The Challenge of Character Limitations
Now, the second problem we have to tackle is how to distill a comprehensive assistant into less than 10,000 characters for Microsoft Copilot. Let's examine how Deepak Negotiator, our very comprehensive assistant, works. I've already uploaded the usual role play instructions for the seller and provided a very simple prompt: "Please analyze the touch role play."
I'm using Gemini 2.5/preview, and here's what we get back.
The system provides a complete negotiation preparation sheet based on the relationship assessment. It notes they have no prior contacts and determines:
Our interests
Their likely interests
The internal stakeholders
Issues and options
Objective arguments
Alternatives
BATNA power leverage
Aspirations
Questions
Opportunities to make trades
Negotiation strategy
The analysis also explores plausible scenarios for the other party. We receive three different scenarios, along with anticipated key objections and prepared responses. For example, addressing objections like "Your price is too high" with responses about shipping costs, the age of some books, and how loose-leaf services require ongoing updates which are costly.
It even provides additional creative options for value creation and transparency in the process, explaining which steps have been taken. This analysis is extremely comprehensive.
What we did next was distill these comprehensive system instructions into new ones that we can use in Microsoft Copilot. The character count has been reduced from 78,000 to 9,100. Microsoft Copilot Web has a limit of 10,000 characters.
We should consider doing another distillation to make it suitable for ChatGPT projects and custom GPTs, as they have a limit of 8,000 characters.
Let's see how our distilled version works with Copilot Web. We've attached the law firm seller instructions and again used a very simple prompt, but now we're using our condensed system instructions.
The results are still impressively comprehensive. We receive information about:
Who we're negotiating with
The objective
Our interests
Their interests (both likely and certain)
Issues and options
Creative resolution options
Objective arguments
Book conditions, urgency, and logistics
Addressing likely objections
Alternatives and power dynamics
Aspirations
Questions to uncover key information
Opportunities for trades
Negotiation strategy (including style, approach, and key tactics)
We can still explore more ideas within this condensed framework. For instance, when I ask "Could you provide more creative options to enlarge the pie?" the system provides:
Bundled knowledge access
Structuring deals for long-term benefit with tiered payment options
Subscription models
Future referral arrangements
Logistics and efficiency enhancements
As you can see, even with a limited character count, we've created system instructions that remain comprehensive and effective. They successfully analyze, explore, and provide a full preparation framework before entering negotiations.
Thank you for listening. If you're interested in AI and negotiation, please subscribe to our newsletter at negoai.ai.
Now, the second problem we have to tackle is how to distill a comprehensive assistant into less than 10,000 characters for Microsoft Copilot. Let's examine how Deepak Negotiator, our very comprehensive assistant, works. I've already uploaded the usual role play instructions for the seller and provided a very simple prompt: "Please analyze the touch role play."
I'm using Gemini 2.5/preview, and here's what we get back.
The system provides a complete negotiation preparation sheet based on the relationship assessment. It notes they have no prior contacts and determines:
Our interests
Their likely interests
The internal stakeholders
Issues and options
Objective arguments
Alternatives
BATNA power leverage
Aspirations
Questions
Opportunities to make trades
Negotiation strategy
The analysis also explores plausible scenarios for the other party. We receive three different scenarios, along with anticipated key objections and prepared responses. For example, addressing objections like "Your price is too high" with responses about shipping costs, the age of some books, and how loose-leaf services require ongoing updates which are costly.
It even provides additional creative options for value creation and transparency in the process, explaining which steps have been taken. This analysis is extremely comprehensive.
What we did next was distill these comprehensive system instructions into new ones that we can use in Microsoft Copilot. The character count has been reduced from 78,000 to 9,100. Microsoft Copilot Web has a limit of 10,000 characters.
We should consider doing another distillation to make it suitable for ChatGPT projects and custom GPTs, as they have a limit of 8,000 characters.
Let's see how our distilled version works with Copilot Web. We've attached the law firm seller instructions and again used a very simple prompt, but now we're using our condensed system instructions.
The results are still impressively comprehensive. We receive information about:
Who we're negotiating with
The objective
Our interests
Their interests (both likely and certain)
Issues and options
Creative resolution options
Objective arguments
Book conditions, urgency, and logistics
Addressing likely objections
Alternatives and power dynamics
Aspirations
Questions to uncover key information
Opportunities for trades
Negotiation strategy (including style, approach, and key tactics)
We can still explore more ideas within this condensed framework. For instance, when I ask "Could you provide more creative options to enlarge the pie?" the system provides:
Bundled knowledge access
Structuring deals for long-term benefit with tiered payment options
Subscription models
Future referral arrangements
Logistics and efficiency enhancements
As you can see, even with a limited character count, we've created system instructions that remain comprehensive and effective. They successfully analyze, explore, and provide a full preparation framework before entering negotiations.
Thank you for listening. If you're interested in AI and negotiation, please subscribe to our newsletter at negoai.ai.
