AI in IB
Hi, by the title this might be a bit redundant but wanted to know what people on the street have been seeing. I am a current rising junior interning in IB this summer and next summer. Recently saw Bloomberg's post on Jamie Dimon saying that JP plans to hire less analysts and more AI specialists. I am a finance major at an Ivy planning to specialize in analytics in AI.
Would love to hear what current bankers are seeing at their desks. Beyond a major push to use AI, do you think your departments are being fundamentally restructured? Do you plan to give less return offers to your summer analysts? Is a finance degree going to be worth anything in ~5 years time? Trying to gauge how to best hedge against this change and keep my job. Would appreciate any type of insight.
Bump, curious if anyone has insight on future analyst class sizes/RO rates
Mainly use Gemini and work is testing Claude. Have to say Claude is miles ahead of Gemini, more accurate, higher quality responses but probably more expensive.
What used to take me couple hours now takes 20 minutes and it'll only get quicker/better
Do you use Claude Code at all? Or is it more general prompting to complete models/research/etc
Yes! I finally got access to Claude Code through work and it's been awesome.
EB, mainly in chat-box phase still. Compliance/tech can't handle artifacts, custom agents, MCPs, RAGS yet. Not that it would matter b/c AI still can't do multi-step accounting sustainably, Or it requires so many clarification / error-checking iterations, that it ends up being faster and more accurate to build by hand (adjusting templates). Piece meal modeling and error checking is working well though. If you want to tell me I'm not using it correctly than please let me know how are you specifically getting it to not make ANY errors.
That’s really interesting — given that the models will likely improve over time and that they’ll eventually be able to host on internal servers, do you think that there’ll even still be a need for junior bankers? And as an intern this summer, would it be frowned upon for me to use AI since I haven’t exactly proved my technical skills yet? Or is efficiency being prioritized?
It’s honestly pretty weak right now at my firm. copilot and rogo but they both are essentially chatbots. It’s very helpful for legal docs, cursory reviews of CIMs, materials etc. but they are completely hopeless and generating slides, models, and useful content. So it’s essentially a second set of eyes but it’s not replacing anyone.
I think we are still 2-3 years away from meaningful AI at work. Right now it’s a convenient tool but the productivity gain is marginal. As AIs improve and can actually learn a job specific to a company, we will likely start seeing the culling of headcount. I definitely think we will get there, but it ain’t close today. Could see it being anything from 2-7 years, but with the rate of innovation AI is generating I don’t see it taking a decade. I would definitely focus my studies on AI if I had to start college today, but understanding finance will still be critical. No one is hiring someone who just knows AI without knowing the underlying material themselves
Try Claude Co work or Code , especially since claude code and cowork allow to use models on MAX effort and has special plugins while web never goes above above medium effort. Same with Codex.
I think people are underestimating their ability to build models and slides (its more prompt and reasoning than mode quality tbh - more reasoning its just building more subtask and plans and more iterations) given that most of it is pretty simple to automate with script and does not involve thinking its more so transferring knowledge. You can give it a plan and it will build many automated Python script behind the scene and execute them for you.
Sometime ago I tried building RAG on top of annual reports and did a demo of it for a hedge fund tbh I was impressed that it decently. I personally think its not hard to automate to good chunk of finance tasks, you just need budget for tokens and people to build you the things and experiment with it.
Just FYI banks don't hire a lot of americans in their tech departments its mostly offshored or people on visa.
"you just need budget for tokens" is doing some insane lifting here. I have heard several firsthand reports of Claude getting pulled after the company realized that hundreds of dollars per person were being burned before lunch by them simply asking it questions. No productivity gains whatsoever. The cost of compute is absolutely staggering. The true investment required to pull meaningful gains out of AI has to be through the roof. It will be years, in my opinion, before the systems of medium/large companies can flex enough to be truly parsed through and have any real work be automated. IDK. Who knows?
100% this
We are still in the 2nd inning or relatively early days of AI. However what should really scare people is the next generation of industry and job specific AI tools that are being built on top of popular LLMs. In a year or two they will be completed and pretty sufficient at taking on many entry and even mid-level tasks.
Only experience is with Rogo, it's getting better but still miles off the finished article. From what I hear, Claude is becoming very good
It's still miles away from producing slides and models, it can read a complex excel workbook and decompress it, but no way should it be the sole reliance
Let me chime in here with a perspective from a MBB consultant. When I read such threads, I truly think people underestimate AI because they use the wrong tools (i.e., Gemini Web UI, well its nice, but not really a gamechanger in terms of productivity).
Being at (my) MBB is great, because essentailly they approved all sorts of enterprise version of AI tools with limited oversight on fees/spending which means I played around a TON with different tools.
The current "frontier" in terms of knowledge/professional services work for me is clearly the combination of Cursor + Claude Opus 4.6/7. Cursor is essentially a software which you run from your desktop and with which you can use a variety of models. It will also have access to folders you give it access to (i.e., load in tons of presentation/pdfs/etc. to learn from/read through / manipulate).
The benefit of using it has been off-the charts insane. Using an analogy: working with ChatGPT felt like having a somewhat decent (but often very goofy) intern. It helped you but you needed to repeat instructions often, check work product meticously, etc.
Now Cursor + Opus 4.6/7 feels like having a combination of so many things. Remember that one insanely smart friend in college that helped you out on difficult problems? Yeah -- for complex analytical works its like chatting with him. Having to build a super complex market model, estimating the spend of some niche industrial verticals across different geographies? It will set you up the driver trees, pressure test them with you and point you to the most obscure data-sets on the web (i.e., Chinese price indeces of some niche end product).
It will also actively build analysis in excel yourself. Example: you need to extrapolate your model to different geographies. You load into the input folder a variety of unstructured macro-data for several geographies. Within ~5minutes, it will build you a cleanly formatted excel (looking better than 90% of corporate excels) in which it buckets countries in different archetypes, writes clear rationales as to why they land in which archetypes, and computes weighted metrics to use in your model.
This is just one example of ANALYTICAL work. The difference to Chatgpt/Gemini/etc. is to me that it stays SHARP. I never have the feeling of getting nervous that it pulls out some evidence/numbers/etc. of completely thin air. There are tons of additional "mini" use-cases .. i.e., once I had to build a operating model for a retail/grocery store chain of a CIM. The data was super messy and presented across various chapters of the CIM. It pulled me within 30 seconds a table of all historical store numbers + categories (e.g., franchise vs. owned) with a clear reference to where it have found it in the CIM.
For non analytical work (i.e., knowledge work) it is like speaking to a seasoned expert. It is great to get up to speed on industries / dynamics / understanding topics. I basically run A/B test where I first spend 1H dabbling around with Cursor/Claude and then interviewed 2-3 experts (10-20Y experience, industry C-levels, etc.) and as of now, most of the initial hypothesis/industry understanding I derived with cursor has matched ~90-95% of what experts told me later.
At this point, I am just wondering where we go with this. Of course there are several flaws and it cannot replace a well educated senior consultant/IB-assoc. The question I always have in my mind though is if it can do this NOW what can it do in 2030/2035/+?
As a management consultant I have seen a lot blue-chip corporations in which probably 40-50% of people are just doing (bad) BS work. Especially in fields like accounting/controlling where a lot of the work is crunching numbers and doing very structured/repeatable analysis I truly think that Claude is already up there doing this better than a lot of employees.
Sounds a bit doomy and gloomy. Not sure if this is the actual experience or more wishful thinking at this stage….
Look it is just what I observe. I tend to disagree with you that it is "wishful thinking". Maybe I am too much of a "glass half empty" person (you can give me that) but I personally have the feeling that a lot of people in banking/consulting are rather on the coping side by trying to talk down the usefullness of AI.
Time willt ell.
Perferendis corporis quia molestiae voluptatem rerum illum eum. Et dolorem beatae vitae excepturi nulla dolor. Laborum dignissimos ipsum velit quia impedit. Consequuntur non ducimus dolores quos est totam. Iste placeat numquam ut omnis.
Ut harum tenetur deserunt. Veniam consequatur laboriosam necessitatibus soluta pariatur aut. Maxime a consequuntur et asperiores debitis nesciunt est.
In molestias unde deleniti. Ut maiores cumque eum delectus. Aperiam aliquid natus et sequi. Sed laboriosam quia enim. Debitis aut sit molestiae distinctio aut.
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...