Sometimes I run when it's dark…in the early mornings or late evenings. Whether it's too late or too early, and in those moments I wear a headlamp, but the trails and paths I run on still seem hard to navigate and downright unnerving (I really don't like running in these hours). To be honest, it's scary most of the time. I bet a lot of people reading this feel the same way about AI and its impact on the world of M&A as I do about being on a trail in the dark.

So here is what I think business owners considering a sale, the buyers of their companies and the advisors who represent them, need to know about AI and its future in this space.

AI is a powerful tool. Compare it to the laying of the railway across North America versus going across the continent by horse and buggy, on foot, by boat or in Canada…canoe. What was once a many month journey full of perils became substantially quicker and unequivocally safer. It was an astronomical game changer. Generations before couldn't imagine crossing North America by anything other than the methods they had experienced. What the introduction of the railway and steam train did was a nation building exercise. More notably, it changed the game. Travel became easier. Goods could move across the continent with significantly more ease and less cost. It created new industries, jobs, new product demand and equally destroyed others.

It's not egregious to say AI is in this same bucket of game changing as the railroad and perhaps even much more so. It has the power to change nearly every industry, cure diseases once thought impossible to cure, improve technologies and assist the human race in ways we can't imagine. Equally, it has its negatives. Whether it's mass job layoffs, reducing knowledge work or extremes such as AI taking over control of our destiny…it goes without saying that AI will, like the railroad, create many new opportunities and destroy many old.

So too does it go for the world of M&A and investment banking at a broader level.

The analytical layer has already gone

For M&A, the tool of AI is very relevant when it comes to the analytical component of the work we do. Think of building out CIMs, financial models, detailed analysis of the company's financial workings, understanding the competitive framework, the technology stack and so forth. What used to take many analysts many hours, and for some, being technically strong was a competitive advantage…is actually something that AI can do with ease.

The numbers on this are no longer speculative. McKinsey documented a leading bank cutting investment brief production from nine hours to roughly thirty minutes. Citigroup found that 54% of financial jobs carry high automation potential. JPMorgan put its in-house language model in front of something on the order of 250,000 employees. OpenAI reportedly hired more than a hundred former bankers out of JPMorgan, Morgan Stanley and Goldman for the specific purpose of teaching models to build LBO and restructuring models to Wall Street convention. And across the bulge bracket, junior analyst intakes are reportedly being cut by as much as two thirds.

Set that against a different number. A mid-market transaction consumes somewhere between 400 and 800 hours of work from start to close. The part of that AI has taken is the part that was always measured in hours. The part it has not touched is the part that was never measured at all.

This element of the M&A process is now, at a basic and mid-level, generally, a commodity. Which is a bad thing for the plethora of young people who start their careers in investment banking as an analyst, because it's now that much harder of an industry to get into an gain experience.

But it is very relevant to the business owner looking to sell, because it makes all of the firms they may consider using to sell their company appear similar. Across 130 mandates in sixteen countries and billion of dollars in transaction value, the single most reliable thing I have watched owners get wrong is confusing the quality of what a firm shows them with the quality of what that firm can actually deliver. That was already a hard distinction to make. It is now very close to impossible to make from the materials alone.

A one person firm offering a success-only model with no upfront fees can produce work that looks similar to that of the very best investment banks with many employees and big upfront work fees. In fact, if you know what a Claude CIM output looks like…you can spot this right away. And it goes further than simply a CIM. They can appear bigger, with a more impressive website that was also created by AI, staffed by AI agents who are seemingly people to the untrained eye.

Output is not outcome

Here is the problem this creates for the business owner looking to sell, and it is worth stating explicitly because nobody selling advisory services has an incentive to say it.

The document is no longer evidence of anything.

I watched this play out with the founder of an engineering firm. We had done the work, run the analysis, and arrived at a valuation we could defend to a buyer. A competing advisor arrived with a number north of two hundred million dollars...more than eighty million more than our analysis and one that would put this engineering firm as the single biggest EBITDA multiple / market value against any of this public market comparisons. He took that number as gospel. Sixteen months later the business still had not sold, and he sits today roughly where he sat when we first met him.

Nothing in the competing advisor's materials would have looked wrong to him. That is precisely the point. The presentation was clean, the model was coherent, the number was thrilling. What he could not assess, because the output gave him no way to assess it, was whether the number was grounded in anything real and whether the person handing it to him had any intention of standing behind it in front of a buyer.

Representation quality in the mid-market varies enormously, and it always has. What has changed is that the easiest visible signal of quality has been erased. The business owner selling their company used to be able to look at the work and infer something about the firm. They can't anymore. The work looks the same coming from everyone.

Which means the owner is now forced back onto the only signals that remain. Who has actually closed deals like this one. Who has the horsepower to push hard on deals when the time comes. Who is willing to tell them a number they don't want to hear. Who picks up the phone at nine at night. Who has the confidence to tell them the hard truth and push to get deals done.

The noise problem

As M&A moves into an era where the technicality behind the role is being replaced with AI, a very notable skillset is emerging as the one thing AI cannot replace and where deals are definitively won and lost. That is relationships and the ability to deal with people. I would argue that relationships and reputation in the world of M&A matter more than ever before. In a way, we are going back to what the industry was like in the sixties through the eighties...where the industry operated on these metrics alone.

Why? Because AI reach-outs make cold emails incredibly easy and accessible. This accessibility creates a profound amount of noise that is increasingly being ignored. I get approached twenty or more times per day with some cold outreach from someone wanting to buy our firm or promising us leads, and I delete it or it goes straight to spam. I'm sure you do too.

The data on this is the clearest confirmation of a thesis I have seen in some time. According to Instantly's 2026 cold email benchmark report, which measured across more than 700,000 businesses, average reply rates have fallen from 8.5% in 2019, to 7% in 2023, to 5% in 2025, and to 3.43% in 2026. One breakdown of a thousand sends puts it in starker terms: roughly three hundred never reach the inbox, four hundred and fifty are deleted unopened, two hundred are opened and ignored, forty-five reply, and thirty of those replies are a decline. Fifteen produce anything real.

Generic outreach lands somewhere between 1% and 3%. Outreach that is genuinely specific to the recipient lands between 5% and 18%.

Read this data as an M&A person rather than a marketer. The tool that made outreach free also made it worthless, and it made it worthless fastest for the people who leaned on it hardest. Many smaller M&A firms are increasingly reliant on cold outreach technologies, inclusive of AI-generated contact lists of targets in a process. They rely on these lists because building your own, and building a name that carries weight in this industry, takes years and a considerable amount of money. The more this technique becomes adopted, the less it generates results. This is relevant because when firms rely heavily on AI for this key portion of an M&A marketing process, they are removing the human element that makes selling a business for many millions of dollars such a unique and complicated experience.

Perhaps more concerning is when new entrants into the market do not know any other method than these approaches. There is a generation of advisors being formed right now who have never been taught that a buyer list is a set of relationships rather than a spreadsheet. It's getting to know people. It's sending Christmas cards, meeting up for coffees, countless calls and knowing more about them than a name on a list.

Deal mortality

Where AI really goes astray is when it is tasked with dealing with complex human challenges that require a deep psychological understanding of the people at the table.

The market data has started to show this in a way it did not five years ago. Axial surveyed lower middle market dealmakers this summer. Valuation expectations were named the single biggest reason deals failed to close in the first half of 2026 by 57% of respondents, more than double the 28% recorded a year earlier. Over the same period, diligence findings fell from 25% to 10% as a stated cause of failure. Fifty-one percent of advisors reported more deals going on hold.

Sit with that pairing for a moment, because it is the entire argument of this essay expressed in someone else's data.

Deal mortality is migrating out of the data room and into the conversation. The analytical failure mode, the one where something ugly turns up in the numbers, is shrinking. The human failure mode, the one where a seller's expectation and a buyer's number cannot be brought together, is growing (due partly to AI valuation guidance that is built on a fictious understanding of the market). The work that is dying is the work AI does. The work that is deciding outcomes is the work AI cannot do.

I have watched this at close range more times than I can count. I have had deals fall apart over very small earn-outs, small percentages of the transaction, simply because they were based solely on a factor that was not in the business owner's control. The buyers failed to recognize that the actual issue was the lack of control for the seller, not the earn-out itself. Every model in the room said the economics were immaterial. Every model in the room was right, and the deal died anyway.

This point is incredibly relevant to note. It is not in the financials, it is not in the diligence list, and there is no version of it that surfaces in a document.

The buyer who missed the window

Recently we worked on a deal that started to stall out. Our client had a very personal reason for wanting to sell their company, the kind of reason that does not fit on a CIM and does not get raised on a management call, but that fundamentally drives every decision the seller makes through the process. The buyer, on the other side, was very interested in the business itself and focused on making sure both the business and the deal were as de-risked as possible.

At face value, this is what mid-market M&A is supposed to look like. A motivated seller. An interested buyer. A clean fit on paper.

Our client needed speed. They had a big life event happening that required their full attention, and the buyer entirely overlooked the meaning of that. They proceeded with a slow and steady approach, methodical and self-paced, without acknowledging the human reality shaping the seller's decision to proceed at all. To be fair to the buyer, the pace they kept was not unreasonable on its own terms, just slow. It was simply blind to the seller's requirements. They missed the window for our client, and in missing it they caused our client to begin to question whether selling still made sense at all.

By the time they came back, the price had moved. Re-engaging our client required a higher total transaction value with twenty-five percent more cash at close, as the business had gained momentum in the interim.

Nothing about that outcome was analytical. Every piece of work the buyer's team produced was competent. They lost eight figures of value to a variable that appeared nowhere in their model, and they lost it to a person and their human requirements…not to a number.

What I actually use AI for

I should be honest here, because we use AI as a firm and I use it personally.

I don't think you can compete as an M&A firm now without it. We use it for CIM building, for modelling and for analysis purposes. It cuts down on work dramatically. But it still doesn’t take away from our requirement to know the business, the industry and how to position our clients for success.

But the reality is that it is a tool. It is imperative that the people using it understand the outcomes they are using it for, and most specifically how those outcomes affect whether deals actually get done.

Closing deal is more complex

AI cannot be in the room. It cannot be at the dinner where the buyer lets something slip. It cannot pick up on the nuances that cause a deal to go astray, or, just as often, the nuances that tell you a deal which looks dead is not. This is the part of being an M&A advisor that is increasingly the critical element to deals closing or not.

A few months ago, we held a multi-hour meeting in our boardroom between our client, the buyer, and the seller and his advisors. The seller wanted an additional twelve million dollars on account of a deal term his own advisors had mis understood. It began the way these things do, with logic. Numbers on a page, positions stated, a shared attempt to find a rational path through a gap that was, in my read, entirely closeable.

Then it tipped. Not over the twelve million. It tipped the moment the conversation went from logic to ego, and once ego entered, neither side could hear what the other was saying anymore. The words kept coming but the listening had stopped. The room erupted into yelling and pointing and visibly hurt people. The seller stood up mid-meeting and walked out of the building, and his advisors followed him.

By every visible measure the transaction was over. Our own client said so. The seller told me just as much.

I did not think it was. I had been listening, and what I heard underneath the argument was different from what most people in that room heard. The actual words being used. How each of them was struggling with what the proposed structure change meant for them as people rather than as parties to a transaction. And beneath all of the noise, a technical structure that was the real source of the problem and that nobody had yet named.

I told both sides I wanted to keep trying. They each told me separately that I was welcome to, but that the deal was finished. Within a week the structural issue had been reframed, both parties were back at the table, and the transaction closed a few weeks later. Both of them told me afterwards that it only closed because of the work that went in when it got hard.

Here is why that story belongs in an essay about AI. Every person in that room had the same facts. The buyer's advisors had them. The seller's advisors had them. A model handed the full transcript would have had them as well, and it would have reached the same conclusion everyone else in the building reached, because it was the conclusion the evidence supported. The deal was dead.

The read that said otherwise was not in the evidence. It was in knowing those two people.

Closing deals has become more complex, and I predict it will get more complex when AI agents and AI-backed investment bankers, buyers, sellers and advisors are all sitting around the table using the same playbook and the same theories.

The honest objection to everything I have written is worth stating, because it is the one I would raise if I were reading this rather than writing it.

Models are getting better at reading people, and they are getting better quickly. The claim that a machine cannot detect hesitation or discomfort or a shift in register is a claim about today's capability, not a permanent law. And the preference I am describing may well be generational. The owner who is forty-five today and sells at sixty will have spent another fifteen years handing consequential decisions to software. There is no guarantee they will want a person across the table the way a sixty-five year old does now.

I don't think that objection is wrong. I think it is incomplete.

What an owner actually wants in a transaction of this size is not analysis, and it is not even judgement in the abstract. It is accountability. Someone whose name and reputation are attached to the outcome, who can be called at nine at night, and who is still standing there when the deal goes sideways. That is not a capability problem, and no amount of capability solves it. A model can become arbitrarily good at reading a room and still have nothing at stake in it.

For all the things AI can do, it cannot meet a client on their boat, on a walk, in a restaurant…and talk about their fears, their aspirations and their goals. Selling your business is often the biggest and scariest business and personal transaction of an owner's life. Its highly intimate and they want to know that the person selling their business is real, is capable, is attuned to their needs, has their back and can close deals.

There is a saying in business that I love and think about often. It came from an esteemed CEO who was asked what the biggest thing a CEO needs. His answer was good judgement. He was then asked how a person gets good judgement. His answer was bad judgement.

That applies to investment banking and M&A as directly as it applies to anything. I tell clients that I know how to get deals done, and that I know a great many more ways not to get them done. The second body of knowledge is the more valuable of the two, and there is no way to acquire it other than having been there.

The railway did not remove the need for people who knew the territory. It removed the need for people who could walk it.

That distinction is the whole of it. The walking was the hard part, and it is gone, and it is not coming back. Knowing the territory was never the hard part. It was just hidden underneath the walking, and now that the walking is done for us, it turns out to be the only part that was ever load-bearing.

Which brings me back to the trail in the dark. The headlamp only ever shows me a few feet ahead. It has never been enough to navigate by, and it was never meant to be. What gets me through is knowing the trail, and the only way to know a trail is to have been on it before, usually at night, usually having gone the wrong way at least once.

Reece Tomlinson is the founder and CEO of RWT Capital Corp. and the Author of Uncommon Capital.

‍