What a Year of AI Implementation Taught Our Advisory Firm

So firms are integrating: quickly, expensively, and often without a clear picture of what they’re solving for. And the results show it. MIT’s widely cited GenAI Divide report found that roughly 95% of generative AI pilots deliver no measurable impact on the P&L. Not because the technology fails, but because of what the researchers call a “learning gap” in how organizations integrate it into actual workflows.

Ai adoption among RIAs has more than doubled since 2023. Full integration has barely moved.

Over the last year, we have made AI implementation a genuine priority, investing across both ends of the spectrum: third-party tools connected to our existing stack, and all-in-one platforms with AI built in. That hands-on experience taught us things no vendor demo ever would. It also gave us a standard.

Every AI decision we make has to answer five questions before it goes anywhere near a client. Here they are, and what each one has cost us to learn.

Here is the uncomfortable truth: something very tech-forward can actually slow down team members who are not tech-savvy. This is not a resistance-to-adopt problem, and your team is not stubborn. It is the fine line leadership rarely sees until too late, between making things easier for managers and making more work for users. The confidence gap is real and measurable. An SBA Office of Advocacy report found only 27% of small businesses feel confident adopting AI effectively, compared to 82% of mid-sized firms. Most independent advisory practices sit in that small-business range, so we never assume proficiency will take care of itself.

We learned this one the hard way, working with high-end developers and tools.

There are two broad paths to bringing AI into your operations: connect third-party tools to your existing stack, or move to a comprehensive platform with AI built in. We have invested in both, and each has real costs that never show up on the pricing page. The big all-in-one platforms promise everything under one roof; what they do not advertise is that getting them to do what a specific firm needs usually takes custom development, real setup time, and serious ongoing investment to maintain. You are not buying a solution. You are buying a foundation and a construction project.

The fragmented route has its own trap. A stack of best-in-class point solutions wired together with middleware and connector workflows can work, right up until it doesn’t. Every connection is a potential failure point, every workflow is logic someone has to remember and debug, and “which system is the source of truth?” becomes a weekly conversation. The typical AI-using small business now runs a median of five AI tools, an operational stack whether or not anyone designed it as one.

After investing seriously in both, our takeaway is simpler than I expected: choose a platform or tool you actually understand and can edit yourself. The ability to open the hood and make a change without filing a ticket or hiring a consultant is worth more than almost any feature on a comparison chart. MIT’s research found that purchased tools succeed about twice as often as internal custom builds. Buy what works, but make sure you can operate what you buy.

And before connecting anything, we interrogate the impulse: what are we actually trying to do, and why? We think about it the way we think about investing, with a human in the loop. AI should accelerate the work. It should never become an unsupervised system making decisions no one on the team can trace or explain to a client, or to a regulator.

Every AI investment should answer one question in a sentence: who does this help, and how? We hold each one to a measurable improvement, for clients or for our team.

Is it improving the client experience, with faster responses, better personalization, fewer dropped balls? Is it making the team more efficient, with less data entry and less time hunting for information? Both are legitimate. “Everyone, eventually, probably” is not. And “usable” is the bar we set up front, not after: it means the team works in the system by default, the data coming out is trustworthy, and the workflows run without anyone babysitting them. We define what that looks like before we start, with concrete measures like adoption rates, time saved on specific tasks, error reduction, and client response times. Skip that step and “success” becomes whatever the sunk cost forces you to call it.

The data suggests most firms have not done this. MIT found that of all the organizations evaluating AI tools, 60% looked seriously, 20% reached a pilot, and just 5% made it into production. Most never get past experimentation, investing in one or two use cases without a clear problem to solve or a way to measure whether it worked. If you cannot point to a specific group of people a tool helps, and how, it is fair to ask whether it is worth the time and capital at all. The fear of falling behind is not a use case.

Few AI projects survive the trip from demo to daily use.

This is the lesson that surprised me most, obvious as it seems in hindsight: implementing technology is a job, with real hours attached, and we treat it that way.

Someone has to map the current process, design how the new system fits, build it, test it, break it, fix it, document it, and train everyone else. You cannot expect people who already carry full-time responsibilities, and in our industry fiduciary responsibilities, to absorb that in the margins of their week and produce a good result. We learned that if no one owns the rollout and the ongoing training and support, one of two things happens: the work stalls, or someone quietly sacrifices their own responsibilities to carry it. Even when we bring in an outside developer, someone inside the firm has to lead. MIT’s 95% figure is not a technology failure rate. It is an ownership failure rate.

Ownership does not end at launch, either. There is a real difference between using AI, a consumer assistant helping someone draft an email or summarize a meeting, and building on AI: custom platforms and workflows with AI embedded in their operations. The first is low-stakes and nearly maintenance-free. The second is infrastructure. Models change, APIs update, and a workflow that ran flawlessly in March can behave differently in September. Someone has to monitor outputs, catch the drift, and maintain the system indefinitely. That is not a reason to avoid building. It is a reason to budget for the operational reality, not just the build. Too many firms price the project and forget to price the upkeep.

This is the question we treat as non-negotiable, and the one where careful vetting has saved us the most. It is also the part of an AI rollout our industry can least afford to treat casually.

A platform can encrypt your data and still not be secure enough for a financial firm handling client information. “Encrypted” is not the same as “secure for handling client PII.” Encryption usually means protection in transit and at rest: protection against interception or a stolen server. It does not tell you who holds the keys, which systems can reach your content, or what happens if the vendor itself is compromised. That is different from access being truly restricted to us, and different again from a properly vetted vendor relationship, with documented security controls, confidentiality obligations, breach notification, and audit rights.

We hold platforms to that standard, and it matters. In one evaluation, a CRM described its connection to another system as a “direct integration.” As we worked through onboarding, we found that the “direct” connection actually routed through third-party middleware that did not meet the standard we hold for client data, so we did not move forward. No client information was ever exposed. The difference only surfaced because we went looking for it before committing, not after.

Widespread habits, largely ungoverned in independent firms.

AI is worth integrating into your firm. We have seen the gains firsthand, and we are continuing to invest. InspereX’s survey found 63% of advisors believe AI will let smaller practices compete with much larger firms, and I think they are right. But the firms getting real value are not the ones moving fastest or buying the most. They are the ones asking the unglamorous questions first: Does this work for our whole team? Can we change it ourselves? Who does it help? Who owns and maintains it? And where, exactly, does our clients’ data live? Those are the same questions we ask on behalf of every client whose financial life depends on us getting this right.

The second is the quality of our own decisions. The same discipline we apply to client-facing tools applies to how we look at our own practice: understanding what is working, where we can improve, and how to be more intentional about where we invest in growth. Better information leads to better decisions, but it never replaces judgment. It sharpens it.

That is the version of AI worth building toward: not the flashiest tool in a demo, but the one that quietly makes the client experience better and the firm sharper at the same time.

Sources

• Schwab Advisor Services, RIA AI Adoption Study (January 2026)
• InspereX, Spring 2026 Pulse Survey of 783 financial advisors (May 2026)
• MIT NANDA Initiative, The GenAI Divide: State of AI in Business 2025
• Gallup, AI in the Workplace (2025–2026)
• WalkMe, AI in the Workplace Survey (2025)
• LayerX Security, Enterprise AI and SaaS Data Security Report 2025
• ISS Market Intelligence, Advisor Pulse Survey (June 2025)
• SEC Regulation S-P Amendments (adopted May 2024; smaller-entity compliance in effect June 3, 2026)

Disclosures

This material is provided for informational and educational purposes only and should not be construed as individualized investment, tax, or legal advice. The opinions expressed are those of Strategic Advisory Partners as of the date published and are subject to change without notice. All investing involves risk, including the possible loss of principal.

Any statistics or third-party information referenced are believed to be reliable but cannot be guaranteed for accuracy or completeness. Examples provided are hypothetical and for illustrative purposes only and do not represent actual client experiences or guaranteed outcomes.

Retirement planning strategies should be evaluated based on an individual’s unique financial situation, goals, and risk tolerance. Before making financial decisions, individuals should consult with their financial, tax, and legal professionals.

Strategic Advisory Partners is a registered investment adviser. Registration does not imply a certain level of skill or training. More information about Strategic Advisory Partners, including its investment strategies and objectives, can be found in its ADV Part 2 or Form CRS, which are available upon request.