Council Post: The AI Budget Hangover: Why Banks Need Forward Deployed Engineers
Lukas Haffer is CEO of CASCA.
getty
Large enterprises are adopting AI quickly, and banking is no exception. The promise is obvious: more revenue, lower costs and better products for happier customers and employees.
But another reality is emerging. Many organizations are asking, “Are we getting value? Are we spending efficiently? Where can AI actually be helpful?”
Those questions are increasingly relevant because the worst-case scenario is no longer theoretical. A company gets excited, gives everyone a generic AI tool and hopes productivity will self-organize. Then the bills come in, the token budget is gone and what the organization has to show for it is not transformation. It is noise.
In some companies, AI has become a way to produce more artifacts without solving more problems. Every meeting has a summary, every summary has follow-ups and those follow-ups become emails that do not quite make sense. The danger is that organizations spend both salary and the extra AI money while less quality work gets done.
AI Activity Is Not Business Impact
Banking leaders should be especially careful not to mistake AI activity for business value. A model can generate a 400-page credit memo in seconds. But if no one wants to read it, and it does not help the bank make a better decision, nothing has really improved. The organization has simply created 400 pages that someone else’s AI now has to review.
The good reality is much more measurable. For a bank, AI should answer practical questions: Are we processing more loan applications? Originating more loans? Moving customers faster? Improving Net Promoter Score? Helping employees handle more volume without burning out? Giving underwriters more time to analyze the business instead of typing from PDFs into Excel?
These are the outcomes banking leaders should care about. AI should not be judged by how many people have access to a tool or how many documents it can generate.
The Diffusion Dilemma
The reason many AI deployments fail to create value is what I think of as the diffusion dilemma.
The models are really smart. They can do a lot. But the people who know the models often do not know where that capability needs to diffuse, while the people who know the business problem often do not know how to leverage the technology.
In banking, that gap is especially important. A banker may want AI to write a credit memo, but the model cannot do that well without the right information gathered from credit reports, application data, financial statements, customer information and dozens of other sources assembled at the moment the model is asked to work.
That is not underwriting. That is not useful. And in a regulated, high-stakes environment, it is not enough.
Why Forward Deployed Engineers Matter
This is where forward deployed engineers become essential.
A forward deployed engineer does not build technology in isolation and hand it over to the customer. This engineer works directly with the client, observes the problem and prototypes a solution in context.
This is different from giving every client the same cookie-cutter product and asking them to adapt. The engineer engages with the people experiencing the problem and helps build something that actually solves it.
In banking, forward deployed engineers serve as a direct intermediary between what the technology can do and what bankers need to solve for customers. They sit with loan closing officers, learn eligibility rules, understand what underwriters need to review and figure out what information must be pulled before the model is asked.
In lending, banks often receive hundreds of pages of loan documents. Someone must extract financial and customer information, update the system of record and prepare analysis for an underwriter.
AI can help, but only if it is implemented in the workflow correctly. The value is better extraction accuracy, more time for underwriters to analyze the business, faster customer turnaround times and more time for loan managers to respond directly.
Improved extraction accuracy, cycle times, customer speed through the process and whether employees can handle more volume with the tools they have been given are all measurable outcomes.
The Best Engineers Are Empathetic Generalists
The forward deployed engineering model works when engineers are technically competent and curious about the industry. They have to be empathetic generalists, excited about learning the problems of banking, not just showing off what the technology can do.
Their role is to observe, prototype, test and measure whether AI is causing positive business impact by providing more revenue, higher profit, lower cost, better customer satisfaction and better employee experience.
Banks Also Need Cost Discipline
AI costs can spiral quickly. If organizations procure AI directly from foundation model providers, they need to understand the incentive structure. Those providers make money when customers burn more tokens.
As AI usage grows, banks will need more intelligent routing. Not every task requires the smartest, most expensive frontier model. Some tasks simply need a model smart enough to compare whether two addresses match.
For repeated internal use cases, more companies will likely use open-source models or routing layers that choose the right model for the right job at the right cost: enough intelligence to solve the task efficiently.
Banks do not need a vendor incentivized to maximize consumption. They need a trusted AI partner focused on outcomes.
Real Outcomes Over Hype
The next phase of AI in banking will be defined by who turns intelligent models into measurable improvements in real banking workflows.
Forward deployed engineers are the bridge between two realities. On one side is the AI budget hangover of money burned, tokens spent, documents generated and little value created. On the other is AI that helps banks process more loans, serve customers faster, improve underwriting quality, reduce manual work and control costs.
The difference is implementation. Smart models need the right context, data, workflow and measurement. Forward deployed engineers bring those pieces together.
For banks, the goal is not the AI hype cycle. The goal is real outcomes. That requires more than buying tools. It requires a team close enough to the business problem to understand what needs solving and technical enough to build it.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?