Reimagining investment research from the ground up.
Joined August 2024
- Tweets22
- Following44
- Followers640
- Likes50
Pinned Tweet
Every fund has a chatbot. Most investors tell us their process doesn't feel much faster. The time AI saves on search gets spent stitching everything else together across an already fragmented stack.
The mistake is treating AI as one more tool. We believe it's the first real opportunity to rethink the research process from scratch, and we've spent the last two and a half years building exactly that: one system for the entire workflow, run by an agent that knows your names, learns from live information, and works ahead of you.
There's a lot to build, and a race to build it. The first funds to get there lock in a multi-year lead. So what does it take? Not to deploy a chatbot, but to actually build an AI-native research platform? We just published our first post on exactly that. Link in the reply.
Implied retweeted
AI could do well at shorter-term alpha. Further out, more compute can’t solve every uncertainty (as long-range weather forecasts show). And if much of the uncertainty in markets comes from human behavior, you’d think at least some humans would still have an edge
Ken Griffin does not believe that artificial intelligence will ever replace investors
According to the Citadel CEO, the fundamental job of an investor is research, akin to being a journalist. Trading is simply how that research is monetized.
There are bits and pieces of the research process that artificial intelligence can automate, but the narrative still needs to be assembled by a human
Implied retweeted
At shorter time horizons, investing becomes more like poker. You’re playing the other players and their expectations, positioning and constraints as much as your own hand (the fundamentals)
An earnings beat can still trigger a selloff. Kelly Granat on whisper numbers, crowded positions and pod incentives.
Kelly Granat, CIO of Lone Pine, explains:
"I have never worked at a pod but have several friends who have, still do, and have come out of there."
"You have 20, 30, 40 desks covering a sector. They look at the same data, sit in the same meetings, on the same calls."
"You'll have 20, 25 books long a stock into a quarterly print."
"The numbers come out, and it was better, but it wasn’t better enough because the whisper number, the data was suggesting they were gonna beat by three points, and it only beat by two points, and so the stock’s actually gonna sell off."
"Hand me an earnings release and I'll tell you what the stock does was a fundamental game we were good at a decade ago. Now it's a coin toss."
"The opening reaction vs the full day: everyone's covering the short because it was bad, but it wasn't worse."
"They have drawdown limits — monthly, quarterly. Their capital gets pulled. That's the lives they lead."
"Market reactions to events are a function of those incentives and those structures."
AI can sound like a great analyst without knowing how to invest.
It can build models, forecast KPIs, and write convincing investment theses. But how much of that translates into actual investment judgment?
Take a seemingly simple task: given the actual earnings results, will the stock go up or down?
We removed the need to forecast the numbers by giving seven AI models the actual earnings results and asking whether the stock would go up or down. Across 26 earnings cases, the models achieved accuracy rates ranging from 49% to 69%. A simple strategy of always predicting “up” achieved 54%.
Even with the numbers in hand, the models struggled to interpret what they meant for the stock.
Our view is that what’s missing is the context and nuance behind investment judgment: which metrics matter for a particular company, what’s already priced in, and how the same results can trigger different reactions depending on the market.
We’re expanding this initial study this earnings season. We have a few ideas for closing that gap, and we’re eager to put them to the test.
AI modeling doesn't need a better agent. It needs a better spreadsheet.
Most tools ask an agent to write code that populates or edits Excel. As the workbook grows, so does everything the agent has to keep straight: numbers, cell locations, formulas, sources, and company-specific instructions. A longer prompt can't fix the wrong infrastructure.
So we built an AI-native spreadsheet from scratch.
In Implied Sheet, AI lives in the formulas. Instead of remembering and reproducing every number, the agent builds source-linked formulas that stay inside the model.
Analysts can inspect citations, edit the model directly, and export it to Excel with formulas intact.
This foundation opens up a much broader set of workflows. A few examples: agents can update models in parallel, incorporate new guidance into forecasts as an earnings call unfolds, and compare commentary with consensus across your coverage.
Here's a first look at building a model from scratch and updating it after earnings. More soon.
A new 200-page 10-Q comes in. You ask: “What changed vs. last quarter?”
Does the agent run out of context? If it breaks the filings into smaller pieces, does it lose context across them?
Even if everything fits, how do you know it caught every important change, and how do you verify it?
Giving an AI agent SEC filings is easy. Making a common investment workflow accurate, complete, and auditable takes a lot more than that.
Implied built custom tools to align filings by section and process text, tables, and images separately. We also built custom visuals where every citation opens directly into a redline view, so investors can see the takeaway and the underlying change together.
Building AI for investors is not just about asking an agent to read raw documents. It means building a system with high-quality processed data, custom tooling, and the right visual design so the analysis is complete and accurate, and every result is auditable.
A convincing AI answer isn’t proof that the research was done thoroughly.
Implied handles this by using our proprietary AI-native spreadsheet as an intermediate research step for complex, multi-step processes. It serves as both a research scratchpad for the agent and an auditable workspace for the investor.
This is just one example of what Implied’s AI-native foundation makes possible. We have more capabilities to show you soon. Stay tuned!
Implied retweeted
Earnings have become harder to predict despite more data is available. Can AI help with that?
Ken Griffin says short-term alpha from calling the quarter with alt data is dying, and the next era of stock-picking belongs to those with a long-term horizon.
"I think there'll be more focus and emphasis on those who have really good vision about what companies are actually creating transformative products that will change society. The market will reward that far more intensely in the future than will a company beat this quarter's earnings or not.
I think the question of will a company beat this quarter's earnings has gotten far more difficult over the last 10 years because, for example, the rise of alternative data. I have access to the credit cards of millions of Americans. What are they spending money on? What's that mean for Starbucks revenues this quarter? What's it mean for McDonald's this quarter?
This is a decade-old transformation, but the here and now is just becoming far more transparent, far more readily understood and triaged by the combination of really bright people and really good AI technology. Where this will leave us is those who are able to see what is unfolding over years to come will be in a very valued position on a relative basis."