Q&A: AI will automate many roles in the IB/PE world. A live Q&A with Arctic, who are recruiting finance professionals to help manage that change
At Arctic, we have been working with firms in the Financial Services sector over the past six years to implement Artificial Intelligence into their toolkit. As we are rapidly growing across the Americas and actively recruiting regional GMs, we wanted to host a Q&A session to address the common questions.
Early adopters of AI are now leading the way with functioning systems to expand their deal origination and due diligence operations. These firms are using these tools to ingest larger amounts of structured and unstructured data and output standardized reports within minutes. These systems are currently able to automate approximately 50% of these functions and are improving with every day of usage. What does this mean for the future of the Analyst role? With this competitive advantage, how will firms who are late adopters of this technology going to fare? Are there certain tasks that AI just cannot do as well as we can?
Firms are also funding the implementation of AI solutions in their portfolio companies to increase the exit valuations of their investments. Solutions involving market monitoring, sales optimization, computer vision, automation and operational efficiency are leading the way in value creation at portfolio company level. How successful and mature are these solutions and how can the buyside think about them for their portfolio?
These are the questions we answer at Arctic AI and the solutions we build. We have open roles for P&L-owning VPs in various cities across the USA, Canada and Europe. We are targeting individuals with experience in IB and PE, who are keen to help design and consult on AI solutions for the same clientele.
We are here to answer any and all questions – how these solutions work, what kinds of firms can profit most from them, as well as the roles that are available at Arctic!
Just to start things off:
Great questions!
1. We do build very customized solutions but generally, we use AI to perform research and market analysis for deal origination (AI finds and ranks companies according to a fund's investment hypothesis or an IB's market focus, in simple terms), to automate due diligence (creating dealsheets, pulling data from DRs and analyzing it, etc) and to automate processes at portfolio companies to expand margins. Super custom so lots of varied case studies here!
2. This is absolutely right - that's why we always build IP for our clients. We aim never to keep the IP to the AI models we build, essentially. We even work, on occasion, with on-premise servers, such is the sensitivity of some data. We are big believers that AI companies who sell in off-the-shelf products are only suitable for a small number of generic tasks. Other businesses with valuable proprietary data - including the training their professionals can give the AI - need to keep hold of the IP to the AI that data can build.
3. VPs sit on our global exec team and are responsible for generating revenue by pitching AI technologies to PE and IB, consulting them to a great solution, and liaising between them and our technical Hub to represent the client as we build out solutions. Most clients then go on to ask for multiple phases of work to iterate on the solution in question. Depending on revenue growth, there's a great opportunity to build out a client-facing consultancy team.
Thanks for these great questions!
I’m curious on how AI can replace a coverage banker with a Pitchbook / CapIQ login or do anything they couldn’t. I understand that you make said banker more efficient, but the marginal cost of time for a junior is $0 to a bank.
Can you explain how AI would be helpful given that most of the reason you pay an IB is because of non-public insights? (E.g. ‘estimated’ financials based on information from a prior lender presentation, or figures recorded in a bank’s CRM from a coffee chat with a loose lipped PE VP).
Ideally, I’d be curious to see an example of results this origination would pick up. Can you provide the top 3 non-publicly listed acquisition targets for a U.S. based private equity firm in the insurance brokerage space with an enterprise value of USD$1B+? Let me know what other inputs you might need, but I want to see what the AI pulls. Because when I’ve used ChatGPT for things like this, it’s just incorrect (and I am accounting for their data set being 2021 and prior).
I’d like to also say that I think AI is great and there is absolutely an opportunity here. So congrats to your team on this and apologies if my questions are too blunt :)
Yeah you can have this happen--use autoGPT, the github repo for it is floating around somewhere, that is connected to the internet and spawns threads of other AI agents to accomplish tasks. Could easily answer that question, probably build out good models for each of those targets too
Not blunt at all, thanks for the questions! To address the question regarding non-public information, AI can also be trained to read from CRMs (via API connectors just like, for example, reading from public data via a CapIQ connector) and to understand when certain information is super-valuable, unique insight (such as intel from a coffee chat). Prior prospectuses or CIMs can be read by AI and each data point stored so that, in one example from our case studies, a TAM figure for a subsector can be pulled next time we see a company in a similar space. One of the reasons that we build custom solutions is that each of these examples of private info (there are 100s more) is actually pretty specialized for a given use case, fund, bank or even desk. So the task is often to knit the data sources together and train and AI to read them in a way that's useful.
Relatedly, public but non-quantitative data - like news articles, filings, blog posts, social media posts even, CEO presentations, analyst reports - can also be read and analyzed for what are often called "soft signals". Examples could be "growth", "reputational risk", "upcoming product launch", "supply chain headwinds". And of course, supplier and end markets can be read about and assessed in a similar way. That works by training an ML model to become a "soft signal researcher", and performed at huge scale and very quickly, those can be indicators from public, but non-financial, data, including about quite small or MM businesses that otherwise have zero information about them, especially in the US.
Hope that makes sense! In terms of seeing a demo of how that actually works, we'd welcome setting that up as a demo if you want to DM us. Or, if there's interest in that from this thread over the coming weeks, we can set up some webinars for further discussion... If that happens we'll announce that on here, otherwise feel free to reach out in DM!
...Sorry, one further response / comment - we often say that AI shouldn't be seen as replacing what a human could do, but doing it at a scale that would require 1,000 humans otherwise. We often start workshops and planning sessions with a thought experiment imagining what a team would do with 1,000 analysts, for an example. So, could a human spot that a company is growing and doing something vaguely interesting if LinkedIn job titles change, there's a few changes to product lines on their website, and analyst reports about the industry / macros are positive? Absolutely - but to do that for an entire company universe at scale, when often those signals aren't binary bands from public financials and would require reading every article / report / etc, is the task we can set an AI solution.
Again, hope that makes sense!
Very cool opportunity here...encourage members to ask questions. Here are a few from me:
Does the VP position require any technical background or just an interest/passion for AI?
How did you get into this / what are your backgrounds? all bootstrapped currently with enough funds from current clients to continue or any plans for funding rounds?
Thanks!
Patrick
Interest and passion is fine for the start! We'll be pretty comprehensive in training folks on AI solutions - and what it means for AI to generate genuine value - and an Engineering background at university is helpful, but not essential.
Our backgrounds vary from IB through to big tech development to technology consulting, so a varied team. What we are really looking for now is experienced business-focused experts who can be highly credible consultants - on technical topics for sure, but with the advisory / investment capabilities to the fore.
Funding wise, we have dry powder reserved for this expansion and some motivated backers, but very deliberately no VCs via funding rounds and with the vast majority set aside from prior projects.
Thanks for the questions Patrick!
OP, are you a human or AI posting this?
I have a hunch that OP is GPT4 or Bard … the way OP explains stuff just feel very Bardy lol
OP is a human, promise! There's a team of 3 of us monitoring this thread and replying to these great questions over the coming weeks.
What solution do you have off-the-shelf for doing early stage diligence and data room processing for a PE investor?
Sorry for missing this question yesterday!
Off-the-shelf, we only really have data connectors and database tools on Day 1. We focus on use cases where we need to build custom tools. For early stage DD and Data Room analysis, the data aggregation is pretty simple, so that can get going really quickly. The ML required to actually analyze it, though, needs around 2-3 months to build a first iteration, and the same again for implementation (our other answers below might be more detailed on that topic!).
How do you decide with clients on whether to train a model from scratch, fine tune an existing model, or to use an off-the-shelf API or open source model?
What level of involvement do you often see from buyside firms? Is there usually a data scientist on the other end, or do you provide end-to-end services?
Model selection will depend on a range of factors but it's very rare that an off-the shelf model will work well without at least some customization. We often say, AI is like training a team rather than buying a piece of software from a vendor. Training a model of scratch is necessary for highly-specialized tasks. To give one example, we build an AI solution for an automotive port-co to monitor defects, using 12 cameras on top of a conveyor belt. It was the same part, always in the same position, and the defects were one of 3 things (dent, scratch and discoloration) - and still, each camera needed a custom model, so we built 12 ML models, rather than use the same model for each camera.
Hope that answers the first question!
From buyside firms, we usually need them to help us think about their investment hypothesis and internal process in a way that allows us to automate it, but it's rare that they have a Data Scientist on their end unless they are a +$1bn AUM shop. In that case, we work essentially as team aug: we come in for 6-9 months but our client is their ML team, that their client is the investment desk. When a fund has a Data Science team, we get very excited because they are clearly already ahead of the curve on the use case, but at the same time, they can sometimes prefer to keep the work entirely in house.
Let me ask the ultimate question: which role / specialization is the safest from AI ( or have the most bargaining power when the AI thing hit the desk ) ? What skillset is the least threatened by AI so that all human should focus more on ?
Roles that are relationship-based
We would probably agree with the reply, in the very long term: relationship-based roles as well as super creative roles will be impossible to automate. That said, AI is a long way from being able to fully automate most roles. We believe that for a good decade or so, humans will be writing "proficiency in AI tools" on their resumes, rather than not having jobs to apply for!
Will the analyst and associate role exist in 36-48 months? If not, when do you see both roles being completely automated?
Former IBer here - I definitely think that a lot of the work that I did as an entry-level IB Analyst role will be automated very soon. The skills that'll be required most from IB teams are genuine strategic insight, advice, deal-making capabilities, relationship skills, and network, which obviously a lot of juniors have the skills for or at least are worth training in that direction. Churn, turning slides, building models, compiling research, info packs, updating comps, etc, can already be automated even if take-up is not huge so far.
Do you think that entry-level Restructuring Investment Banking roles and PE roles will also be automated? Will HF roles be automated away?
Thanks for the Q&A! Very interesting.
(1) Can you give some guidance on a cost range for this? Upfront + recurring?
(2) Can this AI solution connect to a cloud data warehouse (ie Snowflake)?
(3) Can this AI solution connect to an internal sharepoint drive?
(4) How long does implementation typically take?
(5) How do you 'queue' questions for the AI solution? E.g. is it like ChatGPT where you ask questions, or do you (Arctic AI) have to build some sort of separate algorithm for each separate question? Meaning, if we have this solution in place and want to research XYZ, does Arctic AI have to build that additional capability or can we do it in house?
(6) Has Artic AI ever implemented revenue forecasting solutions, eg demand planning?
Great questions, thanks!
(1) We charge a project fee for the custom work, and then take a recurring fee for any components of software such as our marketplace of third-party data and ML models. Typically, clients are looking at $200k-300k plus $50k p.a. recurring costs, but project sizes vary by complexity of course.
(2) and (3) - yes! Solutions that we build will typically connect into 10-15 different services, such as data storage, data providers, CRMs, email services, knowledge management tools like SharePoint etc.
(4) Full implementation including operational transformation (process and staff training) averages 6-7 months. Again though, each project is different and custom, so that can vary.
(5) Generally speaking, we aren't building solutions that function in the Q&A format like the ChatGPT interface. Typically, the ML models (almost always more than one) will surface companies or collect and analyze research / soft-signals around a company automatically, based on public and private data, or a data room. It'll be presented in a dashboard or some other UI, or sometimes an excel file emailed around the team. To work in a Q&A format, what we would do is use a tool like ChatGPT to understand the question but what it'd be querying would be the data store / data services described above, rather than "the whole internet" as ChatGPT does now. Hope that makes sense and answers the question!
(6) Revenue forecasting, demand planning, revenue estimates for private companies with no public information - yes, we have done a lot of work in that field! Both for PE / IB and for retailers to help with warehousing and supply chain management.
Thanks again!
Who are your biggest competitors and what makes Arctic better?
Thanks for the question!
We compete with two types of firm:
(1) Consulting companies who have started advising on, designing and building AI solutions in the past 1-2 years especially, sometimes longer. Deloitte have "Omnia" for example, their AI practice; McKinsey bought a team called QuantumBlack; Accenture's Technology practice include AI expertise. We compete with those folks (and many more) with our specialism and the fast time-to-value, and sometimes we collaborate with other firms and only pick up the AI part of a project while they focus on TOMs, implementation, transformation, etc.
(2) AI companies offering an off-the-shelf value proposition. This can be very tempting for clients, especially from well-funded SaaS vendors, and there are thousands out there. In most cases, we try to inform the client that an off-the-shelf product either won't work as well, or will require a lot of customization for their needs. In some cases, SaaS products are perfect for a component of a solution, but need to be paired with something custom or implemented into a more complex overall solution. So it's not that SaaS vendors aren't valuable - just that in most cases, there's a lot of work required to get them up-and-running in practice.
What advice do you have for incoming IB Analysts? As this technology becomes more widely used will Analyst roles disappear? Roughly how long until this is implemented industry wide?
I don't think Analyst roles will disappear, necessarily; there's plenty in an Analyst's day-to-day that is about making a genuine "strategic" judgment call (let's say on how to present data rather than how to collect and analyze it) and therefore is nearly impossible to automate. But a ton of the work an IB Analyst does can already be automated and adoption is only going to increase. I do suspect that IB and PE will reduce the size of Analyst cohorts, and tasks will be much more about managing a "team" of AI researchers, in the way that many firms have an outsourced shop on call that collects data, tries to find contact details, creates slideware, etc.
Just to give one example: creating a 100 page investment memo out of a data-room-sized collection of company documents could be an "80/20" task now, although not many firms have adopted it yet. 80% of the time required to assemble this deck can be done automatically, by ingesting vast amounts of data, training AI to prepare a deck according to the firm's standards, and training AI to know what pieces of information to put on each slide. The remaining 20% then needs a human to fix errors, make qualitative judgments (quality of comps, EBITDA adjustments, etc etc), write insightful commentary or take a considered position where necessary. AI is quite far from being able to automate that final 20%, but could pretty easily perform the 80%.
I would estimate that AI is widespread in the industry by the end of this year, in the sense of it being rare for any firm not to use AI to some extent. Most funds we meet are looking into using AI for origination and due diligence, often prompted by LPs who are reading about early movers doing so. Ultimately though, much of the key drivers of a deal are human - trustworthy advice, sound strategic insight, market knowledge, informed rolodex, interpersonal skills - so I don't think AI adoption will revolutionize the industry, even if it may lead to a re-tooling of "data aggregation" and "data analysis" tasks.
Hope that makes sense / is helpful! Feel free to DM us if you'd like to have a longer conversation about that topic.
How would you say managing a "team" of AI researchers would look? Do you think the entry level roles will move to comp sci/STEM majors or would standard finance/econ students still be important?
From how I read it, these people wouldn't be coding, they'd be prompting the AI's research/screening for data
Do you have any advice on what incoming summer analysts should be doing to better prepare themselves for the analyst reduction? Should we be learning Python, R, Tableau, etc ? Should we be ready to pivot to a different subsector within finance?
Mods, stop removing my fucking posts dammit. I asked the dude a simple question. Can he not steal our fucking jobs?!
they took our jobs
Enim iusto nemo aspernatur quia quis perferendis quo. Et est aut magnam nisi sapiente eveniet. Sint quasi assumenda animi quia officia amet odio. Quod ex et id consequuntur quia eius aliquid. Ipsam quibusdam harum est in veniam. Sit id eveniet ullam exercitationem. Asperiores voluptate quisquam accusamus est.
Blanditiis ad molestiae quae modi eaque. Optio cumque libero aut libero recusandae ea eos. Quasi ut sed incidunt a. Earum unde et et voluptas alias doloribus.
Modi est enim recusandae accusamus non fugiat accusamus. Qui repellendus officiis vero corporis et deserunt sed.
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...
Ducimus repellendus accusantium assumenda aperiam tempore pariatur ea. Atque consequatur saepe facilis aspernatur. In itaque voluptate pariatur nesciunt. Dolorem recusandae quia quisquam ratione. Nihil quod et omnis voluptatem doloremque pariatur.
Magnam voluptatem dolorum repellendus. Consequatur dolore dignissimos sit aut qui aut. Et tempore ut perferendis sequi aliquid sunt. Rerum nihil nesciunt assumenda reprehenderit odit expedita. Autem molestias nostrum impedit dicta sint. Quibusdam occaecati modi magni consequatur delectus velit.
Hic sint rerum nostrum ut et sequi. Eos ea quo quis. Deleniti ea nesciunt quia omnis voluptatem optio voluptatem ex. Facere voluptates molestiae deserunt doloribus placeat impedit velit.