We launched Omaha commercially a bit more than a year ago, and we’ve since had hundreds of conversations with PMs and analysts across the globe about the way they work, and crucially, what they would and wouldn't trust a machine to do.
These conversations often contradicted the general discussion on the impact of AI on investment management. The idea that a blind trust in LLMs will almost immediately and unilaterally forge investment decisions is still overrated. Make no mistake: AI is radically transforming investment management, but it will not do so completely overnight and will rely heavily on human judgement and ownership.
We have chosen to focus on three of the main problems encountered by our clients: AI is only as good as the data it is fed with, its models usually carry a trade-off between robustness and flexibility, and it cannot take ownership of its decisions.
Data quality and differentiation
The first issue is the most fundamental one, because everything else in this list depends on it.
“Garbage in garbage out”: in a world where LLMs and MCPs are becoming standard, the quality of your data sets the floor and the ceiling on the quality of your answers. The trust threshold is asymmetric and much higher than some vendors might assume. Both quant and fundamental investors told us versions of the same thing: when data is right 95% of the time but you can't tell which 5% isn't, the whole data set must be checked anyway.
There's also a simpler market logic: if every fund runs the same AI process on the same data, the edge disappears, and what’s wrong becomes exponentially wrong. Crowded AI strategies are no different from any other crowded trade.
As models are becoming a commodity, data and framework become the edge. The Terminalist made a similar point recently about terminal vendors bolting chatbots onto their platforms: a model that only sees what everyone else sees is forced into consensus by construction. Everyone will soon have access to roughly the same intelligence. Very few will have clean, normalised, genuinely comparable data, and even fewer will have a repeatable framework for turning it into a view, plus the firm-specific context that shapes conviction. That combination is far more defensible than any individual's insight, because it compounds and becomes an institutional advantage.
Reconciliation and Flexibility
But clean data on its own isn't the whole constraint if you have no way to audit and properly challenge it.
Only a small minority of our clients/prospects mentioned analysis as their primary bottleneck. More often, the conversation turned to reconciliation issues, comparability, and freeing up the time and capacity to easily challenge a model’s assumptions. A large portion of analysts and PMs still want to work in Excel and it’s not without justification, but because a spreadsheet is transparent, modifiable, and theirs.
Some analysts described spending their week making numbers agree across sources before they could form a view or developing a consistent and correct valuation framework across the firm. Beyond this, accounting allows companies a fairly well-known degree of flexibility and generally its regulations have consistently lagged what happens in the real world: different definitions of adjusted EPS, inconsistent treatment of leases, capitalised R&D, accelerated amortization are all examples, and are also evolving at every new filing release.
Ownership and Implementation
Clean data and flexible tooling improve the analysis but neither answers the question of who ultimately owns the conclusion.
Most asset managers are not IT companies, and agentic AI requires implementation costs, maintenance and institutional trust that can take years to build. Seeing AI constructing one financial model is no doubt impressive. However, building and maintaining thousands at production quality is a task with very different economics. Crucially, when a model gets it wrong, fiduciary responsibility doesn't disappear, and someone has to answer.
There, the binding constraint isn't computational, but rather a general sense of accountability. When markets move against you (in the short term), the decision and most importantly the decision-making process must still be owned and justified. Further improvements in AI will undoubtedly automate more of the analysis, yet the judgement that the work supports, and the accountability that comes with it must remain human.
So, what do you build today if you take all of that seriously? Not a chatbot on top of a database. The conversations pointed to places maybe more boring and yet more durable: get the data right, use AI where it compounds rather than where it demos well, and leave the judgement with the person who has to defend it, that's the order we've built in.
Yes, we're a data vendor shipping AI features in a post arguing that AI features are becoming commodities. The distinction we'd defend is order: AI layered onto unreliable data industrialises the problem, while AI layered onto clean, point-in-time, genuinely comparable data does something else entirely.
Smarter data is the foundation...
...and it has been our core conviction at Omaha from day one. Nearly all of our engineering effort goes into cleaning, transforming and building a framework on top of accounting data so that it reflects the economic reality of businesses across the world. This needs to be the foundation of all analysis, whether human or AI, to allow any conclusion drawn from it to be trusted. That is why we have been assembling a team of talented accountants, analysts and specialists to drive innovative financial analysis, and it is what everything that follows in our approach to AI is constructed on.
The point is also to demonstrate that markets do substantially reward investment decisions that look through accounting data to the economic reality beneath it. That sometimes rather unappealing extra mile. Everything that follows in our approach to AI is constructed on these foundations.
AI as a massive innovation enabler
The first use of AI at Omaha is internal and deliberately used to enhance our core expertise. We integrate AI across data ingest, processing and evaluation. The models help because we have the financial expertise to make the data reflect that economic reality, and AI allows us to do it at scale.
The second use is external, and it rests on the same knowledge base. We built an applied AI team to take those principles beyond our own infrastructure and put them in our users' hands. Our MCP server is now live and “the Oracle”, our first agent will dramatically enhance users’ insights and experience. Further agentic features are following quickly, shaped by what clients tell us they need today, and by what we can already see they'll need once the capability exists. This all is forged by the right context and reasoning, developed through decades or equity research and modelling.
Improving ownership and flexibility
Two investors can look at the exact same company on Omaha's platform, the same data economic data, and reach diverging conclusions. The first decides the stock is far too cheap for a business of this quality. The second reasons that you'd be paying a premium for a moat that won't last. Both are defensible. But note what they're now arguing about: the durability of a competitive advantage, the operating nature of a given asset, not whether the numbers are comparable. Good infrastructure doesn't make the decision for you; it moves the disagreement to where it's worth having.
Even once a user accepts Omaha's overlays as the best representation of reality, the ownership problem persists because the error may not come from pulling the wrong data, but from reasoning badly over the right data. Either Omaha does the pre-work to reduce it, or the user can own part of this subjectivity.
That's why, in the coming weeks, we'll also be introducing "the Lab": the ability to challenge and modify a large share of our model assumptions, acting as a genuine sandbox for analysts and PMs, one where they can apply their assumptions and view of the world on top of our framework. And for those who'd rather stay in Excel, the Omaha Excel Linking Tool is already live, and we're extending it with agentic features.
There is growing pressure on firms to embed AI in their processes at every turn, and rightly so: the computational power that AI models provide is unprecedented. However, mindless implementation carries real danger and can severely weaken a firm's credibility.
Generally speaking; the enthusiasm is often strongest at the top and maybe more nuanced in the middle. People who'd have to defend the output to an investment committee are the cautious ones, because they're the ones who have to justify their decisions.
With all that in mind, we're optimistic about where this goes, and a bit more sceptical about how fast. Not because the technology disappoints, because the constraint was never only computational. It's trust, comparability, and someone's name on the trade, those that move at human speed. The work that pays off is therefore the work that would have paid off anyway: get the data right, build a framework that survives changes in accounting regulations, and use AI where it compounds through your process.
That's our view, and we're curious to hear others, especially from those who disagree. Please reach out to anyone in the team!