6 days ago
Backing Durable AI: What Works When Vetting Emerging Tech
As Vice President at WestBridge Capital, Achal Singi focuses on investment and strategic guidance for emerging technology companies.
Every week brings us a new AI platform promising to change the industry. The AI market is learning quickly, but speed alone doesn't translate into returns. Most products are easy to outpace and even easier to imitate.
For finance leaders, the goal is to find investments that still make sense in five years' time. The fundamentals haven't changed: Strategic fit, capital discipline, timing and moats still matter. But the filters we use to evaluate those fundamentals need to be sharper. AI's velocity amplifies both opportunity and risk, and that puts more pressure on financial discipline.
In my experience, the strongest signals of staying power show up in familiar patterns. This article covers the filters I've learned to trust when evaluating long-term AI investments.
Bet on data moats over feature velocity.
Too many AI investments are judged on performance benchmarks that don't last as long as investors believe. Faster inference and better LLM scores may look impressive today, but they age out quickly. What lasts longer and creates defensibility is a structural data advantage.
This might take the form of customer feedback loops that improve model performance, or industry-specific datasets that are difficult to substitute. A useful test is to strip away access to the firm's unique data: Does the product still hold up, or does it become interchangeable with competitors? If it's the latter, you're probably looking at a vendor dressing up as a platform.
Why does this matter? Access to public data is tightening, and regulatory scrutiny over training data is only growing. Going forward, product differentiation in AI will hinge almost entirely on who owns or controls unique data streams. Smart investments deepen a company's proprietary data edge. That might mean backing products that can mine insights from operational exhaust, or capabilities that allow firms to license internal data externally. What's important is defensibility through access.
While features can be reverse-engineered or leapfrogged, a data moat, especially one that's reinforced by regulation, compounds over time and is far harder to dislodge.
Invest where capabilities are adjacent and compounding.
Some of the strongest AI returns come from initiatives that extend what a company already does well. Before asking how much an AI initiative might cost or save, finance leaders should ask: Is this close to something we're already good at?
When an AI project builds on existing distribution channels or data infrastructure, integration is faster, and iteration can be cheaper and more predictable. Bank of America's 'Erica,' their virtual financial assistant, started as a front-end customer support tool. As of 2025, it has reduced IT service desk calls by over 50%, and more than 90% of employees use Erica for Employees regularly for internal operations. Because it was built on top of familiar workflows and data flows, each expansion required less incremental lift and created more value over time.
These kinds of adjacencies also make capital planning more efficient. Projects that slot into existing operational budgets or pilot programs are easier to pace than those requiring stand-alone capital or GTM overhauls. The former reinforces what already works; the latter introduces execution risk.
The takeaway shouldn't be to 'play it safe,' but to build AI initiatives where your cost of iteration is lowest and your distribution advantages are highest.
Time investments around regulation.
No AI investment grows in a vacuum, least of all in industries such as healthcare, financial services, cybersecurity or critical infrastructure where regulatory pressure is always evolving. In these industries, regulatory inflection points are just as important as product milestones.
Getting in early, before a data portability mandate or AI audit requirement kicks in, can unlock temporary advantages and reorder market positions. In those moments, compliance can be a positive market-shaping force.
This is particularly relevant when backing early-stage companies. For example, compliance with the EU's AI Act can be the difference between market access and market exclusion. The act governs where a product can be sold, how it can be trained and what data it can access from what parts of the world. To complicate things further, because many of these compliance costs can be written off as R&D or administrative overhead, their true cost can be difficult to trace. A project might look efficient on paper but burn cash on back-end overhead.
Finance teams need to dig deeper, and time their entry carefully. Move too early, and you may build to the wrong standard. Move too late, and you'll find yourself overpaying to retrofit compliance. But time it right, and regulatory alignment can create durable tailwinds and, in some sectors, a strong temporary moat. That window might only be open for six months, but in a space this fast-moving, they can define a category.
Focus on long-term value creation, rather than novelty.
What ties these filters together is structural alignment. Each helps identify investments that reinforce the core engine of value creation, rather than distracting from it. More importantly, they remind us to think more carefully about long-term compounding in an environment that's driven by novelty.
The AI market will continue to move fast, and we shouldn't pretend we can forecast where every innovation will land. What we can do is ground our decisions in what holds up over time. AI may feel like a frontier, but the way we evaluate it doesn't have to be.
The information provided here is not investment, tax or financial advice. You should consult with a licensed professional for advice concerning your specific situation.
Forbes Finance Council is an invitation-only organization for executives in successful accounting, financial planning and wealth management firms. Do I qualify?
