By Roger Isaacs
As the power of technology and artificial intelligence increases exponentially, it inevitably plays an ever-greater role in the justice system. There have been reports of AI being used in some foreign jurisdictions to evaluate the risk of recidivism and then to inform criminal sentencing decisions. This article considers the degree to which AI can be used to augment or replace the role of the accountancy expert witness in cases involving business valuations.
Undoubtedly AI and machine learning can be an invaluable tool for the analysis of data and it is frequently used to great effect by forensic accountants. However more recently, a number of firms have started to offer automated business valuations. Such valuations are perhaps one of the most subjective areas on whichaccountancy expert witnesses are instructed to opine.
The proponents of AI are quick to criticise business valuers who do not provide arithmetic calculations to derive the multiples that they apply to their estimates of likely future maintainable earnings. However, arguably there is a danger in thinking that business valuations can ever be entirely formulaic. Indeed, the essence of the role of the expert witness is to provide the court with the benefit of the expert’s professional judgment.
This raises the question as to whether the judgment of a machine can be as good as, if not better, than the judgment of a suitably qualified individual.
Of course, an algorithm is only as effective as the people who designed it and there have been several well-publicised examples of software designers having introduced their own unconscious bias into their programmes. Another challenge for those who try to value businesses using AI, is the lack of relevant statistics on which the AI is able to carry out its computations. The old adage “Garbage in garbage out” has
never been more relevant.
Ironically, the bigger a business is, the easier it can sometimes be to value it. Firstly, large businesses, unlike their smaller counterparts, can be compared meaningfully with listed companies. By contrast, experts who apply statistics from public companies to family businesses have quite rightly been accused by the courts of using the “accounts of Tesco to value the village store in Ambridge”.
Secondly, large businesses tend to produce more accurate data in terms of detailed monthly management accounts and cash flow forecasts. By contrast, a small family business may well only ever produce its accounts once a year for tax purposes.
Thirdly, the accounts of larger businesses tend not to be distorted by the “lifestyle costs” that are often a feature of companies that are owned by their directors. Such costs might, for example comprise wages paid to family members at above market rates or running costs associated with prestige motor vehicles or even helicopters and yachts. The expenses would not be incurred by a hypothetical purchaser of the business and
therefore a valuer needs to identify them and adjust for them in any valuation.
Of course an AI system is more than capable of managing the arithmetic of such adjustments but it will only be able to do so if it is fed the relevant raw data, which will never be apparent from the face of a company’s accounts.
Perhaps the biggest challenge for any AI system is that it relies on published data to calculate the multiples that are applied to earnings. In doing so it is restricted to consideration of a dataset that it inherently skewed because it excludes the very large number of unquoted companies whose shares are never traded because their performance is too lacklustre to command a price that the incumbent shareholders would be willing to accept.
Statistically, only a small percentage of companies are ever sold and those that are have invariably either generated consistent healthy profits or are on a growth trajectory or both. By contrast, the majority companies have good years and bad years. Their performance may be erratic and unpredictable. Often, they will have had years in which they incurred losses. Prospective buyers will consider such businesses as being inherently risky. The greater the perceived risk, the lower the multiple.
When the owners of these types of business seek advice from corporate finance specialists about how to achieve a sale, they can be disappointed to learn how little their companies are worth. They may well then decide not to sell unless they are forced to do so.
The dataset of published multiples takes no account of these types of business and any AI system would need to be able to make appropriate adjustments to take this factor or other qualitative factors into account.
The problem is that there is no statistical basis on which to do so because, by definition, the relevant data does not exist. That is where professional judgment plays a key role.
For these reasons, AI is becoming a useful element of the business valuer’s toolkit but it would be premature for forensic accountants to panic that they are in imminent danger of being replaced by machines.