Where AI Actually Helps and Fails in M&A Legal Work

How often do you get to spend 2 hours with hundreds of corp dev leaders who are all working through the same problems you are? That's what the Buyer-Led M&A Summit is on August 18th. I'll be presenting the State of AI in M&A 2026 report live for the first time before it goes public. If you listen to this podcast, this is the kind of conversation you show up for, and August 18th is where it's happening. Every registrant gets a free copy of the report when it drops. It's completely free. It's virtual, 11:30 to 1:30 Eastern. Register at dealroom.net/summit or link in the show notes. That's dealroom.net/summit. Now back to the episode. I'm Kison Patel, and you're listening to M&A Science, where we talk with deal professionals and learn valuable lessons from their experience. This podcast focuses on stories, strategies, and what actually happened during M&A deals. Hello, M&A scientists. Welcome to the M&A Science Podcast. This show exists for one reason, to learn from the best practitioners actually doing M&A, not the consultants and academics talking about it, We track down top operators in the world, get them on the mic, and pull out what's really working and turn what we learn into frameworks, playbooks, and the only certifications built on real practitioner experience. That's the whole engine behind buyer-led M&A, the operating standard for buy-side deals where the buyer drives strategy alignment and value creation from day one. It matters because the old sell-led model is why most deals underdeliver. Buyer-led M&A is how the best operators flip that. You're serious about getting better at this. Check out the M&A Science certifications at mascience.com, built from over 400+ interviews with practitioners who've actually done the deals. Lead the deal, own the outcome. Let's get into it. I'm your host, Kison Patel, Chief Scientist here at M&A Science. Today I'm joined by Aaron Binstock, partner, also co-head of the private equity practice here at Cooley, where he focuses on private equity, M&A, and complex corporate transactions. Aaron's practice focuses on the tech and media industries with a particular emphasis in cybersecurity, artificial intelligence, defense tech, SaaS, and communications. He's worked on deals from under $1 million in value to over $7 billion and advises boards, sponsors, GCs, and executive teams at some of the most active companies and investors in those sectors. Cooley has been an early public adopter of AI and legal services, And Aaron's seen firsthand what it can do, where it breaks, and how client expectations are shifting because of it. Today we're getting into the practical side, the prompts, the governance questions, and where the line is between work AI can handle and judgment that still requires a human. Aaron, how you doing today?

Doing well. Thanks for having me, Kison.

Thanks for hosting live here in New York at Cooley's office in Hudson Yards.

Yeah, it's great to have you.

Appreciate it. Appreciate the office space and the view out here. Can we kick things off with a bit about your background?

I'm a double GW guy, so went to GW Business School undergrad and stuck around for law school. I've been executing M&A and private equity transactions for approaching 20 years now. I joined Cooley about 15 years ago, and there's no better platform on earth to build a tech-focused transactional practice. My work, as you noted in the intro there, is focused on working with federal contractors, and I also do a lot of work in the media space, have bought and sold a lot of publications you probably know, Time, Sports Illustrated, Fortune, Money, as well as TV stations, and then do a lot of work with companies that are producing technologies that support the United States and our allies at home and abroad. My work also is heavily involved with sponsors that are investing in these spaces, both companies and with private equity sponsors and their portfolio companies.

That's like a pretty diverse coverage to go from defense, media, tech. Is there some interlinks between them or sort of things that you take over from one sector to another?

The federal stuff I've been doing since day one of my career, and it's really my sweet spot. The media side really resulted from representing one client, Meredith Corporation, over many, many years, and probably have done 20 transactions over my career with them. But they were a media conglomerate that, as I mentioned, we helped buy, timing, and then divest a number of those brands before splitting that public company in 2 and selling the print and digital publishing business to DoorDash and the TV business to Gray Television back in 2021. Happened to just gain a lot of media expertise by working with them over my career. And as you know, once you start doing deals in that space, you tend to get calls from others who are similarly active in that industry.

Cool. I'm excited about the conversation today. I feel like we need to put a disclaimer out there because I feel like I'm going to ask you some pretty direct questions. I think it's fair to say these are all just opinions, not representation of the firm. Is that the right way? Personal opinions.

This is not investment advice.

Or any kind of advice.

Yeah. Good disclaimer. We're going to talk about that too, in terms of how you use AI in a way that's providing legal services and then what the lawyer's role is relative to the technology's role.

I should have had AI draft a disclaimer. Next time I'm going to do that. So Cooley has publicly, openly committed to AI in terms of what it provides in legal services. I want to break down what that means because I feel like I commit to using AI for legal services too, and we're going to talk about that. But what does that mean when Cooley puts it out there in terms of the direction of the firm and how they operate?

Cooley chose to go public with our AI principles in part because we feel like clients deserve to know how their lawyers are actually working, but also because transparency is part of trust and hiring a lawyer is a trusted relationship. We wanted to make sure that when clients hire us, they understand the tools that we're using to perform on their behalf. Cooley, as you know, is a well-known sort of tech-focused law firm. We've always been at the forefront of adopting technologies and serving clients that are on the forefront of emerging technologies. It's something that we've been really doing well before AI in terms of using different tools to provide the best legal services. And you can go back decades looking at this. If you checked out our Cooligo website, there's free advice up there. There's document generators, services, and tools that we're providing for free to clients to help the ecosystem. More recently evolved into Cooley Go Lab, which was a collaboration with Y Combinator and Legora, which is a popular AI platform for law firms and legal market that we use here. It really was, again, a tool that we're trying to provide to Y Combinator companies to be able to go in and access the best legal information and use AI in a way that's informed, as opposed to just going out and using any model that's publicly available.

A lot of initiatives that you built out in terms of how the firm puts it out. And I think you got a good point. It's very much of almost this open-source effort for putting all this documentation, best practices out there.

AI is definitely a huge step in terms of the evolution of technology and how it impacts practice. But we've been using Kira for years. That's a program that was able to strip provisions out of agreements and help streamline due diligence exercises. We've used internally things like document processors and generators that help us create documents that are used time and time again on the latest state-of-the-art forms in a more cost-effective manner. Really for us, it's just the next step in that evolution, albeit it's going to be a pretty big one.

It is. Let's break down the workflow. So let's say I am either a corp dev team or in-house M&A legal team. We're going to start pursuing a deal. Where does AI sort of fit into the workflow when it comes to legal work and M&A?

The corp dev is one use case. Obviously the use in legal is different. It's funny, I was down in Miami about 3 years ago at the Merger Market Conference and presenting with partners from a number of private equity funds and had asked at the end of my presentation, my panel, how are you using AI in that sort of deal source generation function? And only 3 years ago there was crickets and people really weren't seeing the use case quite yet. Fast forward, I think it's a different ballgame. I was out to dinner earlier this month with a client in New York who is an independent sponsor that is actively using it to find companies to add onto their platform. They're having quite a bit of success in terms of actually using it to find good targets and targets that fit what they're trying to build. I see it much more on the legal side and our teams are using it internally and in-house teams are using it internally on the legal side in terms of actually executing on deals.

I'd love to hear about kind of what are you seeing in terms of how this actually plays out in the real world today?

And I mentioned earlier, we're using Legora as the primary tool within the firm, but we've got tools like Cura and others that we're utilizing depending on the task. It's like anything, trying to find the right tool for the job. There are things that AI does very well and there are things that AI doesn't do very well. Sometimes people are trying to figure out how to prompt it to do things where there's just certain types of projects that AI isn't the right tool for. Understanding that going in can be useful, but in M&A or private equity, there are a lot of things that can be streamlined during the course of the deal through use of AI.

Let's run through some examples because I always feel like just even starting out, you got an NDA. How's this workflow changing? Are you sort of just running through all these documents, doing initial drafts with AI, or are you still using base templates that you already validated and verified? And then—

NDAs are a good example. Adentra's been doing that work with attorney-assisted AI for a while now and has become pretty well known in terms of being a lower-cost provider to mark up lots of NDAs during sale processes. We're using it to draft some documents. From my experience, the more specific you can be with it, the better. But in the prompting, but also in what you're trying to do. For example, I asked an associate on my team to prepare an asset purchase agreement for a transaction we were working on and gave her a number of instructions in the email. This is a good form to start with, but we need to add indemnification and change the purchase price adjustment in this way. And she took my instructions from the email, fed it into Agora with the precedent form that I asked her to start with, and it produced a pretty decent result. There were some things that were surprising about it in terms of how well it performed. Rather than just adding provisions or removing provisions, it was smart enough to, for example, add the related definitions in the definition section. It was able to, on the purchase price adjustment mechanics, not just add that section, but also add into the section where you're talking about how purchase price is calculated to add plus or minus the adjustment defined term. It was making some logical leaps in terms of the way it was drafting in a way that I was surprised that it could do. But I found that it's actually even better when you focus it on certain provisions or specific provisions. So rather than saying, here's a whole document, mark up a whole 80-page document, it tends to get a little bit confused and doesn't do as great of a job in that context compared to take this purchase price adjustment section and make it more buyer favorable or more seller favorable. It tends to do better when you're actually giving it a very specific provision to look at against the database. And of course we train Legora on our information. That's helpful. You got to be careful when you're using public tools because you don't know exactly what it's training on. And you also need to be careful that it's not training on your data that you're feeding into it, but making sure that the right source data for the job is useful as well. With Legora, for example, you can feed that precedent into it, or you can tell it to look at precedent from deals that you've done with a particular client.

That's a good point. It's a lot of, if you look at use cases for AI in an organization, they're mixing their own datasets with it, training it on, and it becomes very proprietary in what that AI is going to have an output versus what do you get off the shelf.

Yeah. But some other good use cases, people think about AI as making a lot of mistakes. It's actually really good at catching mistakes too. If you have it proof or just formatting things or defined terms, It can actually find a lot of stuff that a human may miss. That's a good use of it. We've used it to start developing playbooks for clients that are executing on the same playbook over and over again. It's useful to have that library of knowledge, the client's preferences, the way that they negotiate provisions in one place so that you have continuity across your team. The AI will actually help you develop those playbooks, which is great. It'll create playbooks, but there are a lot of false positives in terms of it will pick up things that it thinks are issues because it's seeing changes that are not real issues. Or it's not appropriately prioritizing them. I had a situation recently where a client had just run an agreement that came back across through a publicly available model and said, create an issues list and send it to us. And it had a red flag on who was going to serve as the independent accountant, which is not something we would even include on an issues list per se, because it's not a material issue, let alone something we need to obviously fill in the blank for. But it doesn't know how to use that judgment that a lawyer would have to be able to prioritize what the client really needs to focus on. And it's also really good at creating the list. It will tell you, here are all the changes. Where I found the lawyer expertise is actually really valuable, because what AI doesn't do well is give you the, what's the suggestion to the client on what are you seeing on this provision? How would you suggest, given the context of this deal, we sort of return a volley on this particular provision? And that's the judgment that clients are really paying lawyers for. It's not like the— distill the agreement down and give me the changes. Anybody can look at the changes to the agreement. It's, what does this change mean in the context of the deal? How should I be thinking about this? And helping them think through what the response is as they're negotiating.

So 3 big areas I heard is quality control, building out playbooks so this becomes repeatable, and then creating that issues list. Yeah.

There's lots of uses. It'll create form documents, it'll populate documents. We've had some success using it where we'll take the final version of merger agreement and terms from the deal, and we'll have it create an information statement. So if you're familiar with mergers, when you go out to solicit the stockholder vote in connection with the deal, you have to provide them under Delaware law an information statement that has all kinds of information about the deal and the stock price and financials. It actually does a pretty decent job taking a precedent information statement and populating that with the terms for the deal. Those are things that are just summary in nature. And so I think wherever it has to summarize something, it tends to do a pretty decent job. Not perfect, but decent.

I want to find out what the ideal workflow is. So right now, my big thing is NDAs. Anytime you talk to potential investors or look at companies, always NDA. And what I will do is I will, a lot of times, if it depends, our paper, their paper, our paper, easy, you got to sign it or there's red lines, send it to a lawyer, look at it. If it's their paper, I'll run it through AI and saying, hey, which basically pull out an issues list, but it overdoes it. It always will pull out like 8 things and then you look through it and it's like, okay, I don't care about this. But there's maybe 2 things that are, hey, this is probably material. I flag those and I send it to our counsel again and just saying, hey, I looked through this. These are the only 2 things that make sense that I want you to pay attention to when you rewrite it.

Yeah.

And so far that's worked out pretty good. He acknowledges it and he appreciates that I don't just run off and just send it after letting the AI mark it up. I don't know, is that the right way? What do you think? Because I'm trying Trying to flip that around, if I'm working with you, is that going to be something that works for your sort of flow or what's that ideal look like when you're working with clients?

Yeah, the NDAs is sort of a very narrow use case in the sense that there's a certain number of standard provisions that get negotiated in a very standard way. And some clients have preferences or things that they require based on type of business or fund that they're running to make sure that they're not getting themselves in trouble. It's very commoditized in the sense that if you've worked with Ontra, you're putting that playbook together for them where they understand this is the default fallback. And then if they don't agree to that, to this. It's something that AI can learn your playbook and provide markups in a pretty efficient manner. When you're starting to deal with 80, 150-page merger agreements and there's a lot of things and the deal's more bespoke, I think it becomes a little bit more difficult to just have it run through that document and do a decent job. Again, it's good on certain provisions and I'll use it as a drafting tool all the time, for example, because we've got Word plugins where we can say, I'm looking at this provision, can you give me an example of this? Or, can you mark this up and track changes? And then I can look and see what it does, and maybe it even thinks of something that I'm not thinking of because it's comparing against a lot of precedent in the background. That's a useful tool. I always like to have it make the changes in Track, by the way, and then I can decide what I want to accept or don't want to accept. Because to your point, sometimes it does put a few things in that you don't care about or that just aren't important.

So before we replace all the lawyers with AI, you had a good story of saving the client from a million-dollar mistake, basically. That was what AI was directing towards and sort of the experience or human judgment should have been overriding that.

The judgment is key. AI doesn't own anything the way humans own the relationship, the way humans own the responsibility over the transaction. You have to look at everything that it produces and make sure that you're checking it. The story that you're talking about was we had a client that was working on a complicated reorganization transaction and they had put together a tax step chart using AI. And this is, if you don't know, an organizational chart that shows each step of the transaction. the tax rules. It's something that lawyers will work in connection with their accounting firms and tax advisors to put together at the outset of a deal frequently to make sure the parties are on track in terms of how it's actually going to be structured. And the client goes, here's this step chart. The email sort of made it sound like it had come from their tax advisors. My tax partner looked at it, got a phone call 5 minutes after sending her the email, and she goes, is the target an S corp? And I said, no, why? And she goes, the whole step chart is premised on the target being an S corp. Everything after slide 1 is wrong and based on this faulty premise, which doesn't work for the transaction and is completely off. And so I called the client, I said, your tax advisor looked at this? And he fessed up to using AI and said, no, but that's a really good catch. And obviously exactly why we're having you take a look at it, but you do need to be careful about it. There can be big consequences to relying on what seems like a small thing, like a step chart, where it can have a big impact on the deal and potentially cost millions of dollars either in Yeah, you hit it spot on.

You need the right balance. You can use AI to really get good results and the quality control, but then that level of human judgment, you just can't get rid of that just yet. But the combination of the two is a deadly combo. I want to talk about real-world examples of prompts that you found to be really valuable in how you're using AI.

I want to go back to what I said at the outset, which is prompts are good and being a good prompter is helpful, and we actually provide training to attorneys at Cooley around how do you prompt it and what are the best ways to prompt it to give you the desired result in a way that's usable. But again, there are some things that it's going to be a good tool for, and there are some things that it's going to be a bad tool for regardless of what those prompts look like. I would go into it thinking, is this a project that AI is a good use case for, or should I be doing this manually or using some other tool for the job? But when you are prompting, the specifics and the more information you can feed it, the better. If you can feed in precedent documentation or forms that you've used, then that's helpful. One thing that we have found a lot of success with is called reverse prompting. You'll give it a very specific prompt, take this provision and make it more buyer favorable, and we want this cap, and you give it all the details that you have. But then at the end of the prompt, you say, is there anything else I can tell you to help you do a better job on this project? And it will actually, the AI tool will ask you questions to help it do better. So for example, I was putting together a checklist for a group of founders that's starting a new LLC that has some complicated nuance to it and was putting together, again, just sort of a list for them to use as an agenda for a meeting to have some preliminary discussion around considerations to think about as they're structuring this new business. And it came out with these questions and then In the reverse prompt, it said to me, wait, is this a Delaware LLC? Yes. Is it a multi-member LLC? Even though I was saying owners or founders in my prompt, it understanding in its language that it was a multi-member LLC, that makes a big difference in terms of how you think about a document like an LLC agreement or other preliminary considerations that are going to be important to figuring out the relationship between multiple founders of a business. That is a good lesson. I would always ask it, do you have any questions for me? after you've given the initial prompt, because usually when you iterate like that with the tool, the AI is able to help you create a better result.

I like that reverse prompting. I've seen the models improve where they're generally improving your prompt before they execute it. And then you can always prompt to create the prompt, which is, I think, really helpful if it's more of a complex output that you're striving for. So then you sort of understand you can make more of a complex prompt and then you can tweak it and then make making something repeatable or turn into a skill depending on which LLM you're using. I think you got a great point on this reverse prompt where you get the output and it's like, okay, what could I have told you to make this result better? And then it's turning around asking you those questions to— Yeah.

And this is all evolving really quickly. The AI's getting better. There are people who are really good at the prompting. I've got a guy in our technology transactions group who is at the forefront of beta testing all these products for our firm. And I was giving him that example I gave you earlier about marking up the asset purchase agreement and how it did some things well, but some things it didn't do well. And He said, well, the more specific you can be and target it to specific provisions, the better. You could actually build a prompt out that says, make this particular purchase price adjustment provision more buyer favorable, make this indemnity, and go sort of prop by prop. But you can have it build a very complicated prompt that sort of takes a long document, breaks it into pieces so that it then is more focused and does a better job doing a markup of a more complicated document. Some of that sort of prompting it to give you the prompt or using agents to sort of create prompts, is where this is all headed and it's only going to get better.

I very much agree. It's exciting to see how that's evolving. You got any other examples?

I use it for lots of different things. There's the standard use case of just getting smart on a subject, whether it's, tell me about this, and it will pull information and provide you sort of treatise-level information on things. In terms of helping with email and inbox management, it can take a very complicated thread where clients and teams have been going back and forth on a particular subject that may be very nuanced, and it will summarize the email thread. Super helpful when you're trying to get up to speed quickly. If you're new to a client relationship and you're trying to get up to speed on their existing documents, it can provide decent summaries of existing documentation. It's not going to replace your review of them, but it's enough to understand the situation and understand what the high-level terms are. So I use it in those cases. On the drafting stuff, a lot of the examples I already gave, we also use it for just working with particular clients. If you're working with a repeat buy-side client or a sponsor who's doing lots of deals and they've got one of those playbooks, You can take all their prior deal docs from that client. If you've done 20 deals with a client, put them into a sandbox in the AI tool and have it create benchmarking. So almost the way an ABA deal term study or whatever, it will create a tabular review chart that says, all right, this is everything that they've agreed to on survival periods for reps and warranties or caps or baskets, and break them all down by whatever categories you want it to create. Super useful detail to have at your fingertips when you're working with those clients because They may say, what did we do in this deal? Or what do we typically do? And to have that information at your fingertips can be really helpful.

I like that, the summarize and then the benchmarking. What about if you're doing a deal that's like a cross-border deal, there's a batch of employees in Denmark, you're trying to figure out what are the HR laws that we have to abide by doing this transaction there and those things. Do you start leaning on AI for that or do you still—

Yeah, I wouldn't rely on the AI for anything. And fortunately I haven't had a bunch of Denmark employees recently in any of my deals, but Recently had a deal where it was an odd one for business reasons. The parties decided, even though it was a US acquisition, to use a European lockbox-style purchase price adjustment in the deal. And that created some weird timing things with respect to the way certain accounting items were adjusted. I was able to ask some questions overnight when my UK tax partner was still asleep to be able to just, again, get up to speed quickly before I could talk to an expert on it. What would the impact of using this lockbox in this style transaction be to the way this particular thing gets adjusted? And it was able to give me a really good response and even go beyond that to say, how is this negotiated? And help me with that. And when I talked to my UK tax partner in the morning, a lot of it was pretty spot on. Again, I don't trust the—

Can you remind me of this? I forgot. The lockbox, how does that price adjustment work?

Just generally?

Yeah.

You basically lock the balance sheet as of a certain date. It's usually before you've signed the agreement. say, okay, we're going to take the end of May and that's going to be sort of the adjustment. And then there are certain items that are leakage or permitted leakage that are allowed from that date to closing and you adjust for those. But otherwise you're essentially fixing the balance sheet date prior to the deal signing.

What benefit does that have? Because most people just do it right at close.

US deals, we use more of a net working capital adjustment, which is typical where you do it as of the closing date. Obviously simpler than trying to having to calculate a closing date balance sheet and you're not adjusting for impact of the deal or things that have happened since people have been thinking about the deal, but it's not a very common tool.

So it's less aggressive negotiating at the closing table, I would assume.

Yeah, it can be. Plus I've still seen people fight over leakage and debt-to-equity bridges, and there's a lot of things that come up in those contexts that still get a fair amount of negotiation, but just different way of doing things.

Between a prompt that really works well and one that you have to redo from scratch? Do you differentiate that in practice?

The more simple, from my perspective, the better it tends to work. I wouldn't say things you have to redo from scratch, but there are prompts that need to be refined. And again, that, like what we were talking about earlier, where you iterate with the AI to be able to create something that is actually a useful markup. Sometimes for more complicated things, it requires that iteration because it's trying to understand the context and you need to give it more detail than you may have initially given it when you prompt. By feeding more information into it, it's able to provide a better result.

Have you seen an NDA that prohibits the use of AI?

I haven't. Are you seeing those?

I had just talked to somebody about it the other day. They were forbidding putting the data in any kind of AI product.

Yeah. Now we've got clients that say, we prefer not to have you use AI on our work. That's a professional ethical thing that firms need to be cognizant of. If you're telling a client, we're going to use AI on your work, and they say, please don't, you need to be able to have a process in place.

That's even one I haven't heard of. Because I'd be like, by all means, use all the AI you want.

Yeah. So we've got a centralized lookup tool. So when you're in the AI tool, you can see before you start using a particular client matter, has this client opted out of using AI?

What drives it? I guess there's 2 conversations to have here, this sort of with the counterparty permitting them to use AI and then your own client that doesn't want you to use AI. What is the driver? Is it just, hey, we're a conservative company, we're kind of like closer to the Amish and want to stay back in time and not Adopt AI just yet? What is the thinking behind that?

It probably varies depending on the client, and some people are more comfortable with the technology than others. There's concern over how the data gets used. Again, when we negotiate our licenses with these enterprise-level tools, we're very careful to make sure it's not training on client data, that the client information is staying on-premises so that you're not worried about blowing attorney-client privilege, for example, or having that go into some aggregate information source that it's then using to provide information to other people? Because we're not going to let it train on our firm's data to then go out and make that available, even if it's on an anonymous basis to others. So I think there's some concerns about that, the privacy elements of it. There's some concern still about how accurate it is and does it miss stuff? It isn't as great as associates and even partners are. Attorneys and humans miss stuff too. It's a little bit of cost-benefit analysis when you're thinking about whether it makes sense to Use it to figure out, listen, is it going to do a 99.9% good job and maybe it misses something, but a human could as well. We're going to save hundreds of thousands of dollars in the meantime. Yeah, sometimes in certain use cases that makes sense. But yeah, with different clients and their willingness to use it or putting a provision into their NDA that says you won't use AI, it's just a level of comfort and maybe they're thinking about some of these different considerations. Listen, again, a lot of our clients are technology companies and so they were almost asking us to use AI. Even before we were using it to the level that we are today, it was expected that we'd be doing it for efficiency purposes. I'd say we probably don't see it as much, but definitely I'm sure there are companies that are more conservative about it.

All right, let's go over the options for learning M&A. You could go to an industry conference. The networking's fine, but the learning's going to be an afterthought. You could go take one of those M&A certifications, but The curriculum was built 10 years ago off of 1 or 2 practitioners. You could go to a big-name business school, $20 grand for an exec ed program that they cram into 4 days. Great networking, but are you learning the best or are you getting professor theory? Those are the options. Here's ours. 9 years of M&A science, learning hot off the press from the best practitioners alive. We extracted the deal stories, the frameworks, the patterns, and organized them into 2 certifications. More contributors than anyone. The best in the industry. All of it AI searchable with real receipts. I'll just say it. It's the best M&A training out there. Not going to dance around it. And the value isn't even close. Conferences cost thousands. The university package is $20,000. Other certifications run $4,000 to $5,000. Ours is $2,000 a year or $200 a month. Cancel any time. Take the $200 bet. Feel the difference. mascience.com. It's the counterparty. That's, hey, I'm going to provide this diligence information, but I don't want you to put in AI. Or I just was talking to somebody about it where they're starting to see that. And to me, I'm like, I would be a little concerned because there's some reason that—

Yeah. You mean to check it or something like that?

Maybe there's a view I'm coming from operating Dealroom, for example, where we're providing a workflow product for the buy side. And a big use case now is putting all these AI features in there that allow you to I don't want to use automate, but definitely automate, I guess, is the right word. Parts of your diligence, that's something that's come up where they've worked on deals and the seller is, I don't want you to put my company information into an AI product.

They're probably worried about the privacy element of it, right? And that information going into an AI tool where they don't have any control or understanding of where their information's going. Listen, I'm sensitive to that. Again, it's part of the reason why with the tools that we use, the law firm, we're very sensitive about how the license works with those tools so that we understand where clients' data is going.

If you're my lawyer, how are you going to put them in place so we can use AI on this deal?

Well, I mean, again, I've had conversations with clients about exactly that. Is your information staying on Clio servers or is it going somewhere around the world and is it training on your data? No. But getting them to understand that initially can be helpful. On the defense deals and the security deals that I do, there's a lot of unclassified sensitive information that can only be, for example, seen by US citizens and you don't want foreign nationals to be looking at it under certain government rules. There are considerations like that where you have information that you're worried about, like who has access to it, where's it going. Law firms get very sensitive information like that in all kinds of contexts, not just that, right? You can understand why they would have it, but understanding how the AI tool uses the information and where it's putting it can be very helpful in explaining to a client that may have concerns about it, whether it's being used in a way that they're comfortable with or not. And listen, again, mindset prerogative to say exactly that. I don't want to use it and I understand it's efficient, but I just prefer it not be used. That'll change over time. Without anything, people become more and more comfortable with technology, but understanding how the tech works and how the information is used is key.

That's a good point. Now I just started recollecting some of the stuff like the ITAR classified information is probably one of the highest levels of security classification for documentation in the US. And I remember there was a period of time where you couldn't even have this stuff on a device with Wi-Fi. So they would actually have laptops where there was completely removed Wi-Fi and you would actually have to physically deliver this laptop. And that was still in the cloud era. You still had to do that.

Yeah.

So I can imagine now, I can imagine you putting that stuff through AI if back then that's how reserved it was around security.

And even when you think about information like that, law firms are sophisticated with this, but we'll have our intake team clear certain attorneys onto particular matters. So if I've got a matter that's dealing with information like that and I know that I can't have non-US citizens having access to it, we can make it so that our document systems provide access to attorneys that are cleared on a particular matter. And we do that routinely. But yeah, AI takes it up a notch when you're thinking about, okay, not only that, but now you're putting it into this tool that's manipulating it and using it to provide an outcome. The diligence use case you're talking about, there's a lot of efficiency there that can come from feeding the data room information into that tool. Think about how many hours we spend as junior lawyers digging through contracts to look at change of control provisions or anti-assignment language. And now you still have to look at it, but it can pull all the information out and put it into a chart for you and saves hundreds of hours.

Does that mean the legal bill's going down?

In those types of deals, we would usually figure out ways to make it a manageable project. It may have been the case that you'd create thresholds where you're going to look at the 100 top contracts or whatever. Now you can look at all 10,000 contracts and it takes the same amount of time, but it's pulling out the provisions for you. Kuli's been in the press recently about this, but I think the value that clients are paying for is changing. This isn't going to be an overnight thing, but if you look at time it takes to do something and fact that law firms bill by the hour, put 2 and 2 together and see that that has some problems in terms of an economic model.

That's why I was curious. I'm surprised it's not changing faster. We both know there's a very short list of firms that do charge essentially deal points like a banker. I was under the impression that's probably going to catch on more, or at least some kind of hybrid model where there's something attached to actual deal value.

It'll be interesting to see how it evolves. If you look at other, outside of law firms, other professional service models, consulting has probably been the this, right? They've gone to more of a project-based billing where they're either charging for completion of a project or success or whatever. It'll be interesting to see if the legal market evolves in that way. But there's also some firms that do this today to some extent on larger matters, like on M&A where they will charge a fixed fee for successful closing of a matter. And so that's sort of like the investment banker model, right? Where it's like the deal closes, it's this, and whatever this is may be higher than what the aggregated hours may have been on the bill. But Also, they're not charging if the deal doesn't close. It'll be interesting to see where law firms go with this.

Right now it's hourly and that's pretty common, but potentially there's a contingency based on a percentage and then there's a flat fixed fee.

Yeah. And you see, I mean, again, hourly is probably the predominant billing method today, but to your point, there are other models out there. And I think it goes back to that judgment. AI is reducing the time that it takes and making certain things more efficient, but it's not replacing judgment in that sort of trusted relationship. They're advising a team or a partner on things that are nuanced and complicated, and you have to really be in the trenches on the deal dynamic to give them a view on it that aren't just comparing it against what's the market. That judgment that leads ultimately to successful transactions is the part that's valuable.

I want to be clear on this. I want to just kind of understand just for the sake of understanding, and I'm definitely cheap. I'm publicly cheap about what I'm trying to do or be efficient as possible. But I don't want to be cheap with the legal counsel because if anybody you don't want to be cheap with in the deal process is your legal advisor. You can cut corners in a lot of places in the deal, but it shouldn't be with your legal counsel. That's your best bet. So I definitely want to work with the best attorney I can find and they should be happy.

Love it.

I'm curious about just what this industry sort of future looks like, especially now with that many signs. We got different certifications and some of the stuff is right towards Early career practitioners breaking into the industry. And I'm curious about the legal track because I've talked to incoming lawyers or people considering it, even my own kids. I've been kind of steering them away. I'm like, I don't know, between— before it was always the cliché, you come up with immigrant parents, it was go be a doctor, lawyer, engineer. That was your 3 career options. And I've only given my kids one. I'm like, just go be an engineer. Give me like your take.

What do you—

your kids are young though.

11 and 14.

Same age.

I've been thinking about it too. We've had a— I mean, my wife and I and parents, friends in our community have had a lot of discussions about this because we're seeing how it's disrupting our professions. And it's interesting to think about how do you advise your kids what's going to be a good career path for them, given industries that may be replaced or changed in a drastic way. Yeah, it's hard to say. There's things like being a chef, things that have an artistic element to them, which in some ways law is like that too, like the creativity aspects of it, that piece of what we do, be hard to replace. And certain professions like that, it's hard to see AI changing them. Others like pure coding, we're seeing a lot of jobs already being eliminated in the tech world for years now. It'll be interesting. Again, we're sort of early stages here, but it definitely is having me think twice about what career path and what type of higher education I would suggest for my kids.

Are you hiring less incoming associates in the firm now?

We haven't been. I don't think this is an overnight thing. And so I think that When you think about that skillset and what junior associate spends their time doing, that's evolving. The time spent on diligence can be used for time spent doing more complicated things. None of us went to law school thinking we want to sit there and pore through documents throughout the course of our career. So you've kind of gone through that because it was a necessary evil at the beginning of your career and because it's necessary to get transactions done, but it allows somebody who has a world-class education to use their brain in a way that's a little bit more productive. There's still going to need to be associates. Do we need as many over time? TBD. But right now we're not seeing a massive need to replace— we're not replacing jobs with it at that scale.

Be more productive or just not use your brain? Because I would just be getting— here you are, get this done. Blah, blah, blah. I'm out the office early today. Yeah.

I mean, you raise a really good question around— and we've had a lot of good conversations about what the expectation is for attorneys and junior attorneys. And again, how they're using this tool in a way that's productive, but also that is the proper way to do it. You can't just use it as a substitute for legal judgment. You can't just feed information into a doc generator and say, create this document, if you've never created that document and you don't know when you're reviewing it, what's good and what's bad and what's wrong for this situation. that doesn't work, then it's really not ethical to be practicing law in that way. You've got to be able to, in one way or another, understand the way provisions and agreements work to be able to check the AI. You can't just throw up your hands and rely on it. I'm sure there's lots of practitioners out there that have either fed something in and they have to make a lot of changes to it, or they receive something from a junior person on their team and you're like, no, this doesn't work. But those are learning lessons, right? And it doesn't replace the practice of law. It's a tool. That's what we got to keep in mind.

When you think about those junior attorneys, the big value is to help them learn those skills to make judgments. How do you actually approach that?

This is all evolving, right? The way firms are training is evolving too. You need to think about how do you get somebody to short circuit exactly that? Okay. You used to read thousands of agreements and you learned through osmosis and you see how things got negotiated and now you're having AI do a markup for you, but you need to understand that. Some of the formal training may evolve. I like to use this kind of law school, barbery, bar prep course metaphor where it's like, okay, you could read thousands of case law and get the one point out of each case that you read. And then at the end of the time, your 3 years in law school, you have a basis in contracts law and constitutional law and you earn the points through that method. And then you go take the bar class and They give you all the substance that you need to know to have that particular topic down in a month's time. And maybe there are ways that we jump to that, what's the conclusion, without having to pore over all those documents. But I also don't think that, again, where we are today, we're there quite yet. If you produce that chart that says, here are all the change of control provisions and assignment provisions, somebody still has to check the work of the AI. You can't just go through and read it. There's still a lot of learning and reading through osmosis and some looking at those documents. more focused and streamlined than having to go find something in the document, but you're still reading the provision and understanding what that particular provision that you're focused on says. A lot of that is still there. The way that you interact with AI can be helpful when you're thinking about engaging with it and prompting. That's using critical thinking alone to think about how you're actually asking the AI to create what you want it to create. You do have to have that foundation, but once you have that, there is a sort of a creative element and critical thinking element to prompting with the AI. Things like the playbooks, right? When you're using AI to create a playbook for a buy-side client, you're looking at that. That's a learning tool of itself, creating a summary or creating a playbook. We still do this. When deals close, we track data on deals the same way the ABA or SRS have these deal term studies that come out. Sometimes pulling those stats out of documents can be a useful learning exercise for an attorney because they're seeing, one, where do I find it and how is it worded in the particular provision, but also being able to extract those key points and understanding what those look like in the aggregate. And then some of the tools, you've probably seen this with AI, will actually explain the reasoning that it's going through. Have you seen this where it will give you the result, but as it's telling you why it came up with that, that's useful too in the sense that if it's telling an associate or a partner, we made this change because of this, or we pulled in this particular source, that's what we're using. Having that explanatory note with the AI result is a good training tool.

You have a good point though. You need the experience in the war room. You need to be on the other side of the table yelling at the other.

Yeah. Yeah. There's a lot of soft stuff too, right? That when you think about, as attorneys get more senior, their ability to conduct a call with a board of directors or with opposing counsel on points and to be able to engage with them on that, or to run a deal process, or to have that one-off call with the CEO to have that sensitive discussion around a particular point, that voice and those things, that's the art that AI isn't replacing. And that's an area where You're still going to see the attorneys that are successful practice and building trust with clients be able to do that in a really good way.

So if we want to break down basically what AI can own, but then specifically break down the areas of human judgment where you'd want somebody that's done hundreds of deals to really lean in on.

Well, I don't think AI can own anything yet. Again, I want to go back to that. The AI technology isn't owning anything. Even the most basic tasks that it does pretty well, it's not owning them. It's just producing a report. The people that own it are the people that own it. That's the legal judgment and the need to quality control check it. And that's why lawyers are trusted advisors and there's a relationship there because we're the ones owning the result. But it does do things. It does create efficiencies by producing reports and drafting more efficiently. It's again, a tool for the owners, maybe the way I would think about it. And then again, a lot of that human interaction stuff that I was just talking about a minute ago is the part that I don't think it's ever going to be able to replicate. It's good at absorbing context, but it doesn't really absorb the context. I was having lunch with some folks today and we were saying there's all kinds of reasons that particular parties to a deal feel one way or another about, and it could be informed by market or their practice, or maybe they just don't want precedent out there because it's a public deal with their name on it because it creates a bad precedent for future deals. There's all kinds of things that the principal is thinking about in deals that inform their negotiating perspective on lots of different things during the course of the deal. Again, maybe we get there eventually, but having AI be able to understand that global context and have that situational awareness of the overall transaction to be able to advise at that sort of way, we're far away off from that, but we'll see.

I'm writing my prompt down. Take on the persona of Aaron Binstock at Cooley and guide me through this deal. And then when I fall short, I'll call you and tell me to fix this problem.

You know where to find me.

Exactly. What is the craziest thing you've seen in M&A?

That's a good question. You didn't prompt me on this one ahead of time.

This is good. I get a real answer from you then.

I've seen deals that are at the signing table where a party decides to walk away when everything feels like it's fully baked and just at the very last second because they have a change of heart. I've seen big events like economic market changes that have led to odd results. You wake up and the world shut down because of COVID and all of a sudden that at least briefly impacts things in ways that are unexpected. Lots of smaller stuff, but I'll let you know if I think of any other ones. Yeah.

Next time I want a big dramatic story, but no, those are all good case in points. You can't predict the unpredictable when it comes to M&A. Aaron, thank you so much for taking time from doing your deals to help me become a better M&A scientist.

Yeah, it's my pleasure. Thank you for having me.

Fellow M&A scientists, let me know what you think of this. If you're interested in chatting more with Aaron, go bug him on LinkedIn. Mention M&A Science. He'll give you a discount on the first engagement. Maybe he will. We'll see. But reach out to him. I had a great time in this conversation and just, we've had a great relationship with Cooley. You guys hosted events over here at the office and just been a great partner for providing different educational resources. Also connect with me on LinkedIn. If you've gotten this far in the podcast, you definitely deserve to connect with me. I love hearing feedback in terms of what you like and didn't like about this. If there's things I could have done better, I'm open to the criticism. And if there's topics I haven't covered, reach out, let me know, but mention, mention something in the podcast. I get so much spam. Till next time, here's to the deal. Thank you for taking the time to explore the world of M&A with our podcast. We love hearing feedback. Tag us on a LinkedIn post, add a review on Apple Podcasts. We'd love to hear from you. If you need help standing up an M&A function or optimizing one that you already have, we're here to help. And if we can't help you, we probably know someone that can. You can reach out to me by email, kison@mascience.com, or you can text me directly at 312-857-3711. If you just want to keep learning at your own pace, visit mascience.com for a lot more content and resources. That's where you can also subscribe to our newsletter. Again, that's mascience.com. Here's to the deal. Views and opinions expressed on M&A Science reflect only those individuals and do not reflect the views of any company or entity mentioned or affiliated with any individual. This podcast is purely educational and is not intended to serve as a basis for any investment or financial decisions.