Let’s be real- we’ve all seen the post-exit victory lap splashed across our LinkedIn feeds. It’s the one claiming, “I knew five years ago exactly how this company would break out.” Or we’ve sat in the internal post-mortem where someone insists every red flag was visible about a certain company from day one.
But none of it was obvious. Hindsight smooths the chaos, rewrites the initial conditions, and convinces us the path was linear. In reality, venture decisions are made amidst conflicting signals and market shifts. We are forced to call the weather long before the sky offers any clarity.
I learned this truth long before I ever invested in a company. In my early career as a weather forecaster (always a great party anecdote!), I went through nearly two years of training, academic meteorology combined with military‑grade forecasting sessions. It was rigorous, technical, and repeatedly forced us to confront a fundamental question: how do you provide a definitive prediction in an inherently uncertain system, especially when the stakes are high?
Similarly, investing in companies at the dawn of the AI era feels incredibly chaotic. Technology cycles that once unfolded over a decade now compress into months. Competitive moats appear and disappear with disorienting speed. Capital flows remain volatile, and every week brings a development that forces us to recalibrate what we thought we knew. In this environment, feeling true conviction when backing a company has never been harder.
The parallels between meteorology and venture investing have always been there for me: both attempt to predict outcomes in chaotic, complex systems with fundamental limits on predictability- weather because of countless atmospheric variables, and venture because of the inherently unpredictable nature of people and markets. In both fields, decision-making means analyzing incomplete data, building mental models from disparate signals, and making high-stakes calls under pressure. And in the chaotic AI era, those parallels are even clearer.
As we continue to bet on the teams and companies that we believe will shape the future, the science of weather forecasting offers venture capital something it desperately needs: a tested framework for making decisions when the environment is more chaotic than ever.This will never mean a perfect ability to forecast the success of a company (and if you’ve figured that out, I want you working with me!), but over decades, meteorologists have developed principles for operating effectively amid uncertainty. Those principles are surprisingly transferable to venture investing- and in today’s rapidly shifting tech “climate,” perhaps more valuable than ever.
The Prediction Paradox: The U-Shaped Curve
Here’s the counterintuitive reality of atmospheric forecasting: accuracy doesn’t degrade in a straight line over time.
Short-term weather forecasts are remarkably accurate, mainly because “initial conditions” for T=0 like wind, pressure, and temperature are observable by real time measures (like satellites and radiosondes). 1-3 day forecasts can hit 95% accuracy. 5 day forecasts are around 90%.But by day 10, forecasts are barely better than chance. The atmosphere is a chaotic system where small errors in initial conditions compound exponentially- this is the butterfly effect that makes predicting next Tuesday’s weather from two weeks out essentially impossible.
But here’s the paradox: while ten-day forecasts plummet, ten-year climate projections are surprisingly better.
Climate models built in the 1970s correctly projectedthe warming trends we see today. This is because while weather tries to predict specifics (will it rain at 2:00 PM?), climate models slow-moving structural forces: ocean temperatures, atmospheric composition, and energy balances. The result is a U-shaped accuracy curve, not a linear one:
- Days 1–5:High accuracy
- Days 6–30:Accuracy collapses as chaos dominates
- Months 3–6:Partial recovery as large-scale patterns emerge
- Years to decades:High confidence in structural trends
The Venture Application: A Framework for Decision Making
This U-shaped curve can be a great lens to use when analyzing companies and investment opportunities.
The Execution Layer, or The “Weather” (0–12 Months):
When meeting a company and diligencing its team, product, traction, and market size, there are observable, real-time signals that we can track that can help us predict the company’s ability to execute in the short term. [1]
We can forecast a startup’s “initial conditions” with potential high reliability. We look for metrics like management quality, hiring quality, customers feedback, product velocity and financial metrics. These data points help us predict:
- Near term management execution and discipline
- Ability to expand within existing customer base and acquire new customers at the ICP
- Ability to compete/ win against direct competition
- Near term hiring quality and organizational health
- Projected Churn
- Capital efficiency and burn trajectory
- Qualified buyers and assigned budget lines
But at some point, initial data points are no longer useful in the quest of venture prediction. it’s harder to predict timing of broader market education, how the competitive dynamics will shift with new entrants and incumbent moves, what features customers will love, how platforms will shift, the next model capabilities leaps and so on.
This is the chaos zone. Yet this is where investors often focus- detailed five-year projections, precise market-share assumptions, confident calls on where value will accrue and how competition will look like mid term. Especially in the AI era of building companies, this medium-term horizon is completely chaotic.
What you cannot predict reliably, or The “Chaos Zone” (1–3 years):
- Timing of market education inflection points
- Mid term revenue milestones
- Competitive landscape dynamics (to an extent)
- Which AI capabilities will commoditize faster
Eventually, we are all striving to identify if this company has what it takes to win long term.
When meeting a company at T=0, we can look for some signals that help us decide if this a company that potentially has what it takes to win the market. These are often broader thesis observations- like adoption of AI by a given industry, a demand for a specific set of technologies, or even M&A trends at an industry level.
What regains predictability at longer horizons, or The “Climate (3-10 years)-
- Sustained growth in demand for a specific value associated with the company’s product
- Cost of said technologies going down
- M&A trends and convergence of platforms
- Acceleration of AI replacing labor tasks
- Emergence of regulatory frameworks around AI
And while there is obviously a lot of room for things to change like management to take a different direction or products that lose its value over time- looking for long term indicators can often help us up zoom out and see where the “atmospheric streams” might eventually lead us.
The venture forecasting framework in a nutshell, helps us answer these fundamentals:
- Anchor on near-term execution:Can this team ship, learn, and adapt?
- Align with long-term structural forces:Will this company win in the long term?
- Discount medium-term certainty:are we treating detailed 3–5 year projections as a major piece in our thesis?
Again, this is not an effort to predict the outcome of a specific investment- but a suggestion to use a different perspective when we meet companies and decide if we believe the future that they offer.
From Single-Forecast Thinking to Ensemble Thinking
Weather forecasting ran into a hard limit decades ago. Even with better tools to measure weather indicators and more real time data, forecasts failed in the same way: the chaotic nature of the weather made the tiny shifts in current data grow into massive forecast errors over time.
To deal with this, the science of meteorology introduced a new method called Ensemble thinking: instead of running one simulation, forecasters run dozens – sometimes 50 or more -simultaneously, deliberately introducing slight variations in the starting assumptions (initial conditions) for each one.
The value isn’t in any single outcome, but in the convergence of the group:
- The Probability:If 40 out of 50 simulations predict snow, we report an 80% confidence level.
- The Spread:When the simulation lines on a map stay close together, we have high confidence. When they scatter in every direction- creating what we call a “Spaghetti Plot”- it’s a signal of high uncertainty and low predictability.
The Venture Application: Moving Beyond the Single Narrative
In my view, many investment decisions around VC tables still rely on “Single-Forecast” narratives: one specific market size, one competitive outcome, and one linear path to scale. While this feels like conviction, it is often a delusion; as costs, competitive dynamics, and platform shifts evolve at light speed, rendering a single-path thesis especially dangerous.
Applying Ensemble Thinkingto venture capital means shifting our focus from “The Path” to “The Range”:
- Articulate Multiple Scenarios:Instead of one narrative, identify 3–4 credible ways the market and the company could develop and still win (e.g., focusing on a specific vertical / ICP, relying on a specific asset unique to the company, or articulate a healthy range of potential financial outcomes).
- Quantify the Spread:Stress-test your conviction against less favorable scenarios. If the company only succeeds in one possible scenario of the “simulations”, (e.g, your only reliable outcome is the company being acquired by a narrow set of acquirers)- the spread is too wide.
In venture, as in weather, the goal isn’t to be “right” about one specific future- it’s to ensure your bet is robust enough to survive the chaos of many possible futures.
It’s a Data Game, until it Ain’t
Modern weather models ingest new data from multiple sources including satellites, weather radiosondes and radars continuously and update forecasts accordingly. When data changes, the forecast changes. There is no attachment to yesterday’s view.
Investing often works differently. Thesis form at entry, and update quite slowly if we are being honest, as confirmation bias creeps in. In AI, this lag is costly. Capabilities evolve frequently, and competitive dynamics shift quickly.
A data-assimilation mindset in venture means:
- Reviewing key assumptions frequently, without being attached to the original thesis
- Adding new data sources to your investor plate constantly
- (Try) to Define upfront what evidence / data points would change your view
- Adjusting follow-on strategy as information evolves
But here’s a deeper challenge: recognizing when the entire system has changed.
Sometimes we weather forecasters encounter a situation where the models are just not working- they digest atmospheric conditions that they just can’t solve for. When this happens, experienced forecasters lean more heavily on their own judgment. What looks like “gut feeling” from the outside is actually pattern recognition built from having seen hundreds, sometimes thousands, of prior scenarios: an internal library of outcomes, failures, and edge cases that never fully makes it into a model.
In venture, sometimes markets behave the same way- fundamental conditions change and strategies that worked brilliantly under one set of conditions can degrade rapidly under another. We are living through such a shift right now. The AI era represents a fundamental change in how software is built, how value is captured, and where moats exist.
We can no longer apply 2019 SaaS frameworks to 2026 AI companies. This may be like using a confused model to forecast tomorrow’s weather. We really want to trust it, but the underlying system has changed.
Some hard-earned forecaster truths apply directly to venture:
- Actively question whether your framework still fits the current environment.
- Be willing to abandon playbooks that worked for years when evidence accumulates that conditions have shifted.
- Turn to the “gut feel”, slow down, and trust your internal model.
Good forecasting- and good investing- requires knowing when to trust the model, and when to trust the person who has seen the model fail before. Let us not forget that in many AI driven companies- sometime you really need the ‘human in the loop’ to make the right product decision.
Venture Forecasting in the Age of AI – Summary
My core message is simple: venture capital, like weather forecasting, is a blend of art and science. Both require processing vast amounts of data, yet ultimately depend on human judgment to navigate what remains fundamentally unpredictable.
Venture capitalists can learn a great deal from meteorologists. They succeed not because they can see the future, but because they have built systems that function reliably under uncertainty. I believe that the same discipline applies to venture investing in the AI era:
- Know your prediction horizon: Focus on near-term execution and long-term structural trends, while accepting chaos in the middle.
- Think in ensembles: Acknowledge the full range of possible outcomes, not just the one you prefer.
- Update continuously: Refresh assumptions as new information emerges.
- Recognize when the model breaks: And rely on judgment shaped by experience.
In the age of AI, chaos is not an anomaly, it is a structural feature of the environment. The science of forecasting teaches us not to eliminate uncertainty, but to design decision-making systems that remain effective despite it.
Get in touch with Iren Reznikov:
