Tax planning and the structuring of transactions are fairly central to the work of many tax lawyers. There are elements of this work that would be incredibly challenging for a computer. For example, there is no reason to program a computer to determine if shares in a company qualify for Business Property Relief (BPR). That would be difficult and fairly pointless.
On the other hand, there are aspects of tax planning and the structuring of transactions that a computer would be extremely good at. In particular, once a client’s situation is properly understood (e.g. the client’s assets, family relationships, and goals), a computer could be programmed to identify how that client should structure his or her affairs so as to achieve the desired goals in a tax efficient manner.
This short paper outlines a simple program that can do just that, a task that I refer to as “designing tax efficient solutions” for the sake of simplicity.
In designing tax efficient solutions, human lawyers use a number of rules of thumb. If you pick up any article on a recent tax change, invariably a portion of the article will consider the effect that the change will have on clients and provide some general advise for tax professionals. This might read, “Tax advisers should be aware of this new rule, and if (insert generic scenario) the client might want to consider (insert possible tax efficient solution).” Tax lawyers do not think through every possible scenario and calculate the tax implications before advising a client; rather, tax lawyers use these general rules to find the best solution. This is evident in the fact that a tax lawyer could provide an intelligent answer to the question of when a person should incorporate a company versus set up as a partnership.
This is not how a computer would work. An automated tax adviser, for lack of a better term, would have no general understanding of when certain structures will and will not work for a client. However, computers are vastly superior to humans in terms of iterating possible solutions and running calculations. A computer does not need general rules because it can search over every possible way that a client could structure his or her tax affairs.
I think it is easiest to understand the “automated tax adviser” program as a computer game. We can program the game to have certain characters, namely the client and any relevant friends and family members. We can also assign property to each character, and each piece of property can have its own features. For example, if a client has unquoted shares that a tax lawyer identifies as qualifying for BPR, this can be assigned to those shares as an attribute.
The game’s fundamentals would consist primarily of the mechanical tax rules. That is, the game would not have any idea when shares qualify for BPR and when they do not, but it would know the tax effect of qualifying for BPR along with some of the BPR rules such as the holding period requirement. It would know tax rates, how Potentially Exempt Transfers work, how CGT is calculated, etc. The game’s fundamentals would also consist of the allowable interactions between characters. Characters can, for example, make gratuitous transfers, engage in arm’s length transactions, or vary the will of a person who has died.
One can imagine a tax lawyer “playing” this game by controlling each of the characters, and demonstrating the tax effect of a particular plan that he or she came up with. But we could also program a computer to play this game. At the simplest level, the computer could play the game by choosing a series of random actions. By repeating this random gameplay and recording the result each time, we now have a series of possible solutions and the associated tax implications. Iterate enough times and eventually you reach a point where there is a very high chance that every solution has been discovered.
Of course, having 50,000 solutions to choose from would be of little help to a tax lawyer. However, there are a number of ways in which we can reduce the number to only the solutions that might be useful. For example, we might set certain goals that a client has, and compare solutions against those goals. These goals might be hard rules or they might be flexible. An example of a hard rule would be “granddaughter X to receive at least £20,000”. Any solution that does not satisfy this rule can then be discarded. A flexible rule would be “everything else equal, the granddaughter receiving money is preferable to the mother receiving money”. Any solution that is otherwise equal except has the mother receiving a sum that a different scenario gives to the granddaughter can be discarded. In addition, there is always a Pareto rule: any solution in which a person could be made better off without anyone else being made worse off (except for HMRC) should be discarded.
By applying these types of rules, the program could provide a tax lawyer with a limited set of solutions to consider. These solutions would ideally be sorted in some meaningful way using a decision rule.
The goal of this program is not necessarily to save time for tax lawyers, or even make their job any easier. The goal of the program is to improve the quality of advice that is given. A tax lawyer would be presented with possible ways of structuring a client’s tax affairs that maybe he or she never considered before. In doing so, the computer also helps to teach the lawyer new general rules.
A very simple example of this has already occurred. I wrote a limited version of this program already, and applied it to a simple scenario I made up. In that scenario, I had a man that died, and his spouse dying four years later (I gave it a specific time of death for simplicity). One aspect of the computer’s solution involved the man transferring a certain asset to his spouse, and the spouse immediately transferring this asset to her granddaughter. When I first looked at it, I thought that was weird, as they could vary the will to transfer directly from the man to his granddaughter. But of course, that has a tax cost (as the man had already used the entirety of his nil rate band). By transferring to the spouse (tax free) followed by a PET that becomes chargeable four years later, the IHT rate is reduced to less than 40%.
This is painfully obvious to anyone reading this I am sure, but the important point is that I never told the computer to do that. I could not have, as I had not thought of it myself. It seems almost certain that similar situations would occur if this program were developed and applied to more complex tax problems. The best chess and Go players in the world say that they have learned by watching how a computer plays the game, and I think it would be no different here. And when the computer does outperform the human lawyer, the lawyer is able to give the client better advice than he or she otherwise would be able to in the absence of the computer.
The program could also be written in such a way that allows for vastly improved monitoring of clients’ tax affairs. Once the lawyer has advised the client and the client has chosen a particular course of action, this information could be communicated to the computer. A year later, if certain tax rules change, the computer can re-run its calculations for every client in the system. The computer could then flag certain clients that might benefit by making some change or taking an alternative course of action. This level of service is difficult to achieve in the absence of such a program, as it is impossible for a lawyer to review the effect of a change on every client. It would hopefully be comforting to clients to know that there is also a computer assisting the lawyer in watching over their tax affairs.