How a small corp dev team uses AI, from blind teasers to document review

• Sellers Dorsey uses AI most in early deal work: market research, sourcing and preliminary diligence that used to take hours now take minutes.
• Matthew Golden says AI document review saved hundreds of hours of legal work on the firm's last acquisition, with no discernible errors.
• Golden puts AI at associate counsel level for issue spotting, not yet at higher level analysis, and the team uses no agents for legal work.
• Alex Abbott wants to feed context on both companies, plus the team's own deal postmortems, into AI to plan integration and catch decision biases.
• Protected health information and legal privilege set hard limits: no free tools or personal licences for confidential material, only enterprise tools with known terms.
• Both say the decision on whether to buy a business should stay with human leadership.

When a blind teaser lands at Sellers Dorsey, the corporate development team pastes it into a large language model and asks which company it thinks this is, based on the context in the teaser. Alex Abbott, the firm's associate director of corporate strategy and development, calls it something fun they have been playing with (around 1:13). It shows where AI sits for this team today: in the early grunt work of a deal, the market research, sourcing and preliminary diligence that used to take hours and now takes minutes.

“Something fun that we've been playing with, too, is whenever we receive a blind teaser, we will plug that into an LLM and say, 'Hey, which company do you think this is based on the context information in the teaser?'”

Alex Abbott, Associate Director of Corporate Strategy and Development, Sellers Dorsey

Sellers Dorsey is a PE-backed healthcare firm of about 300 people that has made acquisitions to build its product and technology side. Its in-house M&A team is small, Abbott says, and runs the whole process from deal strategy and sourcing to diligence and long-term integration. Matthew Golden, associate general counsel, heads up the due diligence. There is no dedicated M&A counsel, he says, so the deal work happens on top of the normal work. The session is part of DealRoom's AI for M&A Summer School, and both speakers use DealRoom.

“Issue spotting is where it's great at right now. Associate counsel level work in terms of issue spotting, but that next level analysis, we haven't moved far up the food chain on it.”

Matthew Golden, Associate General Counsel, Sellers Dorsey

Golden's clearest result is in document review. On the first acquisition a few years ago there was no AI-enabled tool, and all the review was very manual, very tedious. On later deals AI cut the majority of that time. He calls document review the lowest hanging fruit in legal work. Gone are the days, in his words, when you had to train it to recognise a certain clause or the phrasing had to be exact. On the last acquisition it saved hundreds of hours with no discernible errors (around 10:39).

“On the legal side, the document review was the fastest adoption. That saved hundreds of hours of legal work on our last acquisition. No discernible errors or anything like that.”

Matthew Golden, Associate General Counsel, Sellers Dorsey

He is just as clear about the ceiling. AI is great at issue spotting, at roughly associate counsel level, Golden says, but has not moved far up the food chain into the next level of analysis, and the team is not using agents for any legal work. Abbott uses it on the corporate development side to pressure test the investment thesis, asking what they haven't thought of yet and what gaps it sees in their rationale. Human judgement, Abbott says, should come before anything you put into AI.

“We could get to a point where AI could answer almost everything analytical except for the question should we do this deal or not.”

Alex Abbott, Associate Director of Corporate Strategy and Development, Sellers Dorsey

Both draw a hard line at the decision itself. Abbott says this is a people job first and foremost. Selling your business is a scary thing, and getting a founder through saying yes, then diligence, then being absorbed by a much larger company takes relationship building and trust that Abbott doesn't see AI ever replacing. The call on whether to buy should remain in the hands of human leadership. Abbott hopes AI could one day answer almost everything analytical except that question (around 6:28).

The bigger upside, in Abbott's view, is integration. AI thrives on context, Abbott says, so the team could build a persona from everything it can provide about its own organisation, gather as much context as possible on a target, feed both in and ask how it would reimagine the org chart or roll out an integration timeline. The team has also written up lessons learned and postmortems from past deals. Abbott wants to feed those in too, so the model acts as a checker that catches decision biases and doesn't let anything slip past the goalie.

The risks are specific to this firm. Recent acquisitions brought it into protected health information, in one of the most regulated industries, so any tool has to be deployed very deliberately, safely and securely, Abbott says, and staff need to know not to put PHI into AI tools. The firm is rolling out Claude across its roughly 300 employees, and Golden says Anthropic provided a high trust compliant version for enterprise use. He also flags privilege issues with LLMs if you don't have an enterprise licence or an agreement you are comfortable with (around 3:45).

Adoption at the firm is voluntary and, in Abbott's words, very person differentiated, with early adopters on one side and the more reticent on the other. Golden sees the same split across the legal profession, enthusiasm on one hand and hesitation over privacy, confidentiality and privilege on the other. He calls it a K-shaped adoption. The advice from both starts with paperwork. Read your organisation's AI policy, Abbott says, and if there isn't one, get one today. Golden adds: read the terms of service, never use free AI tools or a personal licence for anything confidential, and ask where the boundaries are.

Golden's warning about privilege and personal licences applies to anyone reviewing data room documents with AI. Before a deal team runs a target's contracts through a model, check which tool it is, under which licence, and what its terms say about the data.

0:09 I'm Alex Abbott. I'm Sellers Dorsey's Associate Director of Corporate Strategy and Development. Sellers Dorsey is a healthcare impact firm made up of former hospital executives, state Medicaid directors, and health plan presidents. We combine healthcare financing and regulatory expertise to help healthcare providers uncover new revenue opportunities. We also provide strategy consulting to state Medicaid agencies, health plans, and health vendors. But, we spent the past few years making strategic acquisitions to stand up new product and technology functions for the firm and launch our first ever technology product to help providers with population health and value-based care enablement. We are a PE-backed mid-sized organization with a small but very mighty in-house corporate development team that includes myself as well as an executive sponsor, and we manage the entire M&A process end-to-end. So, that includes deal strategy, sourcing, due diligence, and long-term integration management. To do all of this, we rely very heavily on our tools like DealRoom and AI, but we

1:13 really rely on our in-house expertise, especially our legal team. So, it's my pleasure to introduce you to my colleague, Matt Golden. >> As Alex said, I head the legal team at Sellers Dorsey, Associate General Counsel. I am one of the points of contact and leaders on the our M&A team, and I pretty much head up the due diligence and then work with getting the deal structure, the LOI, and all through the process through integration and kind of beginning to end. Pleasure to be here. >> We're leveraging AI the most right now in the early phases of deals, market research, sourcing, early preliminary diligence on a target when we find one that is a potential fit for us. That kind of grunt work that used to take so much time for us is now something that can take place in minutes as opposed to hours. Something fun that we've been playing with, too, is whenever we receive a blind teaser, we will plug that into an LLM and say, "Hey, which company do you think this is based on the context information in the teaser?"

2:17 From the corp dev side, we use it to pressure test our investment thesis. So, like thought partnership I I heard another speaker mention that it'll never take the place of judgment and that human judgment should be the precursor to anything that you're putting into AI, but we certainly use it to say, "Hey, what haven't we thought of yet? What gaps are you seeing in our rationale based on the investment thesis and the information that we've provided?" I'm curious to hear how Matt's been using it on the legal side. >> Our first M&A that we did just a few years ago, we didn't have an AI-enabled tool. We don't have dedicated M&A counsel, so we're doing this work kind of on top of our normal work. All of our document review, due diligence, all that stuff was very manual, very tedious.

3:00 Subsequent acquisitions through deal room's AI enablement, we've been able to cut the majority of that time down that was spent on document review, which is right in the wheelhouse of where AI is right now. >> This is a people job first and foremost. Selling your business is a scary thing and it takes so much trust to get somebody to to come along through the entire process of selling their business. That includes saying yes to the idea in the first place and then going through the crazy process of due diligence and and then finding yourself absorbed by a much larger company. That takes so much relationship building and trust building that I don't see AI ever replacing.

3:45 >> Right. >> And then of course there's the ultimate judgment call of should we buy this business or not? I really hope that people do not use AI to make a decision with as much impact as that one. I think that should remain in the hands of human leadership at your organization. >> Issue spotting is where it's great at right now. Associate counsel level work in terms of issue spotting, but that's kind of like that next level analysis like we haven't moved far up the food chain on it. There's privilege issues too with using LLMs, especially if you don't have an enterprise license and you don't have an agreement in place that you're comfortable with or something like you know, you're using deal room, you know, with any other kind of agreement. You got to like know what the rules are and what the lay of the land is and be comfortable with how you're using it in terms of compliance.

4:35 Document review is the easiest like the lowest hanging fruit for what it could do for in the legal field. Hundreds or thousands of documents, it's improving exponentially in its capabilities both in terms of the amount of documents it can review in one query and also in terms of like the quality of the results. Gone are the days where you have to like train it to recognize a certain clause or the phrasing has to be exact. In terms of batch review of of large amounts documents, it's really good. >> There's so much potential to cut through all of that the clutter of the the high volume of document review during due diligence, but I think that it also stands to have a huge impact on integration. AI thrives on context, right? So I can envision a world in which we are creating a seller's due diligence persona based on all of the context we can provide on our organization and then gathering as much context as we can from an organization that we plan to acquire and feeding both

5:38 of those context documents into AI and saying, "Hey, how would you reimagine this org chart? Hey, how would you roll out an integration timeline given what you know about these two organizations and how you think they might or might not mesh together? Diving deeper into the importance of context, we have done a ton of learning on the job with our small scrappy M&A shop and fortunately as a result of that, we've done a ton of documenting on lessons learned. We've done a ton of postmortems where we can gather information on what went well and what didn't. I envision a future in which we're feeding all of that information into AI so that it can act as a checker to help us not repeat past mistakes, catch any decision biases that we may have been guilty of in the past.

6:28 Be that checker that knows us so well that it's not going to let anything slip past the goalie on a deal. We could get to a point where AI could answer almost everything analytical except for the question should we do this deal or not. >> Right. >> And of course the answer to that question is still going to require corporate development expertise, legal expertise, financial operational expertise, but my dream is that we get to a place where all we have to do as those practitioners is say, yes, we should or no, we should not. We are in the process of rolling out Claude to our enterprise as we speak. We are a firm of about 300 FTEs. So, rolling out a tool like that is no small feat. I should note too that our most recent acquisitions have gotten us involved with PHI, which is protected health information. It's huge in health care and as you guys know, health care is one of the most regulated industries in the country. Any tool that we deploy,

7:34 whether it be AI or otherwise, needs to be done very deliberately, safely, and securely. It's really important that the people who are getting access to AI tools in our organization know how to use them properly and safely and know not to put things like PHI into the tools. >> I would definitely not put PHI into an LLM, yeah. Anthropic was able to provide like a high trust compliant version of Claude for our enterprise use. The biggest barrier is the not sure where to start part. Training could help with that and I think also like the you know the change management part cuz at least at Sellers Dorsey AI use and adoption is voluntary. It's certainly not prescribed in any of our workflows. There are some members of our team that are early adopters and there's others that are more reticent. Training would be great way to kind of increase adoption. And Deal Room's been great in helping us use tools as they've rolled them out.

8:32 >> What's some honest advice to the teams that want to get there but again don't know where to start? >> Read your organization's AI policy. If your organization does not have an AI policy, they need to get one today. Again, you want to make sure that you're not compromising any sensitive company information or I mean even more importantly sensitive client information. That would not be a great way to get started using AI. So please check your organization's policy first. Secondly, I would encourage everyone to think imaginatively about what AI can do. It's introducing a completely new way to think about what work is. I think really the sky is the limit when it comes to ways for you to reimagine how you get work done every day. Think big about what AI can do for you, not just the small like help me write an email but big like reimagining integration, making a persona. It really is a magical tool that can do so so much and we're just in the early phases of it. And then

9:35 thirdly, I mean you guys know the old expression of the the best time to plant a tree was 20 years ago, and the second best time to plant a tree is today. I I think the same applies for AI. I talk to people who feel like they're behind or feel like their organizations are behind in AI adoption. This is new to all of us. You're not behind, but start today. This is the time to learn this technology, get really comfortable with it because it is only going to grow from here, and just get curious and have fun with it and explore it. >> I have 80 hours worth of projects on my to-do list on any given day, and I'm just hoping that the pile doesn't get bigger by the end of the day. AI could help shorten the time it takes to do projects. The next phase would be is workflow management, helping decision point quicker as opposed to just issue spotting. Legal field is on one hand very enthusiastically embracing AI, and on the other hand, I think as people talk about privacy and confidentiality,

10:39 privilege, very, you know, kind of reticent and hesitant. It is kind of aiming at turning into like a K-shaped adoption across the legal field, which is really interesting. So, it changes so fast, I'm not sure I don't know that anybody could say where we'll be in 12 months. So, I'm just excited to be on along for the ride. On the legal side, the document review was the fastest adoption. That saved hundreds of hours of legal work on our last acquisition. No discernible errors or like anything like that. It was It was really It was well done. It saved us tons of time. the holding out stuff is the, you know, the people said, or you know, we said it earlier, it's it's kind of the higher up on the food chain stuff. We don't I don't know that tools exist yet to do it, but you know, we're not using agents to do any legal work or anything like that.

11:30 >> For us, it's it's been less about which functions are adopting it fastest, but more about which individuals are adopting it fastest. Like as Matt mentioned, we don't have an AI mandate at the firm. It's been a lot more, hey what makes sense for you? Personally, I don't like to spend my days on busy work that can be done by AI. So I have been an early adopter. Matt here has been an early adopter as well, but it's it's been very person differentiated for us from what I've seen. Not yet. We've used deal room for this purpose. I love that you can link findings to to integration, and that's something that we want to lean into more going forward with our deals, but I don't see why we wouldn't adopt AI going forward for this purpose if it meets our needs.

12:18 >> Yeah, I would agree with Alan. We don't have any tools, AI based tools, outside of deal room that we use in our M&A process. As satisfied customers with deal room, we will I'm I'm I will anxiously attempt to try to use any new tools as they come out and see how they work. >> No one likes to read the fine print, but you know, I would look at your terms of service. I would not use any, you know, certainly wouldn't use any free AI tools, or even you know, I wouldn't use even like a personal license to put anything in that would be anything to do with confidential information or privacy. If your company doesn't have an enterprise AI tool, you should ask about that and find out what the boundaries are. And anything any price tool that you are using, check what the terms of service are. Make sure you're satisfied.