There’s a number that defines the economics of every investment banking team: one. That’s about how many deals a banker supports per year in a traditionally staffed team. JPMorgan discovered this the hard way when it found that fewer than 30% of its bankers had five or more active clients, amounting to about one deal per banker.
This number is the result of a staffing pyramid that has existed for decades, with layers of analysts doing the production work that keeps a deal moving, topped by a thin tier of senior bankers who own the client relationship and the judgment.
As AI agents become capable junior co-workers, this pyramid is being dismantled.
David Solomon, Goldman Sachs CEO, described what used to happen when a company prepared to go public: a team of about half a dozen people would spend two weeks drafting an S-1, the regulatory prospectus that details a business, its financials, and its risk factors. Now, Solomon said, AI can get that document 95% of the way there in a few minutes. If the 95% is “now a commodity,” he said, the remaining 5% matters a lot, because that’s the margin where Goldman gets its edge over competitors that have access to the same AI models and platforms. (Business Insider)
When every employee can delegate work to teams of agents, the question shifts from how many analysts you need to what team members, from junior to senior, should focus on with the time that is now freed up.
The leverage shift
Solomon’s example captures the shift happening across investment banking. AI increasingly automates the analytical and production work that consumes 60% to 70% of a junior banker’s time. AI tools ingest a company’s financial data and industry context and produce a structured first draft of a CIM in hours, cutting first-draft production time by 60% to 70%. The bottleneck has moved from assembly to review.
The leverage numbers are striking. A traditionally staffed boutique advisory firm runs one deal per junior banker per year. With AI handling the analytical layer, the same banker can support two to three deals simultaneously. Deloitte estimated that generative AI could lift front-office productivity at the world’s largest investment banks by 27% to 35% by 2026, adding roughly $3.5 million in additional revenue per front-office employee.
Real-world examples are emerging that we can test these predictions against. Evercore’s M&A research desk launched an internal AI tool in February 2026 that automates the creation of a precedent-transactions deck. VP-level associates now run their own deal teams without analyst support on the research side. Tools that were experimental AI pilots last year are now in production, changing how deal teams are staffed.
The natural reflex is to treat this as a cost-cutting story: an opportunity for firms to have fewer analysts, leaner teams, higher revenue per head. But that approach misses the bigger opportunity, and it misses the risk.
The new bottleneck
The conversation about AI in private markets has focused almost entirely on the analytical layer, the work that happens after a deal enters the funnel. McKinsey, for example, found a roughly 20% cost reduction on deal processes and a 10% to 30% shorter timeline. GPs report their highest returns from generative AI in deal sourcing and due diligence, the front-of-funnel work.
But when you lift the downstream constraint, the friction moves upstream. Sourcing is the engine that matters most when you unlock the capacity to work more deals.
Eighty percent of corporate and PE organizations now use generative AI in target identification, and 79% use it for target screening. Yet 57% of dealmakers report that senior-level resistance to AI has increased in the past year, driven by accuracy concerns and fiduciary risk. The average firm still sees only about 18% of relevant deals. Bain’s midyear 2026 report describes a “triple shock” that has been weighing on Private Equity firms with slower dealmaking, fundraising, and exits. It leaves GPs hunting for conviction in a bifurcated market. In that environment, sourcing velocity is the difference between winning the deals that matter and watching them go to a competitor who got there first.
The firms starting to figure this out are turning origination from reactive deal flow into proactive market shaping, using AI to score a 500-company universe in 24 hours, compared with the two weeks it previously took to build a 30-company target list.. Of course, a longer list is not the end of a sourcing strategy. The advantage doesn’t come from simply finding more targets, but from what you do with them once you’ve found them. This is where the human premium lives, and where employees should be investing the time AI has saved them on sourcing and analysis.
The human premium is personal relationships
Before I came to WRITER, I managed multi-asset portfolios and worked with distribution teams across private markets. I’ve seen firsthand what separates a banker who sources well from one who doesn’t. It’s never the list of prospects. It’s the conversations.
AI can surface a target, score a universe, and draft a first-pass outreach email. But it cannot do the things that actually win a mandate: probe a management team’s conviction, read the tension in a boardroom, sense when a founder is ready to sell versus when they’re just testing the water. Your edge lies in people who can build the trust that makes a CEO pick up the phone when your number shows up.
A 2026 benchmark built with 502 current and former investment bankers, from firms including Goldman Sachs, JPMorgan, and Evercore, tested nine frontier AI models against 100 junior-banker workflow tasks. Even the best-performing model failed close to half the grading criteria, and bankers rated not a single one of its outputs as ready to send to a client.
The failure modes were specific and instructive: fabricated data presented as properly derived, conflicting figures across slides, hard-coded values instead of formulas. In the researchers’ accompanying survey, only 10% of bankers said current AI tools meet their needs, and 78% cited too many errors as the primary limitation.
That is not a verdict on AI in banking. It is a verdict on ungoverned AI in banking. The failures — fabrication, inconsistency, and unreviewable output — are precisely the failures that governed, grounded AI is built to prevent. When AI is anchored to your firm’s own data, when compliance guardrails govern every generated word, when the output reflects the banker’s voice rather than a generic template, the reliability gap closes.
Redeploy, don’t just reduce
The firms that win the next cycle won’t use AI leverage to shrink their teams. They’ll use it to redirect capacity toward the thing AI can’t do.
When Evercore’s AI tool removed the analyst research layer, the commentary was sharp: the associate-to-VP path just got steeper, because the mid-level support layer that VPs used to develop on got removed). The answer isn’t to eliminate those analysts. It’s to redeploy them into sourcing, the work that builds the judgment, the relationships, and the conviction that defines a senior banker. An analyst who spends their week on origination research and personalized outreach is developing the exact skills that the associate-to-VP path requires. The analyst who gets cut never develops them at all.
This is where the franchise model matters. The best practices of your top sourcer can be encoded into governed playbooks that scale across the team. Every associate’s outreach reflects the MD’s standard, not a generic template. At WRITER, we’ve seen this work. Our voice feature learns an individual banker’s writing style — the phrasing, the habits, and the idiom that makes an email recognizably theirs. One of our financial services clients put it this way: if a client can tell they’re reading something written with AI, it’s offensive. In investment banking, where the product is credibility and senior attention, that standard is even higher. And without built-in compliance guardrails, a personalized draft goes into a review queue and sits. The banker gives up and sends the pre-approved template. WRITER has guardrails built into every generated piece of content, so the firm’s compliance requirements govern every word.
The takeaway
AI is dismantling the analyst pyramid that defined investment banking for decades. The firms that win the next cycle won’t use that leverage to cut headcount. They’ll use it to redirect freed-up capacity into sourcing, where the human premium lives. The constraint moved upstream. The capacity is there. The relationships, the judgment, and the conversations that only a banker can have are what turn that capacity into won mandates. The question isn’t whether AI can write your CIM. It can. The question is whether you’re using the time that gives you back to source the next deal, or just to cut the analyst who would have sourced it for you.
