How many stocks does Bedrock AI cover?
Universe: We cover over 9,000 U.S. and foreign issuers. We cover all EDGAR issuers that have filed a 10-K, 10-Q, 8-K, S-1 or 20-F since 2019. We do not cover Asset-Backed Securities (SIC 6189).
Filing coverage: Our coverage includes S-1s (prospectus precursor), 10-Ks, 10-Qs, 20-Fs (annual and quarterly reports), 8-Ks (SEC-mandated news releases), NT forms (non-timely filing notifications) and SEC comment letters (UPLOAD). We’re always adding coverage so check back for updates.
How quickly does Bedrock AI process filings?
We process SEC filings in near real-time. To see new filings and red flags, hit refresh on your browser. During peak filing times processing times may be longer (up to 5 hours).
Is there human review in the Bedrock process?
No, there is no human intervention in our real-time process. Red flags, severity scores, and risk scoring is all computer generated.
How long is the Bedrock AI trial?
Trial either Bedrock Basic or Bedrock Premium for 10 days for free. To trial Bedrock Premium, you must first book a demo here. Bedrock Premium trials are only available to institutional investors and enterprise customers.
How much does Bedrock AI cost?
Bedrock AI trials are free. Sign up here. Bedrock Basic costs $44/month charged annually. Sign up for a demo to learn about Bedrock Premium pricing. We offer discounts to MBA students, academics, journalists and funds in their first six months of operation.
How does Bedrock AI detect red flags?
Bedrock AI uses language models to find content that has empiric associations with fraud and related outcomes. Learn more about our approach to financial artificial intelligence in our post - “Financial NLP & LLMs - The Bedrock AI Advantage”
What types of red flags are detected by Bedrock AI’s algorithms?
Bedrock AI detects any and all text in securities filings that has an association with fraud, earnings management or malfeasance, regardless of the specific wording used. That means we’re extracting generally any information that would be highlighted by a forensic equity analyst.
Popular red flags types include accounting policy changes and restatements, management turnover, related party transactions, internal control issues and much more. Red flags are categorized/tagged after extraction, as part of a separate process.
Here is the complete list of Bedrock AI’s red flag categories.
How is the Bedrock AI risk score calculated and what does it measure?
Bedrock AI risk scores are machine learned, not additive, and assess the likelihood of high impact downside outcomes related to earnings quality and fraud. Risk scores range from 1 to 100 (low risk to high risk).
Scores above 70 are considered High Risk. Fewer than 10% of mid/large-cap issuers receive a Bedrock AI Risk Score above 70. Companies with a high score have about a 1 in 3 chance of receiving an SEC enforcement action in the next three years. Scores from 60 to 70 are considered medium-high.
Risk scores are optimized for precision over recall. This means that high risk companies are very likely to experience future issues but low risk companies may not all be in fact low risk. For a comprehensive view of risk, refer to the Bedrock Summary View on any company page.
Initial high risk scores tend to precede price collapse by at least 6 months.
Risk scores are trained using 10+ years of SEC enforcement actions, regulatory investigations related to fraud, settled class action lawsuits and other labels related to the earnings management and malfeasance.
Our risk scores ONLY use narrative information (Bedrock red flags) in making predictions.
Why doesn’t this red flag make sense to me?
Bedrock AI covers hundreds of types of red flags/risk areas, some of them are somewhat niche. If you’re subscribed to Bedrock Premium II or III, you have direct access to our in-house forensic experts. Reach out and ask. Or you can find our red flag guides here: Red Flag Guide Part I, Red Flag Guide II.
Language models are great at many things but they have a few failure modes. (Artificial intelligence isn’t magic.) Expect about five percent of red flags to be false positives. Most of these will be in the “yellow flag” section. Lower confidence flags can be filtered out using in-platform filtering. 20-Fs and S-1s have slightly higher false positive rates compared to other filing types.
Possible failure modes:
Materiality: Our models are good at understanding text but bad at math. This means they sometimes mistake a $1 transaction as almost as important as a $1B transaction.
Negation: While really simple for humans, negation is much more complicated for artificial intelligence. It’s strange but true given that our models have a strong understanding of other very subtle aspects of human language .
Why doesn’t Bedrock AI incorporate financial ratios into the risk assessment process?
One of the reasons that auditors are bad at detecting fraud is that it's hard to reliably identify through ratio analysis. Financial reporting is manipulated to look normal in many fraud or fraud-like situations. When ratios start to unwind, it’s often already too late.
Predictions based on qualitative risk factors (related party transactions, aggressive/frequent accounting policy changes, poor governance, off-balance sheet risk etc.) have much higher accuracy rates AND are much earlier signals, compared to traditional forensic financial ratio analysis.
What’s included in Bedrock Basic vs. Bedrock Premium?
Refer to our Product page for details.
How do I get started?
Sign up here for a free 10 day trial and never miss a red flag again.
Can students and journalists access Bedrock?
Contact us at info@bedrock-ai.com to find out if you qualify for our program for students and journalists.
I’m already a subscriber, where do I login?
Login to our market intelligence platform at dash.bedrock-ai.com/login
Why does this risky company have a lower Bedrock AI risk score?
Our risk scores measure earnings quality and fraud risk, not generalized risk. Other types of risk are not incorporated into our scoring. Refer to the Bedrock Summary View for a more comprehensive view of risk in the platform.
Note that we also optimize for precision not recall when we train our models. This means that missing a few risky companies is inevitable. However, even when the risk score is low, we will identify relevant and informative red flags. Here are a few examples of red flags detected at low risk companies that helped us predict earnings outcomes- “Accounting policy changes boost tech earnings”.