Frequently Asked Questions

Our Transparently Risk Ratings are an A+ to F scale that gives an indication of the overall accounting health of a company. These ratings are derived from the Transparently Risk Scores, and how those lie relative to the sector and market.

Yes, the system ingests notes in the statements and auditors reports.

Our system uses data that is standardised and translated by Refinitiv, our data provider. In this way, the system uses a uniformly formatted set of data.

No, it doesn't. This is deliberate since the system's prediction of corporate failure is related to the collapse of the share price. Taking into account price movements would overfit the models.

At the moment, no you cannot. Our long-term goal is to build models that will be data-agnostic and will be able to ingest accounting data held in private databases.

We are embarking on a project to address this need with privacy-protected federating computing, sometimes referred to as federated learning. The idea is that a key service provider would encrypt the hard drive and memory; our customers would encrypt their private data and we would encrypt our algorithm.

All this happens on a cloud with services and security fully certified. The algorithm will run on the private data and return results to the owner without data or algo being compromised.

There are 3 types of data that we process:

1. The annual financial statements sourced from Refinitiv’s Worldscope Fundamentals. We chose Refinitiv as this is the “go-to” gold standard for many academics. We ingest the financial statements and also the notes to the accounts, including the audit opinion report (textual analysis).
2. Incentives to manipulate: Anything that can cause a company to be under stress, for example, compensation structure, performance pressure, moral hazard as a result of agency theory. There is a vast body of academic research dedicated to this and we have created specific models to account for such stresses.
3. Statistical patterns observed by the AI: This is arguably the most critical process since it aims to identify patterns that would otherwise be unobservable by human effort. Its output is “black box” in nature but by combining AI/ML results with those from classical modelling, we are able to offer users meaningful insights as to how and where manipulation occurs. 

We cover almost all stocks that have ever been listed globally in the past 35 years. There are approximately 85,000 stocks in our system currently.

However, there are exceptions:

1. Banks and insurers
2. Companies without a minimum of 3 years worth of data

For these companies, you will see notifications on the Transparently dashboard alerting you to limited data availability.

The system was explicitly designed to minimise false positives.

For example, when we compare in-sample (training data) vs out-of-sample (test data) results we see extremely similar results and an extremely low rate of false positives, In addition, as the risk score increases, the level of certainty increases and consequently the rate of false positives trends towards zero.

In a nutshell, the system can rarely miss manipulation, but when manipulation is identified it is extremely likely it is there.

For specifics, contact us for our whitepaper. 

The dashboard uses an array of approximately 200 financials models, combined with machine-learning processes, to sift through millions of publicly available corporate financial data points and identify accounting red flags.

There are 3 broad types of models that we use:

1. Models for explicit measures of manipulation. These are informed by techniques used by forensic accountants, activist short-sellers and banks for fraud detection
2. Models that track incentives for companies to manipulate their accounts, for example, firm stress.
3. Pure statistical models, which do not relate to any particular part of the financial accounts. They are instead looking for anomalous patterns within the accounts that have historically been associated with higher manipulation risk and so can be challenging to explain.

We used point-in-time data for testing and our model performs with a high level of comparative success for all percentage risk scores for both the training data set (in-sample data that tracks known actual outcomes) and test data (out-of-sample data, which forecasts manipulation).

There are 3 observations of note:

1. There is a high correlation between the two data sets.
2. A monotonic relationship, meaning the variables tend to move in the same relative direction, but not necessarily at a constant rate. We therefore observe that the higher the manipulation risk score, then the greater the probability of prediction success.
3. Our test data shows the risk assessment is conservative. This is important to avoid the risk of false positives.

For specifics, contact us for our whitepaper. 

Pay especially careful attention to factors that have been highlighted as Statistical Anomalies. These are really powerful models where the machine has learnt to spot patterns of manipulation that differ wildly from known explicit measures of manipulation.

For context, there are 3 broad types of models that we use:

1. Models for explicit measures of manipulation, which are informed by techniques used by forensic accountants, activist short-sellers and banks for fraud detection.
2. Models that track incentives for companies to manipulate their accounts, for example, firm stress.
3. Pure statistical models.

These statistical models do not relate to any particular part of the financial accounts. They are instead looking for anomalous patterns within the accounts that have historically been associated with higher manipulation risk, and so can be challenging to explain.

There are 10 different statistical models that we employ, so where you get this flagged as a contributing factor, please pay particular attention to it.

You may be familiar with more traditional multi-factor models with defined thresholds; risks are flagged if values are above or below those thresholds, and enough flagged risks result in some action (e.g. a recommendation).

Our process does not operate this way.

The dashboard uses an array of approximately 200 financials models, combined with machine-learning processes, to sift through millions of publicly available corporate financial data points and identify accounting red flags. Each day, we refresh the universe of stocks, triggering a re-estimation of our models, and the AI dynamically determines weights.

The risk score that is produced is dynamically constructed, not linear or static. The system is constantly learning and evolving. There are no fixed threshold levels for factors that warrant concern. Instead, the system identifies the unique combination of characteristics for the selected company and financial year that is driving its Transparently risk score.

Importantly, this combination is generated from a much larger array of possible factors that could appear in the report. This means that a high risk factor in, for example, “Gearing” cannot be considered in isolation - it is the pattern of gearing risk factors in combination with all the other signals that the system is highlighting for that company. This means, high or low gearing (or no gearing signal at all) could possibly result in a high risk signal, depending on the results of all other factors.

We cannot stress enough the importance of reading the full report generated by the system in a holistic manner. The combination of highlighted features is key, rather than focusing on individual factors.

Accounting errors are likely to happen in isolation. Our system will flag an unusual pattern, but unless the company is engaged in active manipulation elsewhere in the accounts, that error will not be sufficient to generate a high manipulation risk signal.

A combination of features is required to generate high manipulation risk signals.

No, our system goes far beyond the considerations of credit default models. This is an accounting manipulation model. We are automating the analysis of forensic accountants, auditors, equity analysts, credit analysts, activist short sellers, and many others.

Hence, credit risk plays a role in the process, but represents only one out of 14 clusters of characteristics the system evaluates. We do incorporate some aspects of credit default models given credit risk has an impact on managements' incentives to manipulate their accounts. 

At the moment, our models do not account for connections between companies and related entities unless that information is already disclosed in a company's financial statements.

We're working on incorporating datasets that will enable us to identify and consider the impact of these kind of interdependencies. By interdependencies, we refer to relationships such as offshore shell companies, cross shareholdings or interconnectedness with directorships and family members.

The system has determined that a company's market capitalisation influences the quality of its accounting. Smaller companies tend to have lower quality accounts relative to larger companies.

This is dynamically accounted for in the presentation of results. Hence, the full forensic report our system provides for each company and each financial year will highlight when the size of a company has had a material impact on the estimated manipulation risk score (in combination with all other highlighted factors for that company).

In addition, the distributional analysis within the user interface automatically adjusts for this phenomenon and provides results that are agnostic with respect to size. It is therefore possible to see size-adjusted results while also being alerted to the presence of size-related risk.

Yes, they do. Variations in accounting standards are important to account for given they can have a material impact on results.

Yes, the models take into account peers in the market for benchmarking. They also account for specific features of individual markets, countries and groups of these.

Yes, each company is analyzed using both models specific and non-specific to its sector.

Corporate collapse considers severely adverse events such as debt default, delisting on negative terms, and/or substantial decreases in stock price. We have different models to evaluate thresholds for price collapses. The publicly available version of the system employs a threshold fall of 80% or more (in both absolute terms and relative to a benchmark index).

The system uses a dynamic weighting model. All our analytics are effectively relative. When a target company for a selected stock year is analysed, its performance is assessed within its country, sector, historically through time, etc.

The focus is on determining what is "normal" and then looking for anomalies. In addition, all factor models are determined simultaneously. That means no single factor should be considered in isolation. 

A factor is being highlighted as important because the system has specifically selected it as key for the selected company, the selected financial year and in combination with all other highlighted factors.

Consequently, this is a dynamic model. Unlike traditional non-ML/AI techniques that employ, say, weighted factors and those factors are always considered to work in a particular way, this system determines the combination of features and the direction of their impact in a completely dynamic form.

That means, for example, a high value of a certain factor may indicate higher accounting manipulation for a stock, while for a different stock, it may be that a low value of that same factor is associated with higher manipulation risk.

The system is able to evaluate patterns of factor interactions that are much more complex than any human could process (or could possibly imagine).

We get a daily feed from our data providers, and we refresh data daily. This triggers a daily re-estimation of all our models.

Yes, our model weightings are dynamic, not linear or static. Each day, our universe of stocks is refreshed, our models are re-estimated, and the AI dynamically determines weights.

All analytics are effectively relative. When a target company for a selected stock year is analysed, its performance is assessed within its country, sector, historically through time, etc.

The focus is on determining what is "normal" and then looking for anomalies. In addition, all factor models are determined simultaneously. That means no single factor should be considered in isolation.

A factor is being highlighted as important because the system has specifically selected it as key for the selected company, the selected financial year and in combination with all other highlighted factors.

Consequently, this is a dynamic model. Unlike traditional non-ML/AI techniques that employ, say, weighted factors and those factors are always considered to work in a particular way, this system determines the combination of features and the direction of their impact in a completely dynamic form.

That means, for example, a high value of a certain factor may indicate higher accounting manipulation for a stock, while for a different stock, it may be that a low value of that same factor is associated with higher manipulation risk.

The system is able to evaluate patterns of factor interactions that are much more complex than any human could process (or could possibly imagine). 

Yes. You can export search results by selecting the green export icon in the bottom left corner of the dashboard.

Select the filters you require and then choose “Apply Filter” for the results.

Store the filter selection for future use by clicking on “Save Filters”. You will be prompted to choose a name for filters that you save.

You can remove the filters by selecting “Clear." 

Filters include:

• Risk Score
• Sector
• Industry Group
• Auditor
• Domicile
• Exchange
• Country of Incorporation
• Status (Active, Delisted, and Suspended stock)
• 1 Year Change
• Market Cap Decile (Each decile represents 10% of the available stock universe sorted by size each year)
• Year

You can upload a watchlist by clicking on the "Upload Watchlist" button. You will be able to select a CSV file from your device. The CSV file should have a column of ISIN codes with the first row containing the text "isin".

Alternatively, you can create a watchlist by selecting stocks within the dashboard. To add a stock to a watchlist, click on the star located in the top right corner of the screen. You will be prompted to create a new watchlist or add to an existing one.

To view a watchlist, select the radio button of the watchlist you want to see and click on the "Load List" button.

To delete a watchlist, simply select the list and click on the trashcan icon. To export, select the list and click on the corresponding icon.

The search bar allows you to search stocks by typing their name, Bloomberg tickers or ISIN codes. As best practice, we would advise to start with ISIN codes.

Risk cluster and factor model scores of 1-5 represent quintiles, with 1 being the best and 5 the lowest score for a particular characteristic.

In the full report, we provide a list of all risk levels for the clusters and then select up to 3 of the most critical factor risk scores within each cluster that has a score of 3-5. Hence, this is why we only show the factors that are having the biggest impact on the overall manipulation risk score.

The system is dynamic, and its analysis changes as new data is ingested. This is due to the use of AI/ML processes which are constantly learning from every new form of manipulation identified, and then applying that learning to every other company in real time.

So a company's risk assessment and its risk score could change even if it has not announced new accounting results. This would be because the system has learned something new from another company which has announced, and immediately applied that learning to every other company globally.

The text is auto-written by the system and it is conditional upon the magnitude of the risk scores. The text is not static. The report provides risk attribution for the top 3 contributing factors for each risk cluster, ie. the worst signals, for any rating worse than “Investigate Further” (ie. High Caution and Extreme Care).

When we looked at a basket of stocks with manipulation risk scores, more than three-quarters of these exhibited some form of collapse within approximately 3 years. This number increases substantially as the estimated risk score increases.

No, we don't record the information you download. The only activity we record is that reports have, in fact, been downloaded, but we don't record which ones.

No, we can't. The system has been created so that filters and watchlists are saved on your device, and not recorded in your account.

Please note: This means that if you change devices, you will need to create those again. This is done to protect your privacy.

No, the system has been created in a way that anything you search for is not recorded.

We absolutely do not sell your information or any of your data. We have designed our system in a way that Transparently.ai has no visibility on your search history on our tool, the stocks you watchlist, upload or download.

We do limit the number of stocks you can download information on to 3,000 over a consecutive 3-month period.

You can read more about our Privacy Policy.

The user manual contains valuable information about the Transparently.ai tool. Find out what to do if you forget your password, how to use the system, its filters and watchlists and other useful information.

Download the user manual here.

If you need help with anything, search our Academy or extensive FAQs for the relevant article.

If you can't find the answer, email support@transparently.ai.

The dashboard is one of the three avenues through which customers can access the Transparently Risk Engine's predictive analytics.

Through the dashboard, customers can utlise the TRE to assess to what extent a company is manipulating its accounting and the likelihood that it may collapse as a consequence.

The engine uses an array of 200 models which replicate the practices typically utilized by forensic accountants, short-sellers, equity analysts, credit analysts, and academics. We use the latest AI techniques to train our models on decades' worth of data across 85,000 companies, detecting complex patterns of manipulation.

The engine then producess Risk Scores and Ratings and provides a detailed explanation of detected red flags, consolidating its analysis across 14 clusters of characteristics to guide analysis.

All of these are displayed on our dashboard. Through the dashboard,  you can also access our full risk reports and Luca, our GenAI assistant.

The risk score is a 0-100% score produced by the Transparently Risk Engine (TRE).

The TRE uses around 200 models, combined with AI machine-learning processes, to analyze the signs and symptoms of accounting and business manipulation patterns.It then provides a risk score for each company based on 14 distinct clusters of accounting characteristics. The score indicates:

• the probability of a company engaging in accounting manipulation
• the extent of the manipulation
• possibility of negative consequences such as debt default, delisting, or significant decreases in stock price (80% or more)

At Transparently.ai, we aim to help companies measure accounting quality. Our product is specifically aimed at helping solve the challenge of accounting manipulation and fraud in corporate financial statements.

The losses incurred as a result of accounting malfeasance by corporations are staggering. More than USD$1.5 trillion in market value is lost to Investors due to accounting manipulation and fraud for listed companies alone. Unlisted companies will be a multiple of that.

Other sources estimate the losses could amount to US$4.7 trillion, with research estimating it's US$830 billion in the US alone. That same research suggested that only one-third of corporate fraud is detected and estimated that 40% of companies commit accounting violations and, on average, 1 in 10 of large publicly traded companies commit securities fraud yearly.

Sources: Munter,P. (Oct. 2022), The auditor's responsibility for fraud detection; Dyck, A. et al (Jan. 2023), How pervasive is corporate fraud?

Complete this form on our website to request a live demo of Transparently.

Yes. All prices quoted on our website and sales materials are before tax (GST, VAT). Your taxes will be clearly displayed in the invoices.

Yes, we offer discounts for annual billing and for more seats added on top of default seats of the plan.

Contact our sales team for more information.

FastSpring LLC, our authorised payment services provider will send you a reminder 5 days before the credit card on your account is due to expire and on the expiration day. The reminder will contain a link for you to enter new payment details.

Navigate to your Account, Settings and click Edit.

You can add more seats or new markets to your current plan.  When you do, you are billed on a pro-rated basis for the rest of the current month for the price difference, including any applicable discounts.

The new add-ons inherit the same billing frequency (e.g monthly) as your original subscription. In the next billing cycle, you will pay the full subscription, less any applicable discounts.

If you want to change your current plan between Basic, Standard or Advanced, please contact us on sales@transparently.ai.

Like many software subscription services, you pay in advance rather than in arrears. You can pay by credit card, or by bank transfer. In both situations, an invoice will be provided to you when you agree to sign up to our services.

Yes, you can add new seats to your current plan at any time.

We will pro-rate the cost of additional seat(s) to your plan and bill you for the remaining month when you add the new seat(s).

At the start of the following month, your invoice will reflect the total number of seats purchased on the plan.

To add a new seat, the named administrator on your plan account should:

1. Navigate to Settings and Team/Users
2. Select "Add new seat" and follow the prompts
3. Enter the number of new seats you require, along with the email addresses and mobile phone numbers of the users you wish to invite
4. Complete the purchase of the seat(s) and notify your new users. 
   
   If you are an Enterprise User, you can contact your Account Manager to add seats to your plan at any time. 

If you are business and have a valid Value-added-tax (VAT) ID registered in the country you are buying from, you can add your VAT ID at the Checkout page before you complete your purchase. By completing this, the applicable sales tax will be deducted from your final price.

We do not offer refunds. We encourage you to test the functionalities of our tools with a 14-day trial. After this period ends, you will not be automatically enrolled for any paid subscriptions. Please contact our sales team for more information about a trial.

You may cancel your subscription. By cancelling your subscription, you will not be charged future payments, and you may use the tool until the end of the subscription period (either month or annual).

We do not offer refunds. We encourage you to test the functionalities of our tools with a 14-day trial. After this period ends, you will not be automatically enrolled for any paid subscriptions. Please contact our sales team for more information about a trial.

Your invoices are sent from FastSpring LLC, which is Transparently.ai's authorised payment services provider. FastSpring acts as our Merchant of Record. Fast Spring will issue your invoice with the relevant sales taxes that are applicable to you.

When you upgrade your access or purchase add-ons, you are billed on a pro-rated basis for the rest of the current month for the price difference, including any applicable discounts.

The new upgrade or add-ons inherit the same billing frequency (monthly) as your original subscription. In the next billing cycle, you will pay the full subscription, less any applicable discounts.

You are billed monthly in advanced from the time you activate your paid subscription or after your free trial ends.

Discounts apply for Annual Subscriptions, which you can pay annually, at the start of your subscription service or monthly. If you pay an annual up-front fee, you are billed annually from the time you activate your subscription.

If you pay a monthly fee on your Annual Subscription, you will be billed monthly and you will not be able to terminate your subscription until the first anniversary of your service, ie. after 12 months.

FastSpring LLC is Transparently.ai's authorised payment services provider. FastSpring acts as our Merchant of Record and will issue your invoice with the relevant sales taxes that are applicable to you.