not aware of the full extent of the understatement based on her varying degrees of knowledge about what the funds were spent on.
The Tax Court did not differentiate between Karla’s varying levels of knowledge for each of the expenditures. Instead, it concluded, based on Karla’s involvement, that she knew or should have known when signing the 2003-2006 returns that they significantly understated the tax owed. The Tax Court’s assessment can thus be seen as a binary determination of whether Karla should have known of the understatement writ large, rather than a carefully calibrated assessment of the extent to which she could have been found to be aware of each item that contributed to the understatement.
Given that Karla may not have been aware of all of Greg’s purchases or the full amount of the misappropriated funds, we can use machine learning to consider what result may have been possible if the Tax Court considered the extent of the understatement, rather than making an overall finding regarding the full understatement.
C. Disputed Facts: Section 6015(b)
Some of the main points of contention regarding whether Karla knew or had reason to know about the understatement of tax in the innocent spouse relief analysis included whether the understatement was attributable to her, whether she knew or had reason to know of the understatement, and to what extent she was aware of the extent of the understatement.
Karla argued, albeit apparently without citing relevant authority, that at least part of the understatement ought not to have been attributable to her. The Tax Court did not explicitly consider whether Karla should have been granted partial relief under section 6015(b)(2) for a portion of the luxury jewelry or the toy trains stored on company premises.
D. Testing the Facts: Section 6015(b)
To ascertain whether Karla’s asserted ignorance about the full extent of the unreported income would be determinative to the result in this case, we methodically tested the facts using Blue J’s machine-learning technology.
We began our first prediction using the Tax Court’s findings of fact, initially treating Karla’s knowledge of the fraud as indicative of knowledge of the full extent of the purchases, and therefore of the full understated amount on the return. Based on these facts, Blue J’s machine-learning technology predicts with greater than 95 percent confidence that innocent spouse relief is not available to Karla (Scenario 1).
We next used the predictor to test whether the outcome would be different when considering only Karla’s involvement in the understatement as it related to the purchases of the model trains and the portion of luxury jewelry that was kept at LNI headquarters and that she may have sincerely disclaimed as her own (Scenario 2), while holding all other factors constant.
Despite implementing these changes to the model to render a result based on facts more favorable to Karla, her chance of securing innocent spouse relief remained dismal. Therefore, Karla’s outcome could not have been improved solely by proof that she neither knew of nor benefited from these additional purchases made by Greg.
To understand why her odds did not improve, we tested which other factors in the model would need to change for the taxpayer to obtain a different result. For example, we tested various scenarios, including those in which Karla was not involved in the decision-making of major purchases, in which she exercised more due diligence in reviewing her tax return before filing, and in which Greg used a separate bank account that Karla had no access to or knowledge of.
Based on this review, we determined that the most significant factor affecting Karla’s outcome on the more favorable variation of facts was her access to the bank accounts of both shell companies. If Greg had purchased the additional items using a separate bank account, Karla’s likelihood of success rises to a passable 59 percent.
Karla sought relief from joint and several liability for the relevant tax years under section 6015(b) and (f). In her claim, Karla said she did not act with fraudulent intent, saying that she was unable to differentiate between her own jewelry and jewelry that ostensibly belonged to LNI26 — particularly the jewelry that was kept in a secret room at LNI headquarters.27
However, the Tax Court held that she did not qualify for either innocent spouse relief or equitable relief, finding that the understatement on the joint federal income tax return was attributable to her, she had reason to know of the understatement, and it would not be inequitable to hold her liable. The court further noted that it was Karla’s responsibility, and not that of the IRS, to differentiate between jewelry belonging to her and jewelry that was an investment of LNI.28
The ownership of this jewelry and Karla’s imputed knowledge of its existence are key facts in determining her chance of success in a claim for innocent spouse relief under section 6015(b). This information speaks to whether the items are attributable to her under paragraph (1)(B) and whether she had reason to know within the meaning of paragraph (1)(C).
Considerations for Partial Relief
Knowledge of understatement
Knowledge of spouse’s extraordinary personal purchases
Involvement in making decisions on major purchases
Review of tax return
Use of separate bank account by non-requesting spouse
Likelihood of innocent spouse relief
1. Scenario 1: Tax Court’s facts.
In our first scenario in Table 2, Karla was held liable for the full amount of the underpayment. Because she failed to differentiate whether any jewelry belonged to LNI, the court found that the full amount of the understatement was attributable to her. The full values of the mansion29 and the toy trains are also included. Based on this view of the facts adopted by the Tax Court, our machine-learning technology predicts with greater than 95 percent certainty that Karla would not prevail on her claim for innocent spouse relief on appeal.
2. Scenario 2: Partial knowledge.
We next considered whether the outcome could have been different if Karla had been able to isolate the luxury jewelry belonging to her (perhaps by reference to the $2 million that she personally spent) and show that she was unaware that Greg had purchased any toy trains. Under section 6015(b)(2), Karla might qualify for partial relief. She would be claiming that (1) she was not involved in their purchase and did not benefit from the purchases; and (2) she had no knowledge of these additional purchases. However, Blue J’s machine-learning technology still predicts with 92 percent confidence that she would be unable to obtain relief even with these more favorable facts.
3. Scenario 3: Separate accounts.
A closer review of the facts affecting the outcome of the machine-learning process identified an important driver — her access to the bank accounts of the shell companies. Similar to the facts in Freman, Karla had reason to know of these additional purchases made by Greg because she had access to the accounts showing the flow of cash in the account ledgers. The facts in both Freman and Podlucky, as articulated by the courts, are unlike the example in the regulation,30 in which the non-requesting spouse maintained a separate bank account into which he deposited the embezzled funds, transferring only a small portion of the stolen money to the couple’s joint bank account. Taking this into account, a final tweak to the facts produces a different result for Karla. If we assume, as before, that Karla had no knowledge of or benefit from the additional jewelry or Greg’s trains, and we stipulate that Greg routed the money for these items through a separate bank account to which Karla did not have access, then Karla’s chance of success at obtaining partial innocent spouse relief to the extent of these additional items jumps to 59 percent.
While still uncertain, this result produces at least a viable chance of success for Karla. When we submit variations on these facts to Blue J’s machine-learning algorithm, we find that Karla’s chance of obtaining innocent spouse relief for any portion of the understatement shifts in her favor. This result from our machine-learning technology is in keeping with section 6015(b)(2), reg. section 1.6015-2(e), and case precedent.
Blue J’s machine-learning model indicates that Karla could not have prevailed on her claim for innocent spouse relief under section 6015 based on the facts adopted by the Tax Court. Those facts indicate that the understatement was attributable to her, and a reasonable person in her position would have known that the family’s expenditures could not have been supported by the income reported on the couple’s federal income tax return. Given these findings, the court determined it would not be inequitable to hold her jointly and severally liable. Our predictor found a greater than 95 percent certainty that she would not have prevailed on appeal of the Tax Court’s denial of innocent spouse or equitable relief.
However, we also predict that Karla’s liability could have been partially reduced if she had established that she neither benefited from nor had reason to know of the full extent of the fraud. Specifically, she may have been able to obtain relief on the strength of facts indicating that she believed the luxury jewelry stored at company headquarters did not belong to her and that she did not know of Greg’s toy trains kept in the company warehouse.
On these facts, Karla’s chances of success depended heavily on whether Greg purchased these items using a separate bank account to which she did not have access. When we assume that Karla had access to the bank account used to make these purchases, her chance of being denied relief is 92 percent. However, when we assume that Greg did not share access to the relevant bank account with Karla, her ability to obtain innocent spouse relief under section 6015(b)(2) becomes 59 percent. Given the crucial nature of this factual determination, it is vital for a taxpayer in Karla’s situation to raise details such as these when pursuing relief.
This analysis demonstrates the benefits of using machine-learning technology at both the trial and appellate levels for counsel in determining a client’s chance of success. After a trial decision is made, technology can aid in reviewing the overall prediction and identifying the key factors that are critical in determining the most effective strategy on appeal. On appeal, technology can assist in evaluating the client’s prospects and formulating a winning approach, including by helping to make informed arguments about the facts at hand. Further, scenario testing with the help of Blue J can highlight important evidentiary factors and arguments that, although not always necessarily applicable to the case, provide valuable insight and lessons for taxpayers seeking relief under section 6015.
2 Podlucky, T.C. Memo. 2022-45, at 5 (“Greg needed a safe place to store all this jewelry. He directed an employee to build a ‘secret room’ at LNI’s headquarters in Latrobe. The secret room was described as ‘a corner of another room that had been walled off with cinder block.’ To enter the room an individual had to walk through ‘a small metal door, lift a rug, and crawl.’ Criminal investigators executed a search warrant and found in the room commercial grade safes stocked with gemstones, necklaces, watches, bracelets, and diamond rings, as well as filing cabinets cataloging each piece.”).
3 Podlucky, T.C. Memo. 2022-45.
11 See also reg. section 1.6015-2(e)(2), Example (providing partial relief for a spouse who was aware of an understatement of $120,000 but did not know of an additional $1.88 million embezzled by her husband and placed in a separate bank account to support his gambling habit).
12 Podlucky, T.C. Memo. 2022-45, at 23.
13 Id. at 24.
15 Id. at 13.
17 Id. at 13-14.
18 Interestingly, she did not concede that all the jewelry was her own property.
19 Podlucky, T.C. Memo. 2022-45, at 26.
20 Id. at 24.
21 Id. at 27.
22 Id. at 26.
23 Id. at 12.
25 While Karla conceded that some of the jewelry belonged to her, she appears to claim that she was unaware of the full amount spent on the jewelry and assumed all the jewelry was purchased with a $5 million investment for LNI. For this reason, Karla argued that the commissioner never differentiated between her personal jewelry and the investment jewelry for LNI. See id. at 24.
26 Id. at 24.
27 Id. at 25.
28 Id. at 24.
29 Id. at 27.