This article was written by Miles Grauberd – Investment Analyst at I Know First.

AI has Moved From Edge to Baseline

The funds with the longest records of beating the market run on models. What changed is the pace of adoption across the rest of the industry. J.P. Morgan’s 2025 institutional survey put active AI use among hedge funds at 46%, up from 18% a year earlier, with only 24% not engaged. A separate AIMA survey found that 95% of hedge funds use generative AI in some form, against 86% in 2023.

The choice is not whether to use AI. Peers use it. The gap runs between funds that built models into the investment process and funds that bought tools that sit idle while analysts work the old way. The first group covers a wider universe and catches shifts the model flags before they reach the price. The second pays for software and keeps its old output.

Family offices sit earlier on the curve. Most run on discretionary calls and outside managers, which leaves an opening: an office can add an institutional-grade AI forecast to its process without building a quant team or hiring data engineers.

The Models You Know are Now Trading

This is no longer a demo. In May 2026, Robinhood opened its brokerage to AI agents through the Model Context Protocol, the open standard that lets a model call outside tools and act. A user connects Claude, ChatGPT, or any compatible model to a dedicated account and lets it read positions, pull quotes, and place real orders on its own. By July, Robinhood pushed the feature toward 24/7 crypto, and its CEO told CNBC he expects agents to match a human trader. Traders run these agents on live accounts and post results as they happen.

The capability is real, and it is early. A hands-on review in July found the trading interface thin: equities only, few controls, and no sign of the sandboxes, kill switches, and position limits a real quant desk builds before it lets code touch capital. General-purpose models can place a trade. Whether they manage risk across a full cycle is the open question.

That gap is the point. Robinhood’s own CEO framed agentic trading as giving the everyday person the automated, AI-driven tools that institutional and high-frequency desks have run for decades. I Know First built one of those tools and has run it for institutions for more than a decade, and the track record below covers that history.

What I Know First is

I Know First runs a self-learning forecasting algorithm built on genetic programming and artificial neural networks. It models the market as a chaotic system and retrains on new price data every day. The system forecasts more than 13,500 assets across six horizons, from 3 days to 1 year, covering equities, sector and country ETFs, commodities, currencies, and bonds.

Each forecast carries two numbers. The signal gives direction and relative strength: positive points up, negative points down, larger magnitude means higher conviction, and the value ranks against the rest of the universe that day. The predictability runs 0 to 1 and scores how well the algorithm’s past forecasts on that asset matched the moves that followed. A desk reads the two together. High signal with high predictability is a position worth sizing. High signal with low predictability is one to hold off on.

Hedge funds, banks, family offices, and wealth managers run the forecast inside their own investment processes. Here is how they put it to work, followed by the performance.

What Hedge Funds and Family Offices Do With The Product

The product is the same for a hedge fund and a family office. Both run the forecast two ways.

Idea generation and screening. The forecast comes two ways. A client can send I Know First a list of the tickers it wants to track, and I Know First forecasts that list. Or the client can take the pre-made hedge fund and family office forecast, a single daily file that covers everything from equities to indices to currencies and more. That file surfaces I Know First’s top opportunities, so the desk starts from the strongest names instead of searching thousands of tickers for one. A client can run either, or both. The same forecast works the other way as a screen. When the PM builds a thesis, the signal and predictability give an independent read before execution: a match adds conviction and supports larger size, a conflict flags the idea for a second look. The screen carries no narrative bias, so it catches names a discretionary process talks itself into. I Know First opens this to hedge funds and family offices, with a revenue share based on performance, or a licensing fee.

Co-developed systematic strategies. I Know First builds and back-tests rules-based strategies on the forecasting indicators, then runs them with the client on a revenue-share basis. The signal ranks and selects the holdings, and the book runs on a fixed allocation: 60% to GICS Level 1 sector ETFs across at most 3 positions, 20% to individual S&P 500 stocks across at most 5 positions, 10% to GICS Level 2 industry ETFs, and 10% to SPY or S&P 100. I Know First rebalances monthly, on a roughly 4-week cycle. That structure targets positive alpha, so the return depends on the model rather than on market direction. The strategies span mean-reversion and trend-following logic depending on the mandate.

The common result: a small team covers a wider universe, and the client adds a quantitative input that runs every night without fatigue or bias.

The same tool flexes to the mandate. A hedge fund leans on it for alpha and faster turnover. A family office leans on the longer horizons, using the 1-month, 3-month, and 1-year signals to trim and add without the churn of a trading desk, and reads its whole book, equities, ETFs, commodities, currencies, and bonds, through one daily file instead of stitching together separate manager views. For a lean team either way, a daily forecast over a custom universe delivers an in-house quantitative view at a fraction of the cost of a full quant team, which tightens oversight of outside managers and cuts reliance on any single voice.

The Performance

The combined strategy has returned 756% since inception on January 29, 2020, a compound annual growth rate of 39.7%, against 129.1% for the S&P 500 over the same period.

The strategy beat the S&P 500 in every year on record, and it gained 15.36% in 2022 while the S&P fell 19.95%. Its drawdowns held below the benchmark in the market’s worst years, down 9.71% against 33.92% in 2020 and 8.80% against 18.90% in 2025, and over the full period it ran a 1.72 Sharpe and a 2.68 Sortino.

Three Recent Portfolios

The since-inception figure holds up in individual portfolios, and across different market conditions. Three windows from 2026 show the strategy adding return whether the market rose or fell, on both the long and short side.

The March book shows the short side at work. As the S&P fell 3.56%, the portfolio shorted its way to a positive 3.53%, a spread of 7.10 points and proof the model calls direction, not only which names rise. The April book shows upside capture: a long portfolio returned 8.12% against the S&P’s 4.56%, nearly double the benchmark in a rising market. The May to June book shows defense: the long portfolio held positive at 0.07% while the S&P lost 2.16%, keeping clients green through a down tape.

Together the three cover what an allocator cares about: make money when the market falls, beat it when it rises, and hold the line when it drifts down.