Using AI as a Financial Analyst: A Year of Portfolio Fine-Tuning

I've spent the past year using AI to do something I didn't expect: turn my already conservative investment approach into something even more structured, more defensible, and frankly more intellectually engaging than I thought possible. Not to chase returns or beat the market, but to build a portfolio system that can outlast me.

Now that I'm retired, the goal with my investable cash is modest and very conservative: stay 1-2% ahead of inflation, preserve capital, and keep myself mentally engaged. After decades in executive roles at manufacturing companies and making business investments, I have no illusions about getting rich from stock picking. What I wanted was a framework that others could understand and execute if something happened to me, and the mental stimulation that comes from following fields I know well from my engineering and medical device career.

AI turned out to be surprisingly useful for exactly this combination of goals.

The Barbell Within a Barbell Within a Barbell

First, the structure. My total assets follow what I call a barbell approach at multiple levels.

The outermost barbell includes all assets: business interests I still maintain, private placements and direct investments, commercial and personal real estate, family assets, and investable cash. This is the broadest diversification layer.

Inside that, the investable cash itself follows a very conservative barbell structure: 80% in treasuries, treasury-backed securities, and high-quality corporate bonds. Another 15% in dividend aristocrats (JNJ, PEP, ABT, ABBV, NEE, etc.) and quality compounders like Berkshire Hathaway, Markel, Danaher, and Fairfax Financial. The final 5% goes into speculative portfolios.

Then each speculative portfolio has its own internal barbell: conservative anchors, picks-and-shovels plays, and a few pure speculative positions. My five speculative themes are space technology, AI applied to biopharmaceutical discovery, quantum computing, energy/grid resilience, and biomedical companies I believe are ripe for acquisition.

The structure isn't new. What's new is how AI helped me formalize, document, and monitor it.

What AI Actually Did

Over the past year, AI has become something like a junior analyst who never sleeps, never forgets details, and doesn't mind repetitive work. Here's what that looks like in practice.

Building a Documentation Hierarchy That Outlasts Me

This turned out to be AI's most valuable contribution, though I didn't realize it at first. I had portfolio logic scattered across mental models, old spreadsheets, and decades of experience. That works fine until it doesn't.

AI helped me create a three-level documentation structure that anyone could pick up and understand:

Level 1: Family Management Overview. This top-level document describes all assets across all categories: business interests, private investments, real estate holdings, family trusts, and investable cash. It explains the overall philosophy (conservative, inflation-plus-modest, no need to take big risks), how the pieces fit together, and where to find detailed information on each category. If something happens to me, this is the roadmap.

Level 2: Asset Type Strategies. Each major asset category gets its own document. The investable cash strategy explains the 80/15/5 barbell structure, why treasuries and bonds anchor everything, how the dividend and compounder stocks are selected, and what role the speculative portfolios play. It's detailed enough that a financial advisor could execute the approach without me.

Level 3: Speculative Portfolio Strategies. Each of the five speculative portfolios has its own strategy document covering investment thesis, market opportunity, how to evaluate companies in this space, selection criteria, position sizing rules, rebalancing triggers, and exit conditions. These documents run several pages each and represent months of refinement with AI's help.

The exercise of articulating this forced clarity I didn't have before. AI would ask questions like "What specifically do you mean by 'appropriate market cap'?" or "How do you define a 'conservative anchor' in the quantum portfolio?" Answering those questions made me realize I'd been operating on intuition that needed to be codified.

Developing Company Filters and Mini-ETFs

Each speculative portfolio follows what I call a "mini-ETF" approach: 10-15 holdings within a theme, structured with large company anchors that already generate revenue in the space, mid-cap picks-and-shovels plays, and smaller pure speculation positions. But finding companies that fit requires systematic screening.

AI helped me develop filters that balance my engineering background's desire for rigor with the reality that I'm still taking meaningful risks:

Must trade on a major US exchange. NYSE or NASDAQ, no OTC or pink sheets. This ensures basic disclosure requirements and liquidity.

Must have analyst coverage. But not just any analysts. AI helped me identify which research firms and individual analysts actually understand each sector. For quantum computing, that meant focusing on analysts who've covered semiconductor manufacturing and high-performance computing. For space technology, analysts who understand aerospace contracts and government procurement. For biopharma AI, analysts who've covered both drug development and enterprise software.

Market cap in the sweet spot. Large enough for liquidity and to avoid penny stock dynamics (generally $200M minimum, though this goes lower for the Biomedical Acquisition portfolio where smaller companies are often acquisition targets), but small enough to have meaningful growth potential if the thesis works (generally under $5B for the speculative plays). This rules out the mega-caps where I'd just be a passive index investor, and the micro-caps where one bad quarter can crater the stock.

No day trading, no options, no leverage. These are long-term positions, often held 2-3 years or more. The whole point is intellectual engagement with fields I know or have interests in, not trying to time markets.

Here's what the five portfolios actually look like:

Space Technology: Anchored by larger aerospace companies with established space divisions, then mid-cap players like Rocket Lab (launch services and satellite components), Planet Labs (earth imaging), Redwire (space infrastructure), AST SpaceMobile (satellite communications), Spire Global (space-based data), and several others in satellite manufacturing and ground station infrastructure. The thesis is that launch costs keep falling and the space economy is industrializing.

AI Applied to Biopharma Discovery: Recursion Pharmaceuticals (AI-driven drug discovery), Schrodinger (computational chemistry), Absci (generative AI for antibody design), plus several smaller computational biology platforms and drug discovery tools companies. I spent 30 years in medical devices, so the combination of AI and drug development fits my background.

Quantum Computing: Anchored by IBM and Honeywell, which surprises people who don't realize both companies have substantial quantum computing investments and already generate revenue from quantum services. Then pure-plays like IonQ (trapped ion systems), Rigetti Computing (superconducting qubits), D-Wave Quantum (quantum annealing), and pick-and-shovels companies making specialized components for quantum systems. This is high-risk because commercialization timelines are long, but the large anchors provide stability while technical progress accelerates.

Energy/Grid Resilience: Anchored by GE Vernova, then mid-cap specialists like Fluence Energy (battery storage systems), Array Technologies (solar tracking), and positions in grid infrastructure hardening, transmission upgrades, and microgrid control systems. Climate change is making grid reliability critical, and storage technology is reaching economic viability.

Biomedical Acquisition: 15-20 small medical device, diagnostics, and healthcare IT companies (typically $200M-$800M market cap) that fit specific criteria making them attractive acquisition targets for larger players. AI helped me develop these criteria: strong IP portfolios, FDA clearances or regulatory approvals, recurring revenue models, defendable niches, management teams with prior exit experience, and evidence of strategic interest from potential acquirers.

The back-testing was revealing. AI analyzed companies meeting these criteria over the past decade and found that 15-20% get acquired each year, with an average 30% premium to pre-announcement trading price. Not every acquisition works out that cleanly, and timing is unpredictable, but the model suggests this portfolio should see 2-3 acquisitions annually. The companies that don't get acquired often "just" become solid small-cap holdings generating steady returns - still profitable investments.

AI didn't pick these companies. I did, after extensive research. But AI helped me build the screening criteria, identified companies I'd missed, and organized the research process.

Modeling the Contribution

One of AI's more sophisticated applications was helping me model how the speculative portfolios contribute to overall returns. The goal was never to beat the market but to stay 1-2% ahead of inflation with my total cash allocation.

AI helped me build scenarios showing how the 5% speculative allocation could contribute roughly 2% uplift to the overall investable cash portfolio over the long term. Not every year, obviously. Some years the speculative portfolios will lag or even lose money. But across a 5-10 year horizon, if the theses play out even partially, that 5% allocation working harder than the conservative 95% can meaningfully improve total returns.

The math is straightforward but the modeling made it concrete. If the conservative 80% generates inflation plus 0.5% and the dividend/compounder 15% generates inflation plus 3%, you need the speculative 5% to deliver mid-teens returns over time to hit that overall inflation-plus-2% target. That's aggressive but not unreasonable for well-researched positions in emerging sectors with real tailwinds. Overall it's significantly lower risk than a more bell-curve portfolio structure with a larger amount in moderate risk investments.

Having this modeled helped me right-size positions and stay disciplined about not letting the speculative allocation creep above 5%. It's easy to get excited about quantum computing breakthroughs or space industry developments and want to increase exposure. The model reminds me that these positions are already doing their job at 5% and increasing allocation would introduce risk I don't need.

Other AI Contributions

Portfolio structure refinement. I described my approach and holdings to the AI, then asked it to identify inconsistencies, gaps, and risk concentrations I might have missed. It caught things I'd overlooked, like having too much exposure to satellite communications across both the space and energy portfolios.

Monitoring spreadsheets. AI helped design spreadsheets tracking metrics that matter for each portfolio, and AI agents keep them updated. For space technology: contract wins, launch success rates, revenue growth. For quantum computing: qubit counts, error rates, partnership announcements. The spreadsheets keep me focused on fundamentals rather than stock price noise.

Weekly news digests. Every week, an AI agent scans news and filings for all speculative holdings and generates a digest with impact analysis. A partnership announcement here, an insider sale there, a competitor's product launch somewhere else. This takes what used to require hours and condenses it to 15 minutes of reading.

The Limits and Guardrails

I'm not naive about this. AI doesn't know the future and lacks the judgment that comes from decades of experience in business. Every recommendation gets verified. Every "insight" gets challenged. And absolutely zero personal data goes to AI unredacted. Portfolio values and account numbers all get anonymized before any AI interaction.

What surprises me is how useful that level of assistance actually is. The difference between spending six hours researching something versus one hour makes me more likely to do the work at all. The difference between having vague mental rules about position sizing versus written frameworks makes me more disciplined. The difference between hoping my family could figure out my portfolio logic versus having it clearly documented brings real peace of mind.

Why This Matters Beyond Returns

The speculative portfolios represent 5% of investable cash, which itself is just one slice of total assets. We're talking about a small portion of wealth where the absolute dollar impact of being right or wrong is manageable. But the intellectual engagement is substantial.

Following developments in quantum computing, space infrastructure, synthetic biology, and grid modernization keeps me connected to fields I spent my career in or find interesting. The research is genuinely interesting. The companies are working on problems that matter. And the exercise of building systematic approaches to speculative investing scratches the same itch I had running companies, just with different stakes and far less stress.

AI has made that engagement more structured and sustainable. Instead of scattered research sessions when curiosity strikes, I have a system that surfaces what matters and archives what I've learned. The barrier to staying informed drops low enough that I actually do it consistently. The documentation means my family isn't inheriting a black box if something happens to me.

Could someone achieve all of this without AI? Absolutely. People have been building disciplined investment systems for centuries. But the combination of research assistance, documentation support, systematic screening, back-testing capabilities, and ongoing monitoring at essentially zero marginal cost changes the calculus. What used to require hiring an analyst or spending weekends buried in spreadsheets now takes a few hours a week.

For someone in retirement looking to stay mentally engaged with fields they know well, preserve capital, and build systems that outlast them, that's a valuable combination. The goal isn't alpha. It's a sustainable system that documents itself, stays within defined risk parameters, and keeps the mind sharp.

So far, it's working.