The due diligence process involves requesting, organising, distributing, and preserving the confidentiality of an enormous quantity of material, much of it requested from the target company. Since the process and the material requested tends to be very similar for each M&A transaction, this data-gathering exercise is well suited to automation with AI.
Indeed with natural language processing, due diligence software can scan documents for secure distribution, sending relevant documents to due diligence team members together with a summary of document findings, identification of key risks and a highlight of critical issues.
Through machine learning, AI due diligence systems can be trained to sift through mountains of data to identify and reveal particular signals or information of value. Indeed, the system can be taught to recognise relationships between the nuggets of information it surfaces and offer opinions based on what it “sees.”
This article examines how you can apply an AI-powered solution to due diligence, what things to consider when choosing an AI solution for due diligence, and how current solutions don’t adequately cater for accounting fraud. Finally, it looks at one success story: An asset manager who uses an AI solution from Transparently.AI as part of his due diligence process.
Table of contents
- Applying AI software to the four types of due diligence
- How to choose AI due diligence software
- Due diligence software: Five things to consider
- Problem: Due diligence solutions don't focus on accounting fraud
- Revolutionising due diligence software with AI
- How one fund manager uses AI software for due diligence
Applying AI software to the four types of due diligence
Obviously, the scope and nature of enquiry would vary according to each of the four aspects of due diligence.
Financial due diligence software
FDD would scrutinise financial statements, profit/loss records, tax filings, and annual reports looking for set patterns established by machine learning. Finding these patterns would alert readers to account manipulation.
The idea would be to provide an estimate of the inaccuracy of the statements and most important aspects of the accounts to research further.
Based on these findings the software would suggest adjustments to the targets reported numbers for estimating earnings and cash flow. The software would then take the current estimated data and assume a growth rate based on an analysis of market position, the competitive landscape, and growth potential. These numbers would drive the assessed value of the deal.
Legal due diligence software
LDD software powered by AI would examine legal aspects, contracts, agreements, and relationships with stakeholders looking for unusual clauses, predetermined patterns in certain clauses and specific legal terms that might alert the FDD team of areas of elevated legal concern for further enquiry.
Based on these findings the LDD software would suggest an estimate of probable future legal liability. This would feed into the FDD.
Operational due diligence software
AI software for ODD would run input-output models to provide a deep comparison of the company with industry peers. Such an analysis would identify key areas of weakness in business practices, operational efficiency, and risk management.
These risk areas would be where value-creation opportunities are greatest. The software would then propose potential value enhancement if those aspects of the business were improved. The estimated value of these enhancements would be reported in the FDD.
Tax due diligence software
AI software for TDD could scan for anomalies in the target’s tax and financial reporting data versus industry peers. The software would then allow for a deeper questioning on specific issues. It would scan the records to document evidence that taxes were correct, paid and properly accounted for.
The software would be trained to search for anomalies within a company’s tax filings. Any discovered tax liability or asset would feed into the FDD.
The bottom line is that in the future, machine learning combined with big data could basically prepare the bulk of the due diligence reports for stakeholders, including the acquiring company’s management.
In the fullness of time, we expect that due diligence reports will be 90% researched and written by AI software.
The software could recommend specialist review in key areas. These findings could be introduced into the report as risks to be addressed. These sections would, for example, be answered by forensic accountants, environmental compliance specialists, intellectual property lawyers or regulatory experts.
In the fullness of time, we expect that due diligence reports will be 90% researched and written by AI software.
How to choose AI due diligence software
When choosing software, it is important to recognise that due diligence is multi-faceted. It is unlikely that any single piece of software will ever perform all four due diligence functions or provide all facets of AI such as machine learning, natural language processing and computer vision.
It is likely that different specialist AI providers will prove different aspects of the due diligence process. Some of the software functions we included in the discussion above are not yet perfected and some are yet to be developed.
It seems likely that DD software will fragment into at least six groupings:
-
One group would request, gather, read and distribute the documentation.
-
Another group of software will conduct searches within the scanned documents using natural language processing in combination with some times of machine learning. It is possible that this software could operate across the four types of due diligence but also likely that some will be better at legal and others will be better at tax or operations.
-
Four additional groups of software providers would likely tackle each of the various aspects of the four components of due diligence.
Of course, some providers will offer services across a range of functions but the different parts of due diligence are highly specialised and it is unlikely that a generalist product will do any of the components better than specialists.
Five things to consider when purchasing AI due diligence software
When selecting AI-powered due diligence software, one should consider five key aspects:
1. The most essential tasks
Your software must be able to help you answer five key questions upon which all due diligence hinges. These are:
-
Is there a serious problem with the target’s financial accounts?
-
Does the target face major litigation risk?
-
Does the deal face serious tax consequences or offer opportunities
-
Can cash generation at the target be improved?
-
Are there operational synergies with the target we can exploit?
Your software must be able to answer these core questions well. Everything else is nice to have but not essential.
2. Capability
Not all software platforms are created equal. Some are highly sophisticated while others are hardly better than a simple decision tree running on a database. The first task of the due diligence process should be to select a small team to determine what software solutions will be employed. This will be a learning curve. Some will employ specialist consultants.
Most important, this team should investigate what kind of tech the software is using. For example, what kind of algorithms is it using? Does it employ supervised learning, unsupervised learning or reinforced learning?
Seek a software provider with a strong pedigree, a strong product and strong customer support.
3. Ease of use
For most members of the due diligence teams, due diligence will not be their full-time occupation, it is a side-task to which they have been assigned. It is likely a task they do not relish. Most will have no experience using AI systems, especially those that need training. They will need a plug-and-play solution.
It is suggested that each team must test the software for ease of use and to see how easily it will fit into the due diligence process.
In many instances, report generation is the weakest link of AI software. This is because the people who build the software are not the same people who will use it and because the output is generally the last phase of software building. Anyone who has worked to deadlines will know that the last phase of any project is typically the most squeezed for time.
Those who build software are mostly focussed on making sure that it works. When testing the system these engineers will interact with the software in a different way than eventual end users. They will write code to ask specific questions or check various outputs They will not rely on the software interface because it serves a different purpose.
This approach means that the software engineers are typically less familiar than you might think with how the software presents its results, how the results are explained or how the system should be configured to look at the results in a different way. In many cases, software changes due to requests and complaints from users.
Each team should spend considerable time considering how the software generates output. For example, does the software only show output for one year or can it show a specific item for a run of years. The less effort needed to convert output into the DD report, the better.
4. Interoperability
Each aspect of the due diligence process feeds into several others. Where appropriate, it will be important to ensure that services being used allow for information exchange.
For example, the ODD will require detailed information on the financial accounts and thus will need to share the financial database with the FDD. Similarly, the TDD software will also want to share data with the FDD. The LDD and ODD teams will also have an overlap in data needs.
This need for interoperability might not be essential if the due diligence being conducted for a one-off deal or transaction, but will be critical for firms that continuously engage in due diligence for their business operations, like asset managers, auditors and banks.
If two or more DD systems use common data formats and communication protocols, then they are capable of communicating with each other and they exhibit syntactic interoperability. XML and SQL are examples of common data formats and protocols.
Beyond the ability of two or more computer systems to exchange information, semantic interoperability is the ability to automatically interpret the information exchanged meaningfully and accurately in order to produce useful results as defined by the end users of both systems.
To achieve semantic interoperability, both sides must refer to a common information exchange reference model. The content of the information exchange requests are unambiguously defined: what is sent is the same as what is understood. This capability will eventually sit at the core of all good DD software.
5. Security
The process of due diligence requires a lot of document and results sharing. Companies engaged in M&A transactions are thus more vulnerable to data security breaches than they would normally be.
Moreover, the very nature of M&A transactions means that inside knowledge of these transactions offer the prospect of considerable financial gain.
Thus companies undertaking M&A due diligence will attract the attention of those looking to steal data to benefit, if possible, from trading in the companies involved. The key data risk will be found in the DD software and thus the security features of the software should be carefully considered.
Problem: Current due diligence solutions don't focus on accounting fraud
An internet search for due diligence software will deliver a raft of infomercials and websites selling two types of due diligence software:
-
The first can automate the process of requesting, gathering and sharing the documents needed in due diligence.
-
The second uses natural language programming combined with some form of machine learning to search the documents for typical red flags in contracts, licences, or correspondence.
In other words, most DD software is currently geared towards performing LDD and TDD. Virtually none is geared towards FDD and ODD.
Most DD software is currently geared towards performing LDD and TDD. Virtually none is geared towards FDD and ODD.
This omission is serious because it means that due diligence software currently fails to address three of the five imperative due diligence questions: 1) Is there a serious problem with the target’s financial accounts? 2) Can cash generation at the target be improved? And, 3) are there operational synergies with the target we can exploit?
In fact, questions 2) and 3) cannot be answered until one is assured that the accounts are not manipulated. It turns out that detecting accounting fraud is essential in order to answer three of the five most important questions to be answered by due diligence.
Software to detect accounting fraud does exist. Banks and accounting forms use various kinds of software to look for specific forensic accounting patterns. Forensic accountants working in FDD teams also use some software to identify signs of potential fraud. This software, however, does not use AI.
Transparently.AI has developed an AI-based software platform that specifically analyses the financial statements of public companies for manipulation.
Revolutionising due diligence software with AI
The key AI technology required for the automation of due diligence already exists. With the breakthrough of Transparently.AI, the stage is set for the revolution in FDD in the next several years. The next major advancement will come with AI software specifically designed to optimise corporate cash generation.
It will take several years before a developer attempts to combine the disparate elements from various software companies into a unified and coherent product. However one can imagine every major accounting firm has begun taking steps towards such an end.
What is currently a mind-numbing task of collecting, merging, reading, dissecting and analysing thousands of documents will be reduced to a largely automated process leaving more time for analysis and risk assessment. Each of the highly technical legal, taxation, financial and operational aspects will be performed by AI software that has evolved to perform each specific function. Thanks to machine learning, the systems will get better at each function over time.
But the true revolution will come from greatly improved efficiency across the entire corporate sector. The enormous waste from failed M&A activity and ill-considered investment decisions will diminish. Companies will be able to assess possible targets in a matter of weeks rather than months.
All companies will have to clean their financial accounts as the due diligence process will drive down the relative value of companies with manipulated accounts.
How one fund manager uses AI software for due diligence
Richard Firth uses Transparently.AI’s solution, the Manipulation Risk Analyzer, as part of the 120-question due diligence questionnaire his Firth Investment Management uses to vet the stock picks for his main small-cap fund.
Firth’s due-diligence questionnaire emerged out of necessity. Small-cap stocks are riskier, there is less attention from the market, less information is available, and often there is less transparency. Transparently.AI’s MRA helps Firth filter out high-risk businesses and help him pick out the 50 stocks he holds in his fund from a potential universe of 3,000 listed companies.
“Transparently.AI is now integrated within our process,” Firth said. “It’s a good starting point to filter for stocks that might be interesting for us to spend time on. We don’t have to hold anything that doesn’t pass our checks.”
Read more about Richard Firth’s experience with using AI for due diligence here.
Subscribe to our blog
Sign up for more AI-powered insights