In March this year, I kicked off our Symphony Conference with the following statement:
2026 is the year AI becomes a necessity.
I’d spent the a few weeks over the Christmas/New Year break creating several prototype AI agents and had seen enough improvement in Microsoft’s Copilot platform, and the underlying models, to be convinced AI could finally deliver meaningful value by automating some of the more complex processes in a financial services firm.
For those who were in attendance, the above quote likely became the soundbite that stuck. However, shortly after that statement, I clarified what that meant in practise
businesses need to make sure they have a solid AI strategy in place.
I’ve worked in technology and financial services for the better part of three decades and over that time have witnessed the hype cycle time and time again. Those of you who have shared the Fin365 journey for any length of time can attest I’ve made a point of trying to provide a counterbalance to the hype and help our financial services firms avoid the pitfalls of “shiny object syndrome”. When it comes to AI, the Peak of Inflated Expectations, is larger than any I’ve witnessed throughout my career, driven, of course, by the very companies who want to sell you AI.
Standing behind my Symphony statement, I do believe AI will deliver some amazing benefits. But what the AI companies won’t tell you is that the risks and uncertainties are just as real. Without a solid AI strategy, therefore, financial services firms are likely to waste valuable resources and, potentially, open themselves up to future data security issues.
I understand the desire to adopt new technology. Fin365 was born as an unintentional outcome to my own search for better technology to overcome the inherent inefficiencies in a data and regulation heavy industry. However, that journey also provided some valuable lessons about what not to do.
Any sufficiently advanced technology is indistinguishable from magic. – Arthur C. Clarke
Unlike any technology that has come before, AI’s capabilities, give it the appearance of magic. Not surprisingly, AI’s emergence has been accompanied by a new crop of “AI Native” tech solutions, that have sprouted like magic mushrooms after a heavy autumn rain.
My intent is not to individually assess these tools. Our customers are already trialling most of them and providing feedback. While it’s early days, one theme is already clear … to deliver meaningful value these tools need to be able to access the data that sits in your CRM.
So, while your AI strategy should include controlled testing, it also needs to consider the following:
Who owns the (client) data?
One of Fin365’s founding principles is that the advice business owns the client relationship and, by extension, should own, and control access to, clients’ data.
The delivery of financial advice requires a broader dataset than any other participant in the financial services industry. Client identity, health and family details, financial positions, insurance details, historical communications and compliance records. Protection of this data is mission critical. You can outsource the technology. You cannot outsource the obligation. If an AI model trains on your inputs, or a sub-processor sits in a jurisdiction you’ve never heard of, the exposure doesn’t belong to them. It belongs to you, your AFSL, and your clients.
The questions every regulated business should ask before handing that data over: Where does it go? How is it used? Who can see it? How long is it stored? What steps have been taken to protect it?
This isn’t paranoia. A quick internet search will return real-world examples of data breaches occurring through a range of pathways, including employees unintentionally sharing sensitive data with public AI tools, leakage via coding assistants, models exposing memorized training data, and autonomous AI agents exfiltrating information through APIs and system access; often bypassing traditional security controls
Does this problem need AI?
Most businesses, right now, are pointing AI at “busy work”: Summarise this. Rewrite that. Watch this inbox. Draft this reply. It feels productive but, given the uncertainty around the true cost of AI, may not be delivering a meaningful ROI. Meanwhile the genuinely expensive problems – onboarding bottlenecks, review capacity, dirty client data – are data and process problems, that can largely be solved with existing tools and restructuring business processes. #datadatadata, as the choir at Symphony kept singing.
This is the architectural truth the feature announcements skate over: AI is only as useful as the data foundation beneath it. A clever agent bolted onto a messy foundation just reaches the wrong answer faster; and now you’re paying per token for the privilege. A clean client record, a defined workflow and two systems that talk will out-deliver that every time. So before “can AI do this?”, ask “does this need AI?”. Often, the honest answer is: fix the foundation first, and the problem dissolves without a model anywhere near it.
The actual price is … ?
Fin365 has been actively working on developing some of our own integrated AI tools. Toward the end of last year, we released an AI Filenote flow for Microsoft Teams & Dynamics. At this year’s Symphony conference, we showed off a number of prototypes, including our Compliance and Investment Analysts agents, both of which will be rolled out in Q3 this year.
One of the challenges we’ve identified, which very few seem to be talking about, is getting clarity on exactly what these AI models are going to cost over the long term.
Currently, AI is in a land-grab phase, where vendors subsidise, bundle and absorb the real bill to win market share before the meter gets honest. We don’t have to speculate about what happens next … it’s already started.
Earlier this month, Sam Altman openly admitted that cost concerns are the second most common complaint from OpenAI customers.
In May, Microsoft began cancelling most of its own engineers’ Claude Code licences. The reason wasn’t quality. It was cost. Fortune put it about as plainly as a headline can: in real use, the technology was proving “more expensive than paying human employees.”
Unfortunately, the direction of travel doesn’t appear to be reversing … or even slowing down. If you’d like to dive into the details, I recommend these articles by Josh Bersin, Neil Dhar and Steve Mordue
For those of you who just want the highlights …
- Expect materially higher and more volatile AI costs – driven by massive infrastructure investment that must be repaid, and a shift from flat per-seat fees to usage-based pricing that meters every token
- Prioritise fewer, high-ROI use cases over broad, unfocused adoption – usage-for-its-own-sake (“tokenmaxxing”) and citizen-built micro-agents inflate spend without proving value
- Point AI at expensive work, not annoying work – high labour, delay, error, churn and compliance costs, rather than email drafts and inbox-watching
- Treat AI as a capital investment requiring governance, business cases and phased rollout – tie each use case to a measurable outcome with a real return threshold (~2.5–3×), tracked in 3–6 month increments; if it only works while AI is cheap, don’t build it.
- Ensure AI delivers measurable productivity gains versus the existing human or system cost it replaces
- Embed AI into end-to-end workflows (agents), not standalone tools – and build an orchestration layer that routes each task to the right-sized model for cost and quality
- Make AI a cross-functional, executive owned decision and use it to reshape enterprise cost structures beyond IT, including services and operations
- Maintain vendor flexibility as pricing, competition and market dynamics continue to evolve
A strong AI strategy today is less about rapid adoption and more about disciplined investment: fewer, tightly governed use cases aimed at genuinely expensive problems, measured like any other capital decision, and architected to survive real pricing. In a rising-cost environment, the winners won’t be the firms that deployed the most AI, but the ones that can prove what each dollar of it is actually worth. , ,
We’ve seen this film before
For those of us old enough to remember, and work in, the Dot Com era, the current moment is familiar.
In the late 1990s, the internet was going to make every existing business obsolete. Profits were for the unimaginative. Get big fast, worry about the model later, and whatever you do, don’t get left behind. Webvan burned through the better part of a billion dollars delivering groceries before collapsing; Boo.com torched around US$135 million in eighteen months selling fashion online; Pets.com went from a Super Bowl ad and a beloved sock-puppet mascot to liquidation inside nine months. Beautifully funded, confidently marketed, and gone. The internet turned out to be every bit as transformational as promised. It just didn’t arrive on the timeline, at the prices, or through the companies that the hype machine insisted it would.
That’s the part everyone forgets. The technology was real and many of the businesses built to ride it failed. Both were true at once. I see no reason to believe AI will be different on either count.
So what should you actually do? (FOMO is not a roadmap)
I understand the impatience. The fear of being left behind is real and human. But fear is a terrible architect. Here’s the balanced posture Fin365 is taking, and the one I’d urge on any firm:
- Own your foundation and your data – Clean data, defined process and connected systems, sitting in an environment you own and control rather than one you rent. This pays off in every version of the future, AI or not. It’s the one bet with no downside.
- Make AI-readiness the goal, not AI features – AI is only as good as the data beneath it. Get the foundation right and you’re ready for whatever arrives.
- Reserve AI for genuinely expensive problems – Stop pointing it at annoyances. Point it at real labour, delay, error or compliance cost and only once the process underneath is sound.
- Price every use case at real cost – If it only works on subsidised tokens, it isn’t a use case yet.
- Separate the capability from the company – Be early to the capability and patient about the vendor. Adopt AI through partners with proven economics, transparent data handling, genuine longevity and the ability to scale; not whoever has the loudest launch.
None of this is anti-AI. Fin365 is investing in our AI capabilities and partnering with others. It’s anti-recklessness. The monolithic, all-in-one system was built for a world that no longer exists; modern advice firms need a foundation they control and an ecosystem they can build on at their own pace.
The bottom line
Working in the DotCom era taught me a valuable lesson: a real technological revolution and a speculative bubble can occupy the same room at the same time. The internet changed everything. But it took longer than the early hype suggested and there were many casualties along the way.
AI is no different. The opportunity is real, but so is hype. There will be casualties for those who dive in face first, without first testing the waters.
Your job is not to choose between “all in” and “left behind”. Your job is to be deliberate: own your data, interrogate the economics, fix what doesn’t need AI, and partner with companies that will still be standing when the music stops and the bills come due.
The agentic future is here, and it will reward the firms that engage with it. But the firms that win won’t be the ones who moved fastest. They’ll be the ones who inform themselves of the realities, act deliberately and hasten slowly.
Sources
[1] What Are AI Data Leaks? Causes, Risks, and Prevention
[2] AI Prices Are Going Up, Up, Up – And What This Means For Enterprise AI – JOSH BERSIN
[3] Why the AI Boom Is Running Into a Cost Reckoning
[4] If Your AI Use Case Cannot Survive Real Pricing, Stop Building It Now | LinkedIn
About Stephen Handley
Stephen Handley is the founder and CEO of Fin365, a Microsoft-native operating foundation for financial advice firms across Australia, built by a practising adviser for his own practice.
Fin365 runs inside a firm’s own Microsoft 365 and Azure environment, so client data stays something the firm owns and controls, not something it rents access to. That architecture is also Fin365’s view on AI: enthusiastic about the capability, deliberately conservative about the economics, security and longevity of the companies behind it, and focused on getting the data foundation right so advice firms are genuinely ready for whatever comes next..
Media Contact
E: [email protected]
