The conversation around AI in investment banking often centers on selecting the right platforms, models, or vendors. However, real impact does not come from choosing the most advanced technology, it comes from identifying where AI can consistently create value within existing workflows.
But there's no sense in asking which tool is best before you know exactly what you want it to be best at.
When we began working with a bulge bracket bank on deal origination workflows, we set aside the platform question and instead mapped each step to identify which were high-volume, time-sensitive, and standardized enough for AI to handle. This exercise narrowed the focus to three steps: company profiles, earnings summaries, and newsletters. Not the entire workflow, but three targeted steps. We deployed AI across those areas, established governance around output, and measured outcomes. As a result, average handling time dropped by 20%, while monthly output increased by 26% without adding headcount. The freed-up capacity went straight back into expanding deal coverage.
Together, these results reinforce that impact comes from disciplined execution within workflows, and here are five key takeaways from this case.
1. Workflow Design Over Model Sophistication
Our work with the bulge bracket bank highlights a critical principle: success in AI adoption is driven by workflow discipline, not platform selection. The results came from knowing which steps to start with. And in practice, none of the most powerful frontier models were required as lighter-weight models handled the tasks effectively. It wasn't the model choice that drove success. It was the plumbing of the workflow.
Plumbing is unglamorous and often invisible work. When done right, systems run smoothly, outputs stay clean, and nobody thinks about what's underneath. You only notice when something breaks. That's how well-deployed AI should function in a bank's operations and why the infrastructure beneath it matters more than the technology on top.
2. Categorizing Work by AI Suitability
A key enabler of our workflow first approach was the categorization of workflows based on their suitability for AI. In other words, which parts of the workflow are best suited for AI-driven execution?
We found that standardized, high-volume tasks delivered the most value, often achieving efficiency gains of 25–40%. More complex tasks requiring judgment, such as industry analysis, benefited from partial AI support, while highly judgment-driven activities like financial modeling and pitchbook story lining largely remained human-led.
This structured prioritization ensured that AI was deployed where it could deliver consistent, scalable impact rather than being applied uniformly across all processes.
3. Enabling Scale Through Standardization
Another important factor to the success of the bank’s AI implementation was the emphasis on standardization and scale. Instead of allowing fragmented experimentation across teams, the bank adopted a “one use case, one tool” approach—testing different tools for each workflow, selecting the best-performing option, and deploying it consistently.
This prevented “agent sprawl,” where multiple disconnected tools create inefficiencies and limit enterprise-wide value. By standardizing tools, prompts, and processes, the bank ensured that outputs were consistent, measurable, and continuously improving.
4. Governance as a Foundation for Quality
Governance played a central role in enabling AI adoption at scale. Given the high-stakes nature of investment banking deliverables, every AI-generated output required strict quality controls, including human review, prompt versioning, and full traceability of sources. Governance frameworks were embedded from the outset, ensuring that increased output did not come at the cost of accuracy or risk exposure. This approach allowed the bank to scale AI confidently, knowing that quality and compliance standards were maintained.
5. Driving Accountability and Measurable Outcomes
Finally, the model relied on clear ownership and accountability for output. Rather than simply enabling analysts to use AI tools, the workflow ensured that accountable teams validated and delivered the final outputs. This shift—from tool adoption to outcome ownership—was essential in translating AI capabilities into tangible business results, including reduced turnaround times, increased throughput, and higher productivity.
Conclusion
In summary, the experience of the bulge bracket bank illustrates that the primary challenge in AI adoption is not the technology itself, but how it is deployed. Banks that succeed will be those that take a disciplined, use-case-driven approach: identifying the right workflows, embedding AI within standardized processes, enforcing governance, and scaling through consistency.
Ultimately, sustainable impact comes not from the tools a bank adopts, but from how effectively those tools are integrated into production to drive measurable, compounding gains.
