''' I see it all the time. Teams chasing the latest shiny model from a paper they just read, burning months on a feature that goes nowhere. They’re using the same old prioritization frameworks they learned in a generic product management course. That’s a fatal mistake when it comes to AI.
Having been in the trenches building AI products that reached millions, and now investing in the next wave of AI giants like Anthropic and Scale AI, I can tell you this: the standard playbook is broken. AI is not just another feature. It’s a different beast entirely. The uncertainty is higher, the data dependencies are more complex, and the feedback loops are unlike anything in traditional software.
At RemoteTeam, before we were acquired by Gusto, we weren’t using RICE or MoSCoW. We had our own system, forged from the lessons learned building AI at scale. It’s not complicated, but it forces you to confront the three questions that actually matter. I call it the Impact-Feasibility-Data (IFD) Matrix.
Your Old Framework Is Costing You Millions
Why do traditional frameworks fail for AI? They assume a level of predictability that simply doesn’t exist. You can’t just estimate story points for a research-heavy project. You can’t neatly slot an experimental model into a "high-impact, low-effort" box.
AI product development is a game of managing uncertainty. You’re dealing with three core variables:
- User Impact: Will this actually solve a painful problem for the user?
- Technical Feasibility: Can we actually build this? Is the model performance good enough for the real world?
- Data Availability: Do we have the right data to train, test, and validate this thing?
Most frameworks only focus on the first two. They completely miss the third, which is the fuel for any AI system. An amazing idea with a brilliant engineering team is worthless without the right data.
The Impact-Feasibility-Data (IFD) Matrix
The IFD Matrix is simple. It’s not a complex spreadsheet. It’s a way of thinking. For every potential feature, you score it from 1 to 5 across our three core variables.
Impact (1-5): How much will this move the needle for the customer? A score of 1 is a minor convenience. A 5 is a "hair on fire" problem solver that they would gladly pay for. This is where my "Top 1%" philosophy comes in. We’re not looking for incremental improvements; we’re looking for step-changes.
Feasibility (1-5): How confident are we that we can build and deploy this? A 1 is a pure research project with no clear path to production. A 5 is something we can build with existing, well-understood technology. Be brutally honest here. Your engineers will thank you.
Data (1-5): Do we have access to the necessary data? A 1 means we have no data and no clear way to get it. A 5 means we have a large, clean, and labeled dataset ready to go.
Now, here’s how you use the scores.
The No-Brainers (Score: 12-15)
These are your high-impact, high-feasibility, high-data features. If you find one of these, you drop everything and do it. At MovieLaLa (acquired by Gfycat), we realized we had a massive dataset of user-generated lists and a clear need for better recommendations. The impact was huge, the collaborative filtering tech was well-established, and the data was right there. It was a no-brainer. We shipped it in a month and engagement shot up by 30%.
The Big Bets (Score: 8-11)
This is where unicorns are made. These are typically high-impact, but with lower feasibility or data scores. They are your strategic R&D projects. You don’t bet the whole company on them, but you allocate a dedicated team to de-risk them. The goal is to turn a 2 in Feasibility or Data into a 4 or 5.
When I was advising an early-stage fintech startup, they wanted to build an AI-powered system to predict stock market movements. The impact was a 5. The feasibility was a 1. The data was a 2. It was a Big Bet, but a bad one. We instead focused on a different problem: automating financial reporting for small businesses. The impact was still a 4, but the feasibility was a 4, and the data (from accounting integrations) was a 5. That company is now a decacorn.
The Quick Wins (Score: 5-7)
These are the low-hanging fruit. Low impact, but high feasibility and data. They’re great for building momentum and keeping the team motivated. The danger is getting addicted to them. A series of small wins doesn’t compound into a big one. Use them sparingly. A good example is building a simple classifier to tag support tickets. It won’t change your business, but it can save your support team a few hours a week.
The Time Sinks (Score: <5)
Avoid these at all costs. These are the low-impact, low-feasibility, low-data ideas. They sound cool in a brainstorming session, but they are a black hole for resources. This is where most AI projects go to die. It’s the "AI for the sake of AI" trap. The classic example is a startup trying to build a generalized AGI to solve a niche problem that could be handled with a simple script.
A Real-World Example: AI-Powered Payroll
At RemoteTeam, we wanted to help companies stay compliant with international labor laws. This is a nightmare of a problem. The impact of solving it would be a 5.
Our first idea was to build a massive NLP model to read and interpret laws from every country. Feasibility? A 1. Data? A 1. It was a Time Sink.
We went back to the IFD Matrix. What if we narrowed the scope? Instead of all laws, what about just overtime rules? The impact was still a 4. Feasibility was now a 3 – it was a hard data extraction and rules engine problem, but solvable. And the data? We could get it. We could partner with legal experts to build a structured dataset. The data score became a 3.
This was a solid Big Bet. We assigned a small team to it. They spent two months just building the dataset. Then they built a simple rules-based system. It wasn’t a fancy deep learning model, but it worked for 80% of the cases. It was a huge win for our customers.
Stop Chasing Ghosts
Too many teams are chasing the ghost of AGI. They’re obsessed with the latest, most complex models, and they forget that the goal is to ship a product that solves a real problem. The IFD Matrix forces you to stay grounded.
Don’t ask "what can we do with this new model?" Ask "what is the most important problem we can solve for our customers, and what is the simplest way to solve it?"
Stop debating in the abstract. Put a score on it. Be honest about the data and the technical challenges. You’ll be surprised how much clarity it brings. This is how you move from building cool tech demos to shipping AI products that actually scale. '''
Frequently Asked Questions
How can I apply this thinking to my own situation?
Start by identifying the core principle behind the opinion, not the specific example. Then ask yourself: does this principle apply to my context? If yes, test it in a small, low-risk way before going all in.
How has this view evolved over time?
My thinking on most topics has changed significantly over the years. Early in my career, I held many conventional views that experience proved wrong. I try to update my beliefs when the evidence changes.
What experience informs this perspective?
This perspective comes from over a decade of building companies in Silicon Valley, two successful exits (RemoteTeam to Gusto, MovieLaLa to Gfycat), and investing in 200+ startups including Anthropic, OpenAI, and Scale AI. I write about what I've lived.
Do all experts agree with this view?
No, and that's fine. The best ideas in business are often contrarian. I share my perspective based on my experience and data, but I encourage you to seek out opposing viewpoints and form your own conclusions.