Build vs. buy: The hidden costs of in-house wealth management AI

It’s a fair question for wealth management firms to ask: If we need AI solutions to stay ahead of the game, why not build them ourselves?
Internal builds are happening—and have been happening—in the largest firms in the industry. These builds are highly visible and newsworthy. Merrill and Bank of America Private Bank spent a year building an AI meeting tool before rolling it out earlier this year. JPMorgan Chase, Morgan Stanley, and Goldman are all building in-house solutions right now.
The thinking goes: If the biggest names are building in-house, shouldn’t we at least consider it too?
Generative AI makes this line of thinking tempting. These tools have drastically lowered the barrier to building a prototype that looks like a finished product. But they haven't eliminated the years of production learning required to make AI solutions reliable. Wealth management firms face real compliance obligations, real household complexity, and real client relationships.
The biggest names in the industry have deep benches of technical talent, and engineering budgets just as deep. More than that, the builds happening now rely on years of preparation to align technical capacity with the real-world repercussions of wealth management.
Even firms that have built in-house are now looking for external solutions, too.
For firms weighing an internal build, the question is less about the technology and more about whether developing and maintaining their own capabilities justifies the investment, especially when specialized platforms already offer production experience, domain knowledge, and infrastructure they can build on.
A prototype is not a production system
A proof of concept can demonstrate, often within weeks, an AI tool’s capability of summarizing meetings, answering questions, updating CRM records, or executing workflows. Those sorts of tasks have genuinely gotten easier to accomplish. They’re also why so many firms feel emboldened to build.
But a demo succeeding is where the easy part ends. The hard part begins in production.
Production introduces problems that don’t show up in a pilot, and that are much more difficult for a solution to address intelligently:
- Incomplete, inconsistent, or conflicting client data across data sources
- Wealth-specific terminology and relationship structures
- Firm-specific workflows—or even team, office, and advisor-specific workflows
- CRM and custodian integrations that need to hold up under real-world usage
- Permissions and approval requirements across teams and roles
- Compliance exceptions that differ across jurisdictions
- Model changes and regressions as underlying large language models (LLMs) are updated
- Token costs that are difficult to predict and control without usage monitoring
- Reliability across thousands of in-house users and their clients
Any one of these problems reduces wealth managers’ ability to rely on an AI solution and makes it impossible for them to truly integrate it into their workflow. That integration is key: The 2025 Kitces AdvisorTech Report reveals that integration—how well a firm’s systems communicate data automatically—is the strongest predictor of how satisfied wealth managers are with their tech. Yet fewer than one in three advisors report having it.
Layering an underdeveloped tool on top of a fragmented tech stack doesn’t fix the disconnects; in fact, the disconnects tend to surface. Building an AI solution that can withstand a firm’s actual messy real-world environment is where most of the development work lives.
Generative tools can accelerate development, but reliable deployment still requires testing, refinement, and ongoing support. Those responsibilities continue long after the prototype works.
The real advantage is in accumulated patterns
A mature wealth-specific AI platform is more than a codebase and a user interface. It reflects years of accumulated learning from real deployment: it accounts for mistakes already made, edge cases already encountered, solutions already developed.
That accumulated knowledge shows up in areas like:
- Wealth-specific ontology and data structures built around real-world cases rather than abstract or generic CRM records
- Advisor and operations workflows shaped by how tasks actually move through a firm
- Household and relationship complexity as varied as clients themselves
- CRM and custodian interactions that have already been stress-tested
- Compliance guardrails informed by how AI-generated records need to satisfy actual regulatory bodies
- Human-in-the-loop requirements calibrated by wealth manager usage
- Edge-case handling for the high-value, high-risk circumstances a demo cannot anticipate
- Ongoing model evaluation and continuous assessment
- Patterns for user adoption and exception management at scale
Buying a mature platform means acquiring years of production learning and acceleration, without having to reinvent the wheel, as it were, in real-time scenarios that impact clients.
The hidden cost of building AI in-house is rediscovery
The visible cost of an internal build is, put simply, engineering time. But the total cost of ownership extends far beyond development.
Firms must also account for product development, integration maintenance, security and compliance standards, model governance and evaluation, monitoring and incident response management, infrastructure, training, adoption, ongoing support, and product iteration. These aren’t one-time costs, either. They’re ongoing and indefinite.
The highest cost of all may be opportunity cost. Internal AI teams spend time recreating foundational capabilities already available elsewhere—meeting-notes pipelines, CRM syncing capabilities, compliance guardrail systems. That’s time that could be dedicated to work actually unique to the firm.
Every firm, no matter its size, has finite resources. The resources given to in-house builds come at the cost of delaying other valuable work that could advance the firm and the client experience.
Even firms that build do not need to build everything
Internal AI expertise and external platforms are not mutually exclusive. For many firms, these are complementary strategies.
It’s worth recognizing that even the big-name institutions choose not to build every layer themselves. JPMorgan Chase, for example, rolled out its initial AI assistant tool (LLM Suite) as a portal to external models rather than developing its own from scratch. That approach separates building an internal application from developing the underlying models, while leaving the firm responsible for its own application and integrations.
Firms don’t have to be one of the big dogs to play ball. Wealth management firms at various scales can, and increasingly do, combine proprietary agents, proprietary data, and firm-specific workflows with specialized technology that provides proven infrastructure and domain expertise.
We are seeing sophisticated firms, with internal AI initiatives well underway, still evaluating external, industry-specific platforms for various portions of their stack. Some problems are simply faster and more resource-efficient to solve through existing production patterns.
Having a dedicated internal AI team does not mean every AI capability should be developed internally. Internal teams can apply the firm’s expertise to organization-specific problems while working with external platforms to put those solutions into practice. That includes shaping proprietary workflows around the firm’s own standards.
Put the firm’s expertise to work
A firm’s expertise shows up in how it works with clients. An advisor knows which questions to ask, when to bring in a specialist, and how to handle a complicated situation. Over time, that experience becomes part of the firm’s playbook.
Turning those playbooks into agentic workflows does not require building the technology from scratch. A specialized platform can provide the infrastructure to coordinate work across systems while allowing the firm defines each workflow around its own requirements, business logic, and approval standards.
Take onboarding for a business owner who has just sold a company. The firm may have developed an approach that brings tax and estate planning specialists into the relationship early. Working with a technology partner, the firm can turn that approach into a workflow that gathers the relevant information and involves those specialists at the appropriate points. The process remains specific to the firm, even though the technology comes from outside it.
Internal teams have an important role in making this work. They can translate the firm’s expertise into clear requirements, connect proprietary data, evaluate outcomes, and refine how workflows support the business. A process being unique to the firm is not, by itself, a reason to build the underlying technology.
Buying also requires scrutiny. A platform may impose limits on customization, data access, or how easily a firm can change providers. Internal development can make sense when a strategically important requirement cannot be met externally, and the firm is prepared to support it over time. The decision depends on the capabilities required, the options available, and the full cost of maintaining them.
The strategic advantage isn’t ownership—it’s allocation
The ability to build an AI solution internally is not the same thing as realizing an advantage by doing so.
Likewise, just because a firm can build its own solutions—and can seem to do so more easily and more often, with the constant improvement of generative AI tools—doesn’t mean investing years in production iteration is worth the opportunity cost. It’s easier and easier to shortcut a functional prototype, but you can't shortcut the production problems that are non-negotiable when serving wealth management clients.
The strongest enterprise AI strategies we’re seeing rely on specialized platforms that provide production maturity, domain expertise, and years of accumulated learning. They also give firms room to apply their own expertise.
The goal is to put the wealth management firm’s resources where they create the greatest impact. Sometimes that means building. Often, it means working with a technology partner to put the firm’s own playbooks into action.
Weighing a build against a buy? Or considering a hybrid path? Zeplyn’s agentic AI is purpose-built for wealth management from years of production experience. Book a demo today.