How Has Your Team Actually Been Using AI (ChatGPT/Copilot?) in IB? What’s Been Game-Changing?
Title says it all. Curious how teams at all levels are actually using AI day-to-day in their workflows. Assuming most banks have access to some version of ChatGPT or Copilot. Has anyone found real day-to-day value yet beyond the chat functionality and implemented real benefits in excel / powerpoint?
I know everyone talks about how “AI will transform the way we work and change the industry”, and I agree it will over time, but I’m trying to get a read on what’s been legitimately useful in practice today for deal teams. I’m generally curious how people are using this at all levels in banking.
Bump, I’m curious about this too. Capiq and factset have been around for years and I still needed to verify data. I think with generative AI, theoretically there will be a day one day when it can draft CIMs and S-1s but maybe I’m just technologically challenged that I can’t get AI to spit out decent analyst type tasks.
Don't get me started on how terrible FactSet is
Curious where it lets you down?
Copilot honestly sucks. gpt is great but a lot of shops have it blocked
This is similar to my experience. So maybe I’m answering my own question here, but then how am I supposed to use AI to be more productive? I concur chat gpt is better but if it’s blocked I can’t use it much less try and use it to create PowerPoints or spreadsheets. And I can’t get copilot to do basic things.
On a broader question, for all the chatter about investment banks replacing people with AI, the ground level experience seems to be AI is nowhere near ready to replace people and automate tasks. So is this just a smoke screen to implement layoffs and slow hiring due to lower deal volume (at least in terms of # of transactions, buoyed by $ amount via mega deals) but investment banks are citing AI as a handy excuse?
not much u can do if gpt is blocked (mine is too). found copilot to be somewhat helpful for reading over files, writing emails, and sometimes converting data to excel (still not great thought)
+1
Agree I've found Copilot to be a watered down version of GPT
In trading, so not 100% the same, but I find it decent for sentiment analysis and looking at a bunch of articles quickly for details and general sentiment about where something can move. It's also good for helping me explain complex ideas in a way that is more simple(I work with derivatives and complex products/models, so breaking these down to many people is hard). I've found that chatgpt and copilot are subpar compared to Claude and Gemini. Gemini's new model is a lot better than anything I've seen openai release.
When I interned at a top 3 BB, I noticed Analysts using ChatGPT to answer questions they had even though it was banned. They sort of used it as a google on steroids to verify information, find sources for things like KPI slides
The common thread of the answers so far is AI isn’t anywhere near eliminating Analyst work. Maybe a tool comes out tomorrow that ages this thread immediately, but not yet.
When used right, it is time-saving for research or small profile tasks etc
But it's not gonna crank out CIM's for you. My view is that AI won't fully replace jobs, but the person who uses AI effectively, will
Worked at a buyside firm but advanced GPT can be useful for initial broad research on industries if you prompt it right
From a SecOps perspective - I worked at one of the largest security operations tech companies, with clients including Nasdaq, BNY Mellon, Regions, and Bridgewater. The most scalable solution for banks isn't blocking AI outright - it's working directly with model providers to develop on-prem deployments that satisfy SOC2, FedRAMP, and data governance requirements.
The approach that actually works (for pretty much every tech deployed in heavily regulated industries): Run models like Llama or Mistral locally on user workstations behind the firewall with strict access controls, then layer in a proxy/gateway that pre-filters, anonymizes, and audits any prompts sent to cloud models. This isn't bleeding-edge stuff - tech companies that already navigate FedRAMP and GovCloud have been implementing similar architectures for years (which is probably why some firms allow Copilot or Gemini but block ChatGPT).
The reality: Most banks are working on this infrastructure now, but it takes time. Implementation and integration with institutional data will likely be a meaningful chunk of the $2-5T in tech spend these firms are planning over the next decade. So the current state - blocked tools, clunky Copilot experiences - is temporary. The firms that move faster on secure, on-prem AI deployment will have a real competitive advantage in productivity over the next 2-3 years.
Other than that, using your preferred model on your personal laptop or phone, in a redacted type of prompt. Youre dabbling at your own risk though.
How should cloud based vendors navigate this to get their software compliant with bank’s needs? Is on prem deployment the only truly safe option or do they have to get FedRAMP to truly prove they are secure or? Trying to understand how we could evaluate current vendors who maybe have Soc 2, iso 270001, ccpa, but not fedramp for example
Really thoughtful question—thanks!
It depends on the bank’s risk appetite and use case.
The compliance ladder:
• SOC 2 Type II + ISO 27001 = baseline for any vendor touching financial data (this is like, onboarding intro material for SecOps)
• FedRAMP Moderate/High = gold standard, but expensive (12-18 months, $2M+ for authorization)
For cloud AI vendors:
On-prem is safest, but not the only option. Cloud vendors can be compliant with FedRAMP + proper data handling (encryption, zero retention, audit logs (why Splunk-Cisco has a huge value covering these solutions)). The real risk is data leakage through training or logging - a vendor can have FedRAMP but if they’re using your prompts to improve models, that’s a deal-breaker.
Quick evaluation framework:
1. SOC 2 + ISO = table stakes
2. FedRAMP Moderate minimum for anything touching MNPI
3. Contractual guarantees on data handling (no training on client data, deletion policies, DPA)
4. Architecture review: prompt logging, data residency, access controls
Some banks accept cloud vendors hitting all of the above. Others require on-prem regardless. Comes down to their compliance interpretation and what data you’re feeding it.
ChatGPT & co are all dogshit and retarded
Geminis new update is extremely helpful
del
Email drafting, Excel cleanup, going through filings, and putting together comps are already being streamlined by tools like Grok and new features firms like BamSEC are testing. Banks and funds are also piloting tech like Anthropic Enterprise that can build models with minimal human input—their demo was pretty impressive, delivering DCF and LBO models within seconds of prompting that are 80-90% ready and just needs minimal finishing touch from a real person, which all models need anyway. OpenAI’s pushing similar ideas with Project Mercury.
Decks and audits will likely follow as teams test AI tools for presentations and financial reviews.
So the junior workload is getting lighter—but does that mean more deals per analyst, or just fewer jobs? And once autonomous agents show up, what happens to mid-level roles like VP/Director?
Its an interesting point. Conversely though if everyone in the future has access to the same high quality AI that’s producing quality models, decks, and analysis, won’t that maybe put more emphasis on Directors/MDs who can bring in business?
This is very good question... What would be the value added (e.g. for the seller) from engaging an IB team at the various levels?
Basic stuff: (LLM? Or better) models + in-house m&a
Outreach: (LLM?) models + perhaps the rolodex of an MD
Process: in-house m&a + m&a advisor (also depending on complexity and needs)
Negotiation: [open for suggestions here]
Structuring etc: I suppose for plain vanilla models, more complex stuff lawyers, tax and m&a advisers
Funding (for buy side): balance sheet bank or resit with AM
I know, overly simplistic, but just highlighting that this may disrupt more than juniors shuffling decks...
What do you guys think?
I find Claude to be quite useful in analyzing strategic acquirers as it grabs what they say publicly and can pull in user feedback to identify gaps in the product stack. Thinking through those gaps used to be basically reviewing Gartner reports and talking to corp dev teams about their focus areas. I’d say those types of questions are well suited for genAI.
That said, there is a whole new wave of companies like Hebbia that are targeting both sell side and buy side. I except tools like those to get very good and also take the regulatory steps to be incredibly helpful on analysis without putting client data at risk. Will still require human verification but if it can get to a rough draft within minutes then it saves a ton of time on the initial turn.
I don’t think we are at the “game changer” moment just yet but it’s going to happen.
.
At Cent they give us enterprise ChatGPT accounts
Used it for strip profiles and data pulls. Nothing that requires detailed scrubbing, but for CS decks where an MD wants a bunch of strip profiles and strategic rationale for potential transactions found it very helpful to lay out to Gemini what fields to populate and for what companies and have gotten decent results. Still need to review but imo comes out good quality and saves time.
Now… if my MD knew that I’m sure they’d just request more strip profiles. Banking always seems to create work so even with agentic AI feel like banking will still find a way to make juniors work 80-100 hour weeks
PE professional here - find it super useful for generating expert call agendas, summarising notes and helping distill ideas into IC docs (we use word so have to write in prose). Generally a long way off in terms of anything numeric - we had a trial of mosaic; interesting tool, but I feel like most models are bespoke and tend to be iterative, so it tends to lose its utility. Have found AI generally to cut down a heap of time on mundane tasks that were previously very time consuming.
JP Morgans internal GPT "LLM" is so awful. Waste of billions of dollars. Why try to beat ChatGPT etc? ChatGPT is blocked on the company PC so analysts resort to using chatgpt on their personal laptops/phones
GS’ AI tool isn’t great either. Gemini we can use within the chrome web browser and that far surpasses what GS’ internal AI tool is able to do
That’s nice we “used” to have access to Gemini early on until our IT folks caught on. Now we just have shitty Copilot
We recently got rogo. Some of the boomer MDs thought that Rogo could automatically produce RFP's or IM's in one-go, but for slides, it's atrocious.
It is good for research and small nitty tasks that can sometimes take a while if there's limited information (ie strip profiles, benchmarking etc), but would say it's below ChatGPT 4/5. It can also pluck facts or metrics out of thin air, so whilst it's good it sources most of the data, it can still be wrong. It is however good at going through annual report's if you're looking for a specific metric etc.
I was prompt engineering and using AI before Ai was a thing. It's called making the outsourcing team scrub your comps and draft up slides
It's a productivity booster but to me i dont see it as a game changer at all - there are a lot of tasks simply too complex. Realstically I dont see this affecting jnr headcount - jnrs are already pretty busy and are a bottleneck on a lot of things - we'll probably see anlaysts just as busy with more expectation to use ai tools to get it done no excuses
How are juniors considered a bottle neck? I'm trying to choose between computer science and finance so im just looking into how automatable the tasks are in both fields.
Always plenty of work to do never enough people to get it done
We have access to Gemini. Can be pretty useful for administrative work (finding emails for buyer lists, desktop research, buyer profiles, etc). Not yet useful to the point where it does meaningful work on models / any analysis requiring data or data interpretation. I think the most useful function I’ve found so far is in CIM creation… I’ll feed it information provided by the company, then give it a screenshot of shelled out PPT slides, and it’ll generate all the text to fill the space. Saves a considerable amount of time
Most teams find AI useful for research, drafting emails, summarizing notes, and pulling data for decks or comps. Tools like ChatGPT, Copilot, Gemini, or Claude save time on repetitive tasks but aren’t reliable for full financial models or CIMs. Adoption is limited by IT restrictions and compliance concerns, so juniors still handle the bulk of detailed work. Real game-changing use likely comes from secure on-prem or integrated AI that can handle more complex workflows.
We actually had Rogo come in to do an hour in-person crash course. They are great salespeople and the fact that they are founded by an ex-Banker (the Rogo rep doing our session was ex-IB too) means that they clearly know it's use in applying it to time-save for day-to-day tasks. Boomer MD's were fascinated but would echo the sentiment - it's great for profiles, research tasks etc - but it's still got a lot of work-ons to do in automating slides etc. That said, if they hold true to the updated and modifications they told us about, it could be very good - but it won't replace juniors, just make them more productive
Omnis ut quaerat sint itaque beatae at illo. Ad saepe qui dolorem est sunt dolor ut qui. Temporibus ad aliquam sed quia velit quaerat. Reprehenderit ratione aut aliquam ut vero.
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...
Sint et dignissimos sunt omnis. Expedita ducimus deserunt ea non reiciendis iure. Enim non amet suscipit est. In quo aliquam mollitia voluptatem expedita. Voluptates consequatur voluptatibus ullam amet voluptatem tempora ratione.
Libero soluta doloribus voluptate debitis omnis consequatur id occaecati. Neque enim odio harum hic. Fugit labore et numquam deserunt delectus blanditiis.
Facilis ratione necessitatibus facere adipisci et quisquam. Nesciunt nesciunt eligendi autem et aliquid et. Rerum eum aut hic vel consectetur. Voluptas consequatur nisi corporis nulla sequi. Fugit culpa nam dolores repellendus quam doloremque ut.