How we do AI: The accounting risk clusters

Mark Jolley

December 11, 2025

This is the second in a series of four articles detailing Transparently.ai’s approach to harnessing artificial intelligence and machine learning to build its tools to detect accounting fraud.

In our previous article, we gave you the basics of Transparently’s approach to artificial intelligence and introduced you to the accounting risk clusters that our tool analyses. Let's dig a little deeper.

You will likely have heard that AI uses pattern recognition to learn. What this means is that AI identifies statistical regularities from massive amounts of data and then applies those learned regularities to make predictions or decisions on new, unseen data.

AI’s “intelligence” emerges from recognizing, combining and extrapolating statistical regularities better and faster than humans can in specific domains. Once an AI is trained to recognize patterns, it can be used to assess the likelihood of almost any event.

Our earlier article likened AI to a highly trained medical diagnostician, and a company's financial data to a patient's complete medical history, with blood tests, X-rays, symptoms, lifestyle habits and so forth. When doctors diagnose an illness, they don't just look at one symptom in isolation. If a patient has a fever, a cough, and body aches, the doctor groups those symptoms together and recognizes a pattern that points to, say, a flu infection. They don't just say "you have a cough," they say "you have the flu," which is a cluster of related symptoms.

Our AI's pattern recognition works in a very similar way with financial data. Instead of just flagging individual "symptoms" like "this number is a bit high" or "that ratio is a bit low," it's constantly looking for groups of related financial clues that, when seen together, form a recognizable pattern of potential risk.

These groups of related clues are what we call risk clusters. The AI doesn't just see an unusual investment figure; it sees that unusual investment figure along with a lack of transparency in related transactions and an unusual change in capital expenditure. When these three "symptoms" appear together, the AI recognizes that pattern as a "risk cluster" related to Investing Activity, one of our fourteen risk clusters.

So, pattern recognition is the process by which the AI identifies these specific combinations of financial behaviors and anomalies, and then it labels them as a particular risk cluster, much like a doctor labels a set of symptoms as a specific illness. It's not just about seeing the forest and the trees, but understanding how the trees form the forest.

The Transparently.ai risk clusters

Each of the 14 clusters that we focus on acts like a specialized lens, focusing on a different critical aspect of a company's financial behavior to uncover potential risks that might otherwise go unnoticed. Let's break down what each of these looks for in a company's financial story.

Accruals management

This cluster looks at how a company handles its "accruals" - revenues earned but not yet received, or expenses incurred but not yet paid. These are often estimates and can be easily manipulated.

Core question: "Is the company honestly estimating its revenues and expenses, or is it using these estimates to artificially inflate profits or hide losses?"
The pattern recognized: The AI would spot a pattern of large, year-end increases in discretionary accruals (like unbilled revenue or warranty provisions) that consistently reverse in the subsequent period, creating a 'smoothing' effect on reported earnings. This suggests management is using estimates to manipulate reported profits.

Asset quality

This cluster focuses on the nature and valuation of a company's assets, especially those that are less tangible or harder for outsiders to verify (like goodwill, certain investments, or complex financial instruments).

Core question: "Are the company's assets genuinely valuable and accurately recorded, or are they being overvalued or used to obscure financial problems?"
The pattern recognized: The AI would identify a pattern of significant, consistently increasing valuations of illiquid or hard-to-value assets (like certain investment properties or complex financial instruments) without corresponding market transactions or clear, consistent valuation methodologies. This suggests potential overvaluation to inflate the balance sheet.

Cash quality

While cash is generally harder to manipulate, this cluster looks for unusual patterns in a company's cash flows and cash balances that might indicate manipulation or underlying difficulties in generating cash.

Core question: "Is the company generating healthy, sustainable cash from its operations, or are there signs of cash flow problems being masked, or even direct manipulation of cash figures?"
The pattern recognized: The AI would flag a pattern of a significant divergence between reported profits and actual cash generated from operations, coupled with unexplained or large 'other' cash flow items, potentially involving related parties. This suggests difficulties in converting profits to cash or attempts to obscure cash movements.

Credit

This cluster examines a company's credit risk, both as a potential source of stress that might encourage manipulation and as an area where manipulation might be used to hide credit problems.

Core question: "Is the company's creditworthiness genuinely strong, or is it using accounting tricks to hide financial distress and appear more creditworthy than it is?"
The pattern recognized: The AI would identify a pattern where loan loss provisions or bad debt write-offs are unusually low relative to the growth of the loan portfolio, industry trends, or deteriorating economic conditions. This suggests understating potential credit losses to inflate current earnings.

Gearing

This cluster focuses on a company's leverage - how much debt it uses to finance its assets. High debt levels can create pressure, and manipulation might be used to obscure the true extent of this leverage.

Core question: "Is the company's debt level manageable and transparently reported, or is it using accounting methods to hide excessive debt or the stress it creates?"
The pattern recognized: The AI would spot a pattern of significant and increasing use of off-balance sheet financing arrangements or complex financial structures that effectively increase leverage but are not fully captured in traditional gearing ratios. This suggests an attempt to mask the true extent of the company's indebtedness.

Corporate governance

This cluster looks at the structures and practices within a company's leadership that might incentivize or allow accounting manipulation. It also identifies high-level accounting indicators that often signal poor governance.

Core question: "Are the company's leadership and oversight structures robust enough to prevent manipulation, or do they create an environment where it's likely to occur or go undetected?"
The pattern recognized: The AI would identify a pattern of concentrated ownership and control, a lack of independent board members, high executive compensation not clearly tied to performance, and frequent, non-arm's length transactions with related parties. This suggests a weak governance structure that could facilitate manipulation for personal gain.

Growth signals

This cluster analyzes the growth rates of specific financial line items. While growth is generally positive, unusual or inconsistent growth patterns can sometimes signal manipulation, especially if management is under pressure to maintain an illusion of rapid expansion.

Core question: "Is the company's reported growth genuine and sustainable, or is it being artificially inflated through accounting adjustments to meet expectations?"
The pattern recognized: The AI would spot a pattern where reported revenue growth is consistently high but is either unsustainable (e.g., requiring massive, increasing marketing spend) or is primarily driven by acquisitions rather than organic expansion, with organic growth slowing after each acquisition. This suggests an attempt to create an illusion of rapid, continuous growth.

Income quality

This cluster scrutinizes where a company's reported income actually comes from. Is it from core, recurring business operations, or from one-off events, unusual sources, or transactions with related parties that might not be at arm's length?

Core question: "Is the company's reported profit truly reflective of its ongoing business success, or is it being propped up by less reliable, non-recurring, or potentially manipulated income sources?"
The pattern recognized: The AI would flag a pattern where a significant and increasing portion of a company's net income is derived from non-operating, irregular, or vaguely described sources (like 'gains on asset sales' or 'income from affiliates'), while core operating profits remain stagnant or decline. This suggests that the company might be using these less transparent or one-off items to artificially boost its reported earnings and mask underlying weaknesses in its primary business.

Investing activity

This cluster focuses on how a company is spending its money on long-term assets, like buying other companies, building new factories, or investing in new technologies. It also looks at how transparent these investments are.

Core question: "Is the company making genuine, value-adding investments, or are its investment activities being used to hide problems, inflate assets, or move money around in an opaque way?"
The pattern recognized: The AI would spot a pattern of consistently high, vaguely described investment outflows (like for 'intangible assets' or 'other investments') that don't seem to translate into tangible business growth or cash generation, coupled with a growing reliance on external funding to support these opaque activities. This suggests that the "investments" might not be genuine growth drivers, or worse, could be used to move money or inflate asset values without clear accountability.

Margin signals

This cluster examines a company's profitability ratios – how much profit it makes from its sales after covering various costs. It looks for unusual or inconsistent trends in these margins compared to what's expected.

Core question: "Are the company's profit margins healthy and consistent with its operations and industry, or are there strange fluctuations or unsustainable levels that suggest manipulation to make the company appear more profitable than it is?"
The pattern recognized: The AI would detect a pattern of gross profit margins that are consistently and significantly higher than industry peers, or show unusual stability/increases despite adverse industry conditions or rising costs. This pattern can indicate manipulation of the cost of goods sold (e.g., by understating inventory costs or delaying expense recognition) to inflate reported profitability and make the company appear more efficient or successful than it truly is.

Business manipulation

This cluster looks at how a company manages its day-to-day operations and sales to influence its reported profits. It's about the decisions management makes that might temporarily boost numbers but aren't sustainable or reflect true performance.

Core question: "Is the company generating real, sustainable profits from its core business, or is it making short-term operational decisions (like cutting essential costs or pushing sales unnaturally) to make its performance look better than it is?"
The pattern recognized: The AI would identify a pattern where sales growth is significantly outstripping the growth in cash collected from customers, and simultaneously, inventory levels are rising disproportionately to sales. This pattern often points to aggressive revenue recognition tactics like "channel stuffing" (forcing more product onto distributors than they can sell) or booking sales before they are truly final, making the company appear more successful than its actual cash flow supports.

Smoothing activity

This cluster looks for patterns where companies try to reduce the perceived volatility of their earnings or other financial metrics. This can involve using accounting discretion to make results appear more stable and predictable than they truly are.

Core question: "Is the company's financial performance naturally stable, or is it using accounting techniques to artificially smooth out fluctuations and hide underlying business volatility?"
The pattern recognized: The AI would spot a pattern of unusually low earnings volatility compared to industry peers, achieved through the strategic timing of discretionary expenses (like restructuring charges or asset write-downs) or the release of reserves to offset fluctuations in operating performance. This suggests an attempt to artificially smooth earnings.

Valuation signals

This cluster examines how a company's market valuation might be influencing its accounting practices, or conversely, how accounting manipulation might be aimed at achieving a desired market valuation.

Core question: "Is the company's accounting reflecting its true economic reality, or is it being influenced by pressures to meet market expectations or justify a high valuation?"
The pattern recognized: The AI would identify a pattern where accounting choices consistently lean towards aggressive interpretations that boost reported earnings or asset values, particularly when the company's market valuation is significantly higher than its fundamental performance would otherwise suggest. This indicates accounting decisions driven by market perception rather than pure economic reality.

Working capital signals

This cluster focuses on a company's short-term assets and liabilities (like inventory, accounts receivable, and accounts payable). Unusual patterns here can indicate manipulation to manage short-term liquidity or profitability.

Core question: "Is the company managing its short-term assets and liabilities efficiently and transparently, or are there signs of manipulation to hide liquidity issues or inflate short-term profits?"
The pattern recognized: The AI would detect a pattern of deteriorating working capital metrics, such as increasing days sales outstanding (customers taking longer to pay), increasing inventory days (inventory sitting longer), and increasing days payable outstanding (company taking longer to pay suppliers). This suggests underlying operational inefficiencies, liquidity pressures, or attempts to manage cash flow through aggressive working capital management.

The core of Transparently.ai’s methodology therefore is intelligent pattern recognition. The system acts as a specialized lens, grouping financial behaviors and anomalies into these 14 distinct risk clusters. This allows the AI to provide a holistic risk assessment by recognizing the full 'symptom cluster' of potential issues, ensuring a detailed, data-driven perspective on a company's financial integrity.

Mark Jolley

Mark has been an investment strategist for almost 40 years and has advised some of the world’s biggest investment funds, public companies and notable investors. In the mid-1990s, as a global investment strategist with Deutsche Bank, Mark produced a daily note read by more than 16,000 investment professionals including voting members of the Federal Reserve and the European Central Bank. In the 2000s, Mark worked as Deutsche Bank’s Asian strategist and then as strategist for China Construction Bank. He is now an independent analyst.