The Un-Automatable Edge

I round up the most relevant AI-in-finance news - the deals being done, who’s rolling out what, and what’s actually working on the front lines

…the stock is down 28% year-to-date and I'm still yet to find anyone who likes it. This article probably tells you why.

Meanwhile at the other end of Wall Street, the big banks cut around 5,000 jobs while booking record profits. Nobody linked the two on the earnings calls. Everyone volunteered bullish AI commentary moments later. Citi alone has 80% of its 224,000 employees using internal AI tools, with trading desks saving 1,700 hours a month.

Elsewhere: Bloomberg's Chris Hughes argues PE is in a worse spot than private credit on AI, KPMG's Q1 Venture Pulse hit a record $330.9bn, and PIMCO is weighing a $14bn debt deal for an Oracle data centre.

But first, my take on sourcing, because it has come up in nearly every conversation I have had this quarter, and I think the consensus answer is about to get a lot more wrong

In This Week’s Issue:

From The Trenches

The Un-Automatable Edge

I wanted to write about sourcing this week because it came up with real depth on two separate calls.

The first was with the head of sourcing at a multi-billion-dollar sponsor. The conversation started as a discussion of whether DealSage should build in sourcing at all. It ended up being a useful download of how their actual playbook is changing right now.

The second was with a reader who got in touch specifically to ask how I was thinking about it, and that conversation sharpened the whole essay. Thanks for the nudge.

On top of that, I was on a panel at a conference in Miami a few weeks back. The first two questions from the floor were about AI's impact on sourcing.

That is three independent signals in the same month, all landing in the same place. So here are the thoughts.

Everyone has the same tools. Grata, SourceScrub and Inven have spent the past decade building pretty good databases of private companies, and at this point the differentiation between them on identification alone is narrow. That part was broadly true before AI.

What has changed is the cost of doing anything with the list once you have it. AI has pushed the cost of sending a personalised email to close to zero.

Scrolling LinkedIn this week I came across a promoted post from Inven announcing their new MCP. One of the two example prompts in the ad, alongside a market analysis, was "Draft personalised outreach emails for every company in my list."

A year ago, a business owner told me he was already getting twenty to thirty emails a week from funds wanting to buy his company. With every buy-side tool now one prompt away from drafting a thousand personalised emails in a minute, that number goes up by an order of magnitude overnight.

What the owner sees is noise. What he reads is close to nothing.

The perceived value of the list goes down in parallel. If every firm is pulling from the same database, sending the same sequences, and using the same AI to write the same "came across you in my research" intros, the list stops being a differentiated input. It becomes a commodity.

The most concrete thing I heard came from that first call with the head of sourcing. They told me they had just churned off SourceScrub. They are still using Grata, but only for market-mapping.

They are not running email sequences off it anymore. A token email gets sent as a multi-touch anchor so the prospect remembers their name, but the real motion is the phone call.

The interesting detail: the firm has its associates doing the cold calling. People on some of the best-paid junior seats in finance, dialling for dollars.

And by their own read, that channel has been their most productive origination method this quarter. Not the database subscriptions, not the email sequences. The cold call.

That is not where anyone expected sourcing to end up in 2026. But it fits the logic.

The cheaper and more commoditised digital outreach gets, the more valuable the un-commoditisable channels become: the phone call (don't @ me with AI voice callers), the industry conference, the operator dinner, the trade show.

The second thing I keep coming back to is personal brand. I first had this thought about bankers. Sell-side coverage runs on pre-existing relationships, and I expected more MDs to start leaning into LinkedIn and long-form writing as a way to build those relationships at scale.

The same logic applies to almost anyone in the outreach business. GPs, independent sponsors, VCs, corp dev. And yet I am still a little surprised how rarely I see it done seriously.

But the owners of the businesses I want to buy aren't on LinkedIn, so why bother? I'm not going to reach them there anyway.

My retort: a) you'd be surprised how many are, and b) even if they're not, the first thing they're going to do when you cold call or cold email them is google you and see who you are.

If the search results pull up a LinkedIn profile with a series of thoughtful takes on the market, on operational challenges, on how the industry is evolving, that changes the conversation before it even starts.

The brand doesn't have to reach the owner directly necessarily. It has to be there when they go looking.

The interesting bit isn't the sourcing mechanics. It's the pattern that sits underneath them. The first instinct with any AI tool is to use it to do more of whatever we were already doing.

More emails. Faster memos. More analysis.

The IC memo that used to take five hours, you can now write in five minutes. So you write five of them. The outreach email that used to take an analyst an hour to personalise, you can now fire at ten thousand companies before lunch.

That is a race to the bottom. The signal disappears in the noise.

The memo that actually gets read is the short one. The email that gets opened is the one with a warm intro. The deal that closes is the one where a partner met the owner in person at a trade show last quarter, or called them, or turned up on their LinkedIn the day after.

AI is a multiplier on whatever direction you point it. Pointed at doing more volume of low-signal work, it produces more low-signal work faster. Pointed at the parts of the job only a human can do, it frees up more time to do them.

This is why we took an early view at DealSage not to build in sourcing. The database players had a head start, the market already felt commoditised, and the promise of AI was never supposed to be sending the same owners a hundred times as many emails as we could previously.

The value in sourcing has always been in the bits that don't scale. The conference, the call, the coffee, the ten-year relationship that means you hear about the deal before the banker is even hired. That was true twenty years ago. It is still true now.

So when I look at the Inven ad, my honest reaction is that it makes me want to use the service less, not more. The moment every firm can draft a thousand personalised outreach emails for every company in their list, emailing any of those companies stops being worth doing.

The value was always in the bit that couldn't be automated. It still is. Which is a useful reminder that getting value from AI is as much about what you choose not to automate as what you do with it.

News Digest

Salesforce's AI Problem Is An Everyone Problem

The WSJ ran a piece on Salesforce this week that caught my eye. Partly because the stock is down 28% year-to-date and the analyst crowd has made the company the poster child for the SaaSpocalypse. But mostly because the name has been coming up on almost every call I've had with a mid-market sponsor or corp dev team for the past two months.

Marc Benioff went on record to push back on the software bears. There are a couple of quotes I think are worth paying closer attention to.

The first one: Agentforce, the AI flagship and the centrepiece of the company's pitch to investors for the past two years, has been, in Benioff's own words, "somewhat slow to gain traction."

The second, which I think is actually more revealing: what Salesforce really needs, he said, is "positive word-of-mouth from clients talking up the value they derive from its AI products."

I have been in hundreds of conversations with finance and deal professionals over the past year. I have not heard a single one of them talk up any value they derive from Salesforce AI. Not once.

The details:

Why it matters: Salesforce is the specific example, but the pattern applies to every seat-based enterprise SaaS business. When the moat is switching cost and the cost of switching is about to get automated, the moat is gone. That is the bet the public markets are making, and this interview did not do much to change it.

My take: The thing that keeps jumping out at me is the time gap. Salesforce has been saying AI is its top priority for five years. Agentforce was launched with enormous fanfare. And the CEO's best answer on traction is that the company needs word-of-mouth, from customers who, as far as I can tell, don't have much to say.

This is the same pattern I wrote about above. Companies default to using AI to do more of what they already did. Salesforce's AI roadmap is, largely, sales automation applied to their existing product. That is why it has not landed. The firms getting real value out of AI are the ones that redesigned the workflow first and then added AI, not the other way around.

For financial services specifically, Salesforce was always a stretch. It was built for territory-based enterprise reps selling into large accounts with quarterly cadences. That has almost nothing to do with how a PE firm, an independent sponsor or a debt fund actually operates. The only reason it was ever the default is there wasn't a better option. There are better options now. And the migration problem, the one thing that made the lock-in real, just became trivial.

The two sponsors who told me they were ready to leave are not outliers. They are early.

Read more

Wall Street Cuts 5,000 Jobs On Record Profits

Bloomberg reported this week that the six biggest US banks collectively trimmed more than 5,000 roles in Q1, on $47.3bn of combined net income. Most of that was Wells Fargo, which cut over 4,000 by itself, on its 23rd consecutive quarter of headcount reductions. JPMorgan and Morgan Stanley actually added staff.

None of the executives explicitly linked the cuts to AI. All of them volunteered bullish AI commentary moments later on the same call.

Citigroup's Jane Fraser disclosed the most concrete numbers of the week. Over 80% of Citi's 224,000 employees are now using the firm's internal AI tools. The trading teams are saving 1,700 hours of work per month. The 10,000-person engineering org used AI to remap an internal system in two days.

Goldman's David Solomon was the most effusive: "The power of the technology, the ability to remake processes, to create efficiencies and also create more capacity to invest in growth, I can't find a CEO that's not talking about that."

The details:

Why it matters: First quarter where the AI productivity story has shown up on the sell-side cost base without anyone having to squint.

My take: Four of six banks cut headcount. Not one named AI as the reason. All of them named AI as a top-three priority. Connect those two statements on the same earnings call and you get the honest version of the story the CEOs aren't quite ready to say out loud.

For PE firms, this is a leading indicator. The sell-side runs six to twelve months ahead of the buy-side on genuine AI deployment. If banks can grow revenue while shrinking headcount in the first full quarter of serious use, the buy-side will face LP pressure to do the same on a two to three quarter lag.

Read more

Private Equity's Bigger Problem

Bloomberg's Chris Hughes ran two opinion columns this week with a genuinely counterintuitive framing. The consensus for the past six months has been that private credit has a SaaS problem. Lenders piled into direct loans against software businesses at 2021 multiples, and AI is about to re-price those businesses.

Hughes flips it. Private equity, he argues, is in the worse position.

Private credit has covenants and seniority. If the business craters, lenders restructure. Private equity owns the equity, and equity takes the first loss.

Sitting alongside that argument is S&P Global's 2026 Private Equity Survey, which has 53% of GPs citing deteriorating credit quality and rising defaults as the top risk in private credit this year. Yet when the same survey asked about AI adoption inside PE firms themselves, the answer was that usage remains limited despite explicit concern about AI-driven disruption to portfolio companies.

That gap is the story.

The details:

Why it matters: The firms sounding the loudest alarm about AI disrupting their portcos are, by their own survey data, the least prepared to respond operationally. That's not a sustainable position for another two or three quarters.

My take: If Hughes is right, and I think he is, the next cycle of PE underperformance is going to be structural rather than cyclical. You can grow out of a slow quarter. You can't grow out of owning the equity of a business whose revenue model just got automated.

The operational response has to come from inside the firm, and right now it mostly isn't. The LPs who read the S&P survey alongside the Bloomberg columns should be asking their GPs one specific question: what are you actually doing, inside the portfolio, to make sure your operating companies aren't the ones a new wave of AI-native startups are about to undercut?

Read more

Other Interesting Things I’ve Read of Seen This Week:

KPMG Q1 Venture Pulse: global VC hits a record $330.9bn (April 15) - A single quarter, ten megadeals above $2bn, with most of the money flowing to OpenAI, Anthropic, xAI, Waymo and Databricks. The rest of the venture market is living on crumbs around the same ten cheques.

Sequoia raises $7bn to double down on AI (April 16) - New leadership, fresh fund, same playbook. If anyone was worried the AI funding cycle was cooling, Sequoia would like a word.

OpenAI acquires Hiro, an AI personal finance startup (April 13) - Product shuts down, ten-person team joins OpenAI. OpenAI is quietly assembling a consumer finance capability the way Google assembled Maps, one small acquihire at a time.

PIMCO weighs $14bn debt deal for Oracle data centre (April 14) - Institutional credit is now willing to single-handedly underwrite an AI data centre build. The story isn't the deal, it's that a deal of this shape is now normal.

Fortune: the three forces dismantling enterprise software margins (April 17) and CIO: AI isn't killing SaaS, it's exposing which platforms matter (April 18) - Two pieces making the same argument from opposite angles. Vertical, workflow-embedded software survives. Horizontal, seat-based software gets re-priced. Pick the flavour you prefer, the conclusion is the same.

S&P Global: middle-market PE consolidates as Q1 exits fall (April 17) - Fewer exits, more manager-level consolidation, uneven AI adoption in deal processes. The middle-market squeeze isn't a forecast anymore, it's showing up in the print.

Bloomberg Opinion: to make a tech unicorn, mix a few workers with some AI hype (April 16) - A nicely sceptical counterweight to the Lean AI cheerleading. The part worth keeping: the best lean-AI businesses and the worst AI-hype unicorn stories often have indistinguishable cap tables.

Acquisition Intelligence is a weekly newsletter on AI in M&A for finance professionals, private equity investors, investment bankers, corp dev teams, and deal-makers.

For questions, feedback, or to share what you're seeing in the market, reply to this email.

P.S. I'm Harry, co-founder of DealSage. We're building an AI-native deal intelligence platform to help professionals turn their institutional knowledge into better decisions. If you're curious what we're up to, check out dealsage.io or just reply here