This leads to more far-reaching questions.
(1) Can technology make better decisions than humans? and
(2) Can it make the best decisions?
"Tax Technologists often talk about AI for Tax when they really mean machine learning, and they invariably over promise and under-deliver. ‘Deep Tax’ is the very opposite of this approach. Using specially developed algorithms it tackles tax problems in a way that a person isn’t capable of replicating, delivering better outcomes, and using a proven technology that far more deserves the description of Artificial Intelligence. Technology can make decisions in tax, but only if you use the correct technology for the problem, and this technology is the right solution".
Graham Tilbury, Tax Technology Expert
We naturally have our own opinions and thought it would be informative if we could provide an example that demonstrates what technology can do. Fortunately, we found an old, but still useful, example in the HMRC manuals.
The HMRC example illustrates the rules for group relief surrenders that apply for groups with non-coterminous APs. CTM80255 - Groups: group relief: non coinciding accounting periods or group relationships - example - HMRC internal manual - GOV.UK (www.gov.uk)
A group has 5 companies of which only 2 companies have the same AP dates. This scenario is not typical, but it does involve a lot of complex inter-connected decision-making concerning the sequence of surrenders and what losses to surrender.
The results were as follows (with the workings shown below).
HMRC manual Deep Tax
Profit after group relief £97k £67k
Unrelieved losses £80k £50k
The results (which have been validated) therefore confirm technology can make decisions that would otherwise be made by humans.
Deep Tax was also very quick, taking less than 1 second.
But has Deep Tax made the “best” decisions? Since the £67k of remaining profits relate entirely to that part of a company’s AP which has no overlap period the results cannot be improved upon (because no losses can be surrendered). Hence this demonstrates that technology can make the best decisions.
Lastly, has Deep Tax made “better” decisions? Looking at the results, it would be tempting to conclude the answer is "yes". However, the HMRC results are probably only intended to show how the rules operate, rather than an attempt to optimise their use. It therefore depends on the individual as to what are the best results that could be produced manually. Lengthy and inter-connected calculations are required, as borne out by the 5 pages of workings in the HMRC manual. We know from a particular set of live data which had similarities that this task is fraught with complexity and is very time consuming, even with CT software. Therefore, it is our view that technology can make better decisions, and sometimes a lot better!
Workings
HMRC (uses monthly time apportionment)
Deep Tax (uses daily time apportionment)
The example only looks at one aspect of decision-making where each company has just one accounting period. The case study referred to on the News page was immensely complex and illustrates the capability of Deep Tax to solve extremely large and complex problems.