Originally published on LinkedIn, John’s Corner is where Fortimize CEO John Hamon shares the perspectives, challenges, and hard truths he sees from the front lines. View original publication.
The banks that will win with AI have already fixed something most banks have not. It isn’t their data stack, their vendor, or their CEO’s ambition. It is how they absorb new technology, and most banks are using a method that has been failing them since long before AI arrived.
I lead an organization that specializes in Salesforce implementations for banks and credit unions. Most of the institutions we work with come to us after a previous attempt stalled.
The vantage
A top-75 credit union came to us three and a half years ago with a failed branch rollout, a data model an earlier partner had mangled, and an executive team that had stopped trusting the platform. Today it runs Data Cloud, Agentforce, and a next-best-action engine that pulls external credit data to surface members carrying mortgages and car loans at competing institutions.
A mid-sized Midwest community bank came to us seeking to realize more from its Salesforce investment. Its dealer onboarding cycle went from six weeks to two weeks within the first year of our engagement.
A large credit union we inherited had budgeted $900,000 for year-one professional services with a prior partner and been change-ordered to nearly twice that, with no complete customer view to show for it.
Across these accounts the product was the same. The delivery model was not.
The gap
Three months ago we were brought into a conversation with a roughly two-billion-dollar bank that has been running Salesforce for five years and just renewed. The president had concluded, aloud, that the platform is a glorified data and case management system. Five years in, the team could not produce the ROI slides the original sales cycle had promised. Seventy-five percent of managed services hours were going to the loan origination system. Salesforce itself was in break-fix mode. The bank was now evaluating Agentforce and Data Cloud for the next budget cycle.
I have had this conversation twenty times in the last year. Another client is halfway through a data warehouse implementation, cannot articulate what business decision the data will inform, and is already asking about Agentforce. Their marketing team builds segments by hand in a spreadsheet. Another client’s leadership cannot explain why Salesforce has not delivered efficiency gains after four years, but wants to add AI to the roadmap anyway.
Big implementation. Change-order treadmill. Partial go-live. Adoption falters. Platform tagged as underperforming. Leadership moves on to the next initiative. AI is now that initiative.
The mechanism
Banks are slow and vendors oversell. Neither explains why the same bank, with the same vendor, can go from red to flagship in three years under one delivery model and stall for five under another.
The default model in enterprise software is the big-bang implementation: a twenty-four month statement of work, a business case that promises outcomes at go-live, and a change-management plan that almost nobody executes—because by the time training begins, the project is eight months late and leadership has already started asking where the ROI is. A feature rolls out that does not fit how a branch manager actually works. She builds a workaround. The workaround becomes the system. Salesforce becomes the place where information goes to be ignored.
Drop AI on top of that. An agent needs clean data, stable workflows, and employees who log their interactions consistently. None of those exist in the environment the big-bang model produces. We set the agent up to fail. Its recommendations draw from household records that exist in three versions across three systems. Its suggested actions reference workflows the branch manager stopped using a year ago. The pilot underperforms. Leadership concludes AI is not ready. What was not ready was what we handed it.
The way out of this is not a better agent. Run quarterly commits instead of multi-year SOWs. Embed change management inside sprints rather than staging it as a post-launch training event. Build the system around the workflow the branch manager actually uses, not the one the org chart says she should. Treat adoption as a design problem. Structure data in small, bounded batches tied to specific decisions, not in unbounded “let’s bring everything into Data 360” projects that produce consumption anxiety and nothing actionable.
I have watched the same institutions, with the same people, produce opposite outcomes the second time through.
The reframe
The productivity gains from AI are real. They are already showing up at institutions that fixed their delivery model before AI got hot. The credit union triggers next-best-action offers to members carrying loans at competing institutions. The community bank cut dealer onboarding from weeks to hours. These institutions will look like they got lucky on timing. They did the unglamorous work of rebuilding how they deliver software before they went looking for AI to deliver.
The institutions that did not do that work will continue to underperform, and the pattern will be obscured because the underperformance cannot be pinned on AI specifically. It will look like ordinary operational disappointment. The CFO dashboard does not not carry an AI line item. The efficiency ratio will drift higher, and the AI investment will dissolve into the general category of technology spend that did not move the needle. Nobody inside the institution will say out loud that the problem was never the AI, because rebuilding delivery is less interesting to a board than announcing Agentforce.
The implication
The headline story in banking tech is all about adoption rates, flashy AI pilots, and bold earnings-call announcements. But the banks actually pulling ahead have largely stopped announcing it.
The big idea is this: AI’s value doesn’t come from bolting intelligent agents onto a broken foundation. It comes from having already rebuilt the organization’s ability to absorb, adapt to, and consistently extract value from new technology. Institutions that mastered incremental, workflow-centric delivery, and which treat adoption and change as ongoing design problems rather than one-time events, now find AI acts as a natural multiplier on a functional platform. Those that didn’t are simply repeating the same stalled pattern, with AI as the latest chapter.
The quiet differentiator isn’t who deploys Agentforce or Data Cloud first. It’s who built the muscle to make those tools (and whatever comes next) actually work in the real world of branches, dealers, and member relationships. Without that, AI investments fade into the background noise of “technology that didn’t move the needle.” With it, the gains compound—and the winners pull further ahead without needing to shout about it.
The question bank leaders ought to be asking their partners is not “can you help us deploy AI?” It is “how have you changed how our organization absorbs new capability?” That question gets answered badly almost everywhere. When it is answered well, AI becomes the next increment on a functional platform instead of a desperate move to extract value from a stalled one.
What’s been your experience with major technology implementations in banking or credit unions? Did the delivery model help or hurt more than the technology itself?
