AI in Modeling Test / Case Study
Currently interviewing for a lateral Associate role at a few MM/EB banks, and I’ve been told the next round will likely be a modeling/case study.
I have solid modeling experience and understand the concepts well, but in my current role I rarely have to build models completely from scratch. Most of my day-to-day is modifying or expanding existing models.
From what I’ve heard, the test is virtual. They’ll email over the case, give you a set amount of time, and then walk through your model afterward.
My question is: what’s really stopping someone from using Claude (or another AI tool with Excel integration) during the exercise? Assuming you understand the model well enough to explain every assumption and walk through the output, how are firms actually controlling for that?
I’m not looking to game the process, just genuinely curious how banks are thinking about this now that AI is becoming so capable.
Also, for those currently at MM/EBs or BBs, how many of your teams are using Claude for Excel in your day-to-day? We use it fairly often at my current shop, and it’s been a massive productivity boost.
Firms are increasingly aware of the potential for candidates to leverage AI tools like Claude or ChatGPT during virtual modeling tests or case studies. However, there are several ways they attempt to mitigate this and ensure the integrity of the process:
Time Constraints: Many modeling tests are designed with tight time limits, making it difficult to rely heavily on AI tools. Even if you use AI, you still need to understand the output, make adjustments, and ensure everything aligns with the case requirements.
Walkthroughs and Explanations: After the test, firms often require candidates to walk through their models, explain assumptions, and justify their approach. This step is critical because it tests your understanding of the model and your ability to articulate the logic behind your work. If your explanations don’t align with the model or seem overly rehearsed, it could raise red flags.
Custom or Proprietary Data: Some firms use unique or proprietary datasets that AI tools may not handle well without significant manual intervention. This ensures that candidates must rely on their own skills to interpret and model the data.
Focus on Problem-Solving: Many tests emphasize problem-solving and critical thinking over just building a model. For example, they might include ambiguous or incomplete data, requiring you to make reasonable assumptions and demonstrate sound judgment.
AI Detection Tools: While not yet widespread, some firms are exploring tools to detect AI-generated content or unusual patterns in Excel files that might indicate external assistance.
As for day-to-day use of AI tools like Claude in MM/EB/BB banks, adoption varies by team and firm. In some cases, AI is being used to automate repetitive tasks, improve efficiency in data analysis, or assist with Excel modeling. However, its use is often limited to non-critical tasks, as firms remain cautious about data security and the accuracy of AI-generated outputs.
If you're preparing for a modeling test, focus on sharpening your ability to build models from scratch, explain your assumptions clearly, and demonstrate a strong grasp of the underlying financial concepts. Even if AI tools are used in the workplace, firms still value candidates who can perform these tasks independently.
Sources: Will robots replace your consulting or financial career?, Case Studies & Modeling Tests: How Exactly Are They Administered?, JP Morgan Ending Campus Visits, Google partners with Goldman Sachs in automating Investment Banking, Q&A: Corporate Banking to FAANG CD
Have you actually tried to use AI for a case study model? Claude on the max tier if you give it the exact case and your explicit directions still spits out juvenile trash, but if you haven’t built client facing models before you wouldn’t be able to tell the difference.
One of the small things that AI can’t do yet is build or modify actual sensitivity tables, so something I’ve seen in Junior models that is an obvious sign they just had a LLM do it is that instead of an actual data table that refreshes automatically, it’s like a basic 4x5 table with *each individual cell in the table manually calculated*. There’s probably other examples but by and large I think the quality of the case study should bar you from getting far with AI usage in of itself on the generation side.
I do think AI usage is great and encouraged for speeding up things like excel formula recall (there’s no valor in trying to remember to use SUMIFS vs SUMPRODUCT), giving it your final model to double check all the formulas are accurate (AI is actually very good at this associate type of task), helping you get up to speed on industries you know less about (ie building a SaaS model coming out of a oil and gas group).
I do agree that using Claude is not amazing at client specific models like a revenue build or expense build. But idk about you, I think it’s fantastic at cleaning data and arranging it in the way I want. I think it’s actually pretty fucking good at PowerPoint too, when I’m building a CIM. My VP likes the work. What do you think AI is bad at from what you’ve seen? But very helpful insight overall man!
Can confirm that AI is actually not great for excel modelling, which is why i regularly see AI companies on linkedin hiring bankers for 100+ an hour to do models lol.
The worst part of it is it all seems fine until you look at the details and you notice it can't be used. Experienced people will definitely know if they audit it closely. Example would be a formula that's super long that technically works, but best practice is to add more rows and make the model easier to read.
Tenetur sapiente molestiae ea eligendi voluptatem vel. Sunt et corporis est. Voluptas deleniti nemo error minus. Sit porro vitae incidunt est voluptatem molestiae. Quod et enim temporibus omnis ut qui consequatur.
Aliquam consectetur cupiditate nostrum quisquam provident quam. Rerum explicabo culpa quibusdam repellat incidunt necessitatibus facilis et. Mollitia necessitatibus blanditiis odio tempora ut qui. Ut et dolores dicta nostrum similique reiciendis. Corrupti modi laboriosam atque vel voluptatem blanditiis et. Officia iste fugiat est.
Consequuntur ex quis provident ut accusamus. Molestiae pariatur ea et iusto delectus commodi.
Reiciendis atque quia autem. Sed unde ut tenetur hic minus. Perspiciatis pariatur neque et qui sit nihil voluptas. Consequatur quis maxime et fugit quia aut. Vitae vel sit nobis et et.
See All Comments - 100% Free
WSO depends on everyone being able to pitch in when they know something. Unlock with your email and get bonus: 6 financial modeling lessons free ($199 value)
or Unlock with your social account...