This week, I share my reaction to and review my negotiation with the open-source M&A Negotiation Agent AI model.

AI will change the world, but how will it change M&A? I want to focus on AI’s impact on M&A in this newsletter. I am not an expert on either M&A or AI, but I want to learn about both topics and how they intersect. I thought there might be others in my situation (or people who are experts in one field or the other) who would find information on M&A and AI helpful in their careers, so I created this newsletter to track and share what I learn.

Reaction and Review of the Senior Level AI M&A Negotiation Agent

A few weeks ago, I discovered the M&A Negotiation Agent available on HuggingFace. It is an open-source AI model that simulates a negotiation between a buyer and a seller engaged in an M&A transaction. Here are links to my previous posts about the negotiation agent: Full senior-level negotiation; full junior-level negotiation; basic overview of the M&A Negotiations Agent.

In this post, I review and react to my negotiation with the “senior level” AI model. I recommend skimming the senior-level negotiations before continuing!

Here are some of my conclusions and reactions to the negotiation.

The difference between the senior level and the junior level negotiations

• The junior-level instructions state that negotiations aim to establish the types of representations and warranties. The junior-level instructions add that “advanced topics” include caps, escrow, baskets, and the survival period of reps and warranties.

• The app states that the senior-level negotiations focus on the cybersecurity incident and suggests that users focus on defining the type of reps and warranties, setting parameters of liability, negotiating the survival period, cap on liability, and baskets.

• I did not see a huge difference between the senior and the junior levels. In fact, the junior-level version immediately wanted to renegotiate the survival period, escrow amount, and damages cap. I will say that senior-level negotiations were slightly more complex than the junior-level ones because they focused on more technical aspects of the representations and warranties. There is also a chance I am using the model wrong and not negotiating the correct terms. This seems unlikely because I follow the app’s instructions, but a more experienced M&A practitioner may know additional terms to negotiate in this situation.

Approach to the senior-level negotiation:

• I took a more investigatory approach to this negotiation, focusing on asking questions about the deal and the cyberattack to gain information. I think this is a more realistic approach to negotiating.

• I took a more direct approach to the junior-level negotiations. This worked well but was probably not realistic (more on that later).

The negotiations:

• In the senior-level negotiation, I proposed negotiating the language of the reps and warranties, a damages cap increase, an escrow amount increase, and hiring a cybersecurity consultant. The model did not add any additional negotiation points. Because I am not yet an M&A expert, I am not sure whether other points should be renegotiated in a scenario like this one. What impressed me here is that the AI did not suggest additional areas, just like a counterparty in a similar situation would likely not suggest changing a deal to benefit the other side. On the other hand, perhaps there are additional points that the model could have proposed to benefit itself.

• I discuss the highlights of the negotiations on each point, below.

  Reps and Warranties language:

  • I tested the model by asking it to “send over the current reps and warranties language,” to which the model responded that it would send the language. Comically, it never happened. I even tried to rush it by saying “My client wants to renegotiate quickly!” (Probably not a great thing to say to a counterparty!).

  • Future iterations of this tool should focus the negotiation on contract language. This, after all, is a lot of what lawyers negotiate. A negotiation tool that can simulate turns of a document would be useful for training new M&A lawyers.

  Damages cap:

  • For the damages cap, I attempted to force the model to negotiate against itself by never mentioning my exact position. I would only say some variation of, “That is too low” and ask fact-finding questions on the model’s position. Funny enough, the model mistakenly thought I suggested a 50% damages cap in addition to a 50% escrow amount. I had to correct the model by saying, “My apologies, but I do not believe I proposed a damages cap of 50%.” It was interesting to see a hallucination in real-time.

  • I challenged the model by asking why the damages cap was so small when TechEase (the seller company) took remedial measures. In addition, I brought up how the model agreed that ShopMaster should not pay for TechEase‘s liabilities. Slowly but surely, the model increased its damages cap offer.

  • I do not think this is realistic. The model increased its offer substantially without much prodding. I never threw out a number for this part of the negotiations until the very end where I put pressure on the model to accept my offer.

  Escrow Amount:

  • I took a different approach to the escrow amount. I immediately started the negotiation by suggesting a figure that made sense to my client, and let the model respond. The model justified its refusal by stating that the TechEase owners needed immediate access to the funds so they could reinvest. I refused to accept that answer and questioned the model’s priorities (the founder’s liquidity vs. the soundness of the investment). The model eventually admitted that protecting the business from liabilities was more important than the founder’s liquidity, and began raising its offer.

  • Eventually the damages cap and escrow amount negotiations merged after the model proposed a 47.5% escrow (I wanted 50%) and a 60% damages cap. I pressured the model with a “final offer” of 50% escrow and a 60% damages cap, which the model accepted. It seems like the model is trained to react strongly to hard pressure from the user. I tried this technique multiple times, and each time the model accepted my offer.

  Overall conclusions

  Here are my overall conclusions on the senior-level negotiations:

  • The model negotiates against itself too easily. It never holds a position and freely increases its offer with little prodding from users. It seems like the default position of the model is to increase its own offer, which obviously is not true in real life.

  • The model did not answer my questions. I’m not sure whether it was a hallucination or on purpose, but I had to repeatedly ask it to answer my questions.

  • Having never played a substantive part in M&A negotiations, I am a little unclear on whether my styles of negotiating are realistic. It’s possible that I am too direct, which may cause problems in a real life negotiation. The direct style of negotiations works extremely well on the model, however.

  • A related point: The model is not emotional. Therefore, there are no consequences to an unreasonable proposal or rude remark. One potential solution is allowing the model to kill the deal. In the real world, deals die all of the time. It seems basically impossible for the model to walk away from the deal. It would be more realistic if walking from the deal was a possibility for the model.

Stay tuned for the partner-level negotiations in the coming weeks! The partner level is substantially more difficult than the senior level because the entire deal is negotiable.

About me

My name is Parker Lawter, and I am a law student pursuing a career as an M&A lawyer. I am in my last semester of law school, and with some extra time on my hands, I decided to create this newsletter. I hope it is informative and helpful to anyone who reads it! I am not an expert at either M&A or AI, but I am actively pursuing knowledge in both areas, and this newsletter is a part of that pursuit. I hope you’ll join me!

Follow me on LinkedIn: www.linkedin.com/in/parker-w-lawter-58a6a41b

All views expressed are my own!