The Bull Case for an AI-native Investment Bank
Building the Next Goldman Sachs - at 1/10th the Headcount
Welcome to Fully Distributed, a newsletter about AI and its impact on business. Join our growing community by subscribing here:
Last year I set out to build a new investment bank, OffDeal.
Not software for banks or a fintech marketplace—a real M&A advisory firm.
When we were raising our $4.7 million seed, the most common pushback was, “Why not just sell software to Goldman Sachs?” Fair question: in our YC batch we were the only team building a service company.
This essay explains why we chose the harder path—to build a full-stack AI investment bank to compete with the large incumbents, rather than sell SaaS directly to them. The thesis is simple: M&A will always matter, but legacy banks are structurally unable to run AI-native workflows, and therefore will be unable to capture the full upside.
We’ll cover:
Why investment banking is a solid business.
Why incumbents can’t reinvent it.
How OffDeal turns AI into client value.
Let’s dig in.
I Investment Banking Is a Solid Business
The first notable M&A advisory mandate dates back to 1895, when JP Morgan advised on a $1.4 billion merger of Federal Steel Company and Carnegie Steel ($45 billion in today’s terms!), which led to the formation of U.S. Steel in 1901. Since then, global M&A has grown to trillions a year, peaking at $5.8 trillion in 2021; advisory fees now run $30–40 billion annually.
But what underlying factors make this industry such an enduring business?
Timeless need → structural demand
Jeff Bezos says it’s easier to bet on what won’t change.
In M&A that constant is clear: sellers of businesses want the highest price, best terms, least hassle—and they always will. No one has ever called Goldman Sachs or J.P. Morgan to complain they got “too much” for their business. Because these motives are permanent, banks could invest heavily in this area, confident the product would stay valuable a century later.
Demand for financial advisory is structural, not cyclical.
Third-Party Validation
Even sophisticated market participants want a neutral set of eyes when selling or buying a business. Private-equity firms—despite employing ex-bankers—still bring in another advisor because a true third party can run a cleaner auction and nudge bidders higher. Boards lean on a bank’s fairness opinion to meet fiduciary duties; founders like hearing an outsider confirm the price. Independence gives the process credibility and usually bumps the final number.
Revenue Density
M&A fees are a slice of deal value, not hours worked. That’s why top firms clear $1 million-plus per employee - numbers other professional services industries like consulting and accounting rarely see.
Even so, top-line productivity hides a margin problem we’ll address next.
II Investment Banks Are Inefficient
Despite enviable revenue per employee, banks are actually very inefficient. For every “revenue generating” dealmaker (Managing Director) there are ~10 supporting staff. The business model is so labor intensive that compensation cost alone absorbs 60–70 % of revenue. Add revenue-per-MD and operating margin and the story changes:
Part of the problem stems from the business model itself - success fees pay only on closed deals, but a much larger contributor is simply how the actual work at banks gets done.
Waterfall handoffs: Instructions flow MD → Director → VP → associate/analyst, and then gradually back up the chain. Each handoff introduces potential for misinterpretation. Two weeks in, the angle shifts and the deck restarts at page one. Thirty iterations per deck is common.
Missing context: Client calls aren’t transcribed, key emails omit half the team, side-bar chats buried in Slack/Teams. An MD who decides to change a proposed financing structure over lunch may only mention it to the deal team the night before the meeting, triggering a last-minute rebuild.
Manual & duplicative work: Even with Bloomberg, CapIQ, FactSet, teams still copy data into Excel, then re-cut it for each MD. Much of it duplicates work buried in Pitch_v32_FINAL2_FF -often faster to rebuild from scratch
Proof-of-work incentives: Page count is treated as a proxy for quality and effort. VPs add slides to impress MDs; MDs add slides to impress clients. In the end, a sixty-page deck is produced.
Taken together, these frictions explain why every deliverable consumes so many staff hours - and why labor cost absorbs most of the revenue. The root cause for the most part is cultural, not technical, which is where we turn next.
III Why Legacy Banks Will Never Change
Excel, Bloomberg, and AI slide builders have raised output but not efficiency; they simply expanded the volume of busy work. Goldman employs ~12,000 engineers; Morgan Stanley spends $4 billion a year on tech—yet margins and staffing look like they did 20 years ago. The primary reason for this stagnation is org design and cultural inertia - not lack of technology.
Legacy apprenticeship model is outdated
The investment-bank ladder—analyst, associate, VP, director, MD—was designed a century ago to teach juniors by having them redo senior work. Evaluation and pay still hinge on visible effort: analysts log hours to prove commitment, mid-levels demonstrate diligence through slide volume, and partners judge quality through the grind they can see. The structure optimizes for apprenticeship, not speed or iteration, and any tool that shortens the grind undercuts the promotion path the pyramid is built on.
Committee governance
Strategic decisions sit with multi-partner committees rather than a single founder-CEO. Those committees prioritize revenue stability and rainmaker retention, so proposals that alter compensation, headcount, or workflow face lengthy debate and often stall. Even leaders who see the benefit of radical change have limited authority to force it through, and few personal incentives to spend political capital on re-architecting a system that already pays them well.
Tech solutions exist, but stall in deployment
Modern tools—Zoom transcripts, CRM capture, data APIs—are standard elsewhere, yet bank analysts still hand-scribe calls. Vendor-risk reviews can take a year, so anything new bolts onto the stack instead of replacing it. In my own early customer discovery interviews, the bankers’ top requests were call transcription and model-version control—features Zoom and SharePoint already ship, but culture and politics keep them switched off.
Incremental gains have limits
Centerview Partners is the usual exhibit for a “lean boutique” done right. The firm launched in 2006, kept headcount deliberately low, and focused almost exclusively on multi-billion-dollar transactions. That structure delivered meaningfully higher productivity:
Two caveats keep the result in perspective:
Purpose-built structure. Centerview was organized from day one to minimize layers and bureaucracy; replicating the same ratios at a 30 000-employee bulge bracket would require dismantling the existing pyramid, not installing new software.
Deal mix matters. The margin lift largely relies on mega-deal fees that rise far faster than staffing needs. If you apply this model to the smaller, more common deal sizes (let alone mid-market deals) and the efficiency advantage narrows sharply.
Centerview therefore demonstrates that thoughtful org design can push margins, but even the best traditional boutique shows a step-change only when headcount stays low and deal sizes stay very large. Point solutions alone, bolted onto legacy structures, are unlikely to reproduce that lift across the wider market.
While point solutions from AI vendors will surely be adopted, they won't fundamentally alter workflows within legacy constraints; productivity gains will likely remain marginal as busy work expands to fill any freed time.
Substantial productivity gains require an AI-first culture, risk-on ownership, and a vertically-integrated system—all built from day one. Incumbents are not configured to deliver that combination.
IV The AI-Native Opportunity
Picture Goldman Sachs was founded this morning and not in 1869.
No legacy tech stack, no pyramid hierarchy, no 60-slide pitchbook template. How would you staff your first M&A mandate, and which tools would you use?
That blank sheet defines the AI-native design space.
Three governing principles drop out:
Structural org redesign. Replace the five-layer hierarchy with the smallest team that can originate and close a deal. Junior time goes to judgment and outreach, not slide formatting.
Unified system of record. Calls, e-mails, financials, buyer notes—everything feeds one datastore that AI agents or humans can query in a single call instead of an endless SharePoint hunt.
Zero-marginal-cost analysis — the “why-not” loop. If an agent can screen and rank 10,000 buyers in less than ten minutes, “uneconomic” tasks become standard, compounding insight on every client touch.
Ironically, technology is the straightforward part; aligning data, org design, and talent is the real lift. An AI‑native bank starts with all three in place, while incumbents would have to unwind decades of habits, process, politics, and legacy code before they could even start.
So what does this look like in practice? Section V walks through how we did this at OffDeal - our first pod, real mandates, and the numbers behind them.
V OffDeal: The First AI-Native Investment Bank
Building a Bank, Not Selling Software
As we discussed, tech only drives step-change when it is fused to appropriate org design, incentives, and culture, so we launched OffDeal, a full-stack investment bank rather than a SaaS business selling tools to incumbents.
Our first battleground is lower-middle-market M&A—deals in the $5-30 m range that big banks ignore because $100-300k fees can’t support a bloated legacy deal team. A two-person pod plus AI agents does make money at that ticket size, and the high deal velocity gives us dozens of live reps each quarter.
Here’s how the system performs in practice:
Operating model: Pod x Platform
To achieve significant operating leverage we had to do two things:
Refactor the atomic unit of a deal team into a two-person pod
Develop a fully-vertically integrated software stack
The Two-Person Pod
Current throughput:
15-20 meetings booked per BD banker per week
One pod runs 8-10 sell-side M&A processes in parallel
Junior bankers involved in high-value tasks since day 1
The Platform — Three Layers on One Source-of-Truth
These AI-native workflows can only reach their full potential in an AI-native organizational structure and work culture that is AI-first.
The Unreasonable Customer Service
A common misconception is that AI replaces humans. The opposite is true: by eliminating busy work, our bankers spend significantly more time on high-value client interactions. The technology simply creates more space for deeper relationships - and for superior customer service.
When marginal cost of analysis approaches zero, you can do things legacy teams label “nice-to-have”. Two recent engagements show this in action.
Regional plumbing company
Hours before a second BD call, an OffDeal banker asked the platform for a quick diagnostic on the client’s financials.
Instant check – The model flagged that labour costs sat in overhead, not COGS, revealing true gross margin well below peer median
Market sweep – Using the client context, the banker asked the system to benchmark the client’s pricing against every local competitor within 50 miles - all in about 30 minutes.
Insight delivered – The banker presented a competitor analysis and a forecast of the upside from a price increase. Impressed by the depth and speed, the owner committed to an engagement with OffDeal on the spot.
100% AI generated OffDeal valuation slide. Business name and financials are illustrative to preserve client anonymity.
Montessori school — sub-$100 k fee
A single-site operator wanted to sell but couldn’t attract a banker on a success fee that small. OffDeal accepted the mandate.
Buyer universe – Our platform scanned every Montessori or private-school operator in North America—more than 10,000 entities—and produced a ranked short-list of 1,802 prospects in minutes.
Outreach – Personalized outreach + NDAs sent the same afternoon; 71 NDAs returned within two weeks.
Outcome – Four firm offers created real tension; the winning bid closed 32 % above the seller’s initial offer 45 days after deal launch.
In both cases, tasks that would have been uneconomic—or impossible—for an incumbent investment bank were completed at near-zero marginal cost and within hours, not weeks. Every slide, buyer list, and diagnostic feeds back into the unified data layer, giving the model more examples of banker-approved output and making the next mandate faster and smarter.
The compounding advantage
Product and deal teams work side-by-side. When a prompt mislabels a metric or a buyer filter needs refinement, engineers hear it in real time and ship a fix the same day. Because every application is internal, we can release features at “good-enough” and harden them through live use rather than long test cycles. This creates a rapid product feedback loop.
External model progress compounds the effect:
Longer context windows. A 1 M-token limit lets an agent ingest the entire deal context in one pass.
Stronger reasoning models. Can handle complex scenario analysis that now matches—or exceeds—human output.
Reliable tool use. Models call internal APIs to run deterministic tasks, widening the set of workflows that can be automated.
Each release frees more advisor time for the work clients value most—strategy, negotiation, and support during the single largest financial decision of their lives.
VI Looking ahead
Launching any full-stack startup, let alone an investment bank is hard:
Many competencies, zero excuses. We must stand up data infrastructure, AI tooling, deal execution playbooks, and a brand that convinces owners to entrust their life’s work - all at once. There is no single “core feature” we can hide behind.
Long lead, lumpy revenue. A SaaS demo converts in weeks; a sell-side mandate must be sourced, launched, and closed before a dollar lands. We carry months of payroll while a single busted deal can wipe a quarter’s pipeline—so the capital curve is steeper than pure software.
Skepticism by default. Industry insiders often view a VC-backed investment bank run by “tech outsiders” as Silicon-Valley hubris. Until we post results, we’re assumed wrong.
The upside: those same hurdles make the model tough to copy. Clearing them builds a barrier to entry and an operating moat incumbents—or fast followers—can’t easily match.
Early traction
Three deals closed in our first few months by 1 M&A banker
Twelve live M&A mandates with expected fees of $1.3M
$12M fee pipeline generated in Q1 alone
Small business owners are resonating with faster value delivery and deeper analysis.
Software ships to production in days, lifting pod capacity with each release.
Junior bankers move straight to origination and execution—no years lost on formatting.
Where we’re headed
Short-term. Remain in SMB M&A until our processes and software reach maturity
Mid-term. Move up-market, competing for $100m+ deals.
Long-term. Move even more up-market ($1bn+ deals) and add adjacent advisory services—capital raises, debt advisory - on the same platform, aiming for a full-service franchise.
The thesis stays constant: an AI-native org, culture, and software stack can match—then exceed—incumbent capability with a fraction of the labour.
Broader implication
Any advice business built on legacy org structures - law, accounting, consulting, wealth management - faces the same constraints. AI-first, vertically-integrated challengers will appear in each field. The post-AI world produces a clean slate for how human work will be done.
In many ways, we've come full circle to what makes investment banking beautiful in the first place: delivering on those timeless client needs for best price, best terms, and least hassle. The underlying value proposition hasn't changed since J.P. Morgan's first M&A mandate in 1895. What's changed is our ability to deliver on those promises more effectively, for more businesses, with dramatically less friction.
Happy building!
As always, pushback and feedback always welcome. Twitter DMs always open @leveredvlad
If you enjoyed reading this, subscribe to my newsletter! I regularly write essays about AI and business.