FinSights AI

FinSights AI is different in three ways that matter.

• 

  Proven-superior data, not internet noise

  Every answer FinSights gives is grounded in New Constructs’ proprietary data. It’s the same forensic accounting work proven superior by The Journal of Financial Economics, Harvard Business School, and Ernst & Young. Our Robo-Analyst technology reads 200-page SEC filings and pulls the data that other firms miss. Then FinSights AI uses that data, and only that data, to answer your questions.

• 

  Built on Google Cloud, by Google Cloud

  FinSights AI runs on enterprise-grade Google Cloud infrastructure. It’s not a wrapper around a consumer chatbot. It’s purpose-built infrastructure for institutional-grade investing intelligence.

• 

  Designed for investing, not everything else

  Generic AI tries to do everything. FinSights AI does one thing: help you make better, faster, more informed investment decisions. That focus means the system is tuned for what investors actually need: reliable data, better metrics, fast comparisons, and answers that don’t require you to second-guess.

Ask anything about any stock.

From SEC filings to your screen.

• 1

  SEC FIlings.

  Public companies file 10-Ks & 10-Qs with hundreds of pages of disclosures. Most of the important information lives in footnotes and management discussion sections that traditional data providers ignore.

• 2

  Robo-Analyst processes everything.

  Our proprietary AI-driven machine learning technology reads every page of every filing, extracts data from the footnotes, and identifies adjustments that turn reported numbers into economic reality. This is the data that's been proven superior by The Journal of Financial Economics, Harvard Business School, and Ernst & Young.

• 3

  New Constructs analysts validate.

  Our analyst team reviews and validates the data before it goes into our models — ratings, reverse DCF, economic earnings, ROIC, EBV, GAP, every metric.

• 4

  FinSights AI delivers it in plain English.

  When you ask a question, FinSights AI draws on this proven-superior dataset, not the open internet, to give you an answer with sources cited. Every claim is traceable back to the underlying SEC filing.
  This is why FinSights AI can do something generic AI can't: tell you what's actually in the footnotes of a company's filing without making it up.

Built for investors at every level.

Get FinSights AI with the right membership.

FAQ

• How is FinSights AI different from other AI?

  Most general-purpose AI are trained on the open internet. That means scraped articles, blog posts, social media, and source data that was already wrong before the AI got to it. FinSights AI is grounded in New Constructs’ proprietary, proven-superior fundamental data — the same data validated by The Journal of Financial Economics, Harvard Business School, and Ernst & Young. Every answer is traceable to a specific SEC filing, not a confident guess.

• Will FinSights AI hallucinate or make things up?

  No system is perfect, but FinSights AI is designed specifically to avoid the most common AI failure: confident answers built on bad data. Every response is grounded in our proprietary dataset. If FinSights doesn’t have the data to answer a question, it will tell you, not invent something.

• What can I ask FinSights AI?

  Anything you’d ask a human analyst: explain a rating, compare two stocks, surface risks in your portfolio, screen the market in plain English, prep talking points for a client meeting. See the examples above.

• How do I get access to FinSights AI?

  FinSights AI is included or available with four New Constructs memberships: Discovery + FinSights AI, Portfolio Insights + FinSights AI, Professional, and Institutional. Pick the membership that fits your investing workflow and FinSights AI is part of it.

  If you are currently a member of New Constructs and want to add FinSights access, simply purchase one of the FinSights enhanced memberships. Your prior membership will be cancelled and you’ll be upgraded to the new membership with FinSights.

• How many questions can I ask per month?

  Depends on your membership. Allocations range from ~50 questions/month (Discovery + FinSights AI) to ~300 questions/month (Institutional). See the pricing table above for full details.

• Can FinSights AI make trades or give specific buy/sell recommendations?

  FinSights AI provides research, analysis, and rating information. It doesn’t execute trades or provide personalized investment advice. New Constructs’ models and ratings are research products, not investment recommendations, and FinSights AI surfaces and explains that research in conversation form.

• The typical question uses ~55,000 tokens. Is that input or output tokens?

  The ~55,000 tokens per typical question is input + output combined.

• Can users see a token-cost estimate before submitting a question?

  There’s no way to estimate token cost before submitting your question. The only reliable way to know is to ask, at which point the tokens are consumed.

• Can users track cumulative usage in real-time during the month?

  Yes, you can track cumulative usage on the FinSights Usage page, which updates after each question.

• Do follow-up questions in the same conversation thread consume fewer tokens due to shared context, or does each submit consume tokens independently?

  Each question consumes tokens independently. FinSights maintains awareness of your prior conversation context, so follow-up and clarifying questions work as you’d expect.  They use approximately the same number of tokens.

   

• What happens when the monthly token cap is met or exceeded?

  If a question would exceed your remaining token balance, it fails immediately. You’ll receive a message from the agent explaining that your included tokens are exhausted and directing you to the Usage page to enable additional token usage if you would like to continue to interact with FinSights.

• Do unused tokens at month-end roll over to the next billing cycle?

  Unused tokens do not roll over. They reset at the start of each month.

• How can I optimize my questions to reduce token costs?

  A significant portion of tokens used per query come from the agent’s system prompt, which runs behind the scenes and isn’t something users can modify or reduce. So while prompt structure does have some impact on total consumption, the ceiling on optimization is lower than it might appear.

  That said, there are marginal gains available on the output side: narrowly scoped questions require less synthesis, requesting a specific format (“table only, no explanation”) shortens responses, and breaking multi-part questions into focused follow-ups can reduce per-call output. These won’t dramatically change your token economics, but they’re the primary levers available to you.

• What is the time lag between an SEC filing and its availability in FinSights?

  SEC filings are retrieved overnight the day they are filed with the SEC and processed the following day. The resulting data is available on FinSights before market open on the next day. Data availability is the same as on our website.

• If a query fails mid-execution are the consumed tokens refunded?

  Tokens consumed during a failed query are still charged to your account. There is no refund or credit for mid-execution failures.

  FinSights in built on and by Google Cloud on the Gemini platform. Google regularly makes updates to Gemini that can sometimes temporarily break certain features of FinSights and other agents on the platform. For example, charting features may sometimes be temporarily unavailable because of changes on the underlying Gemini platform.

• Is there a way to export raw data directly from the FinSights as a CSV or JSON?

  There is currently no mechanism to export raw data (JSON/CSV) directly from the FinSights. CSV download support is something we’re actively considering for a future release.

• What should I do if FinSights stops answering a question it was able to answer before?

  If FinSights is unable to answer a question it should be able to handle, try starting a new conversation and asking again. Prior context in an ongoing conversation can occasionally confuse the agent, and a fresh session often resolves the issue.

• Does Google have access to the questions I ask on FinSights?

  FinSights data is hosted on Google Cloud Platform, which provides enterprise-grade encryption in transit and at rest. Access to stored prompt and conversation data is restricted to select members of our engineering team for purposes of software development and auditing.

  FinSights runs entirely within New Constructs’ own GCP instance. Google operates as an infrastructure provider under a standard enterprise Data Processing Agreement and does not have rights to use or retain your prompt content. Your data does not leave New Constructs’ controlled environment, and no other third parties receive prompt content.

• Does FinSights have access to my portfolio or account data from New Constructs?

  FinSights does not have access to your account or portfolio data. If you want to ask about positions in your portfolio, you would need to provide that information directly in your prompt.