We Built DealMind to Remember What Actually Works in Negotiations

nikhil sathelli — Mon, 28 Sep 2026 14:48:53 +0000

Every negotiation creates experience. The difficult part is making that experience useful when the next negotiation starts.
That was the problem that led us to build DealMind.
Sales teams often have historical information about previous customers, objections, discounts, competitors, and outcomes. But simply storing those records does not mean that the experience becomes useful during the next conversation.
A salesperson may know that the company negotiated with a similar customer before, but finding that experience and understanding what actually worked can still require manual searching and interpretation.
We wanted to build something different.
DealMind is a negotiation intelligence system designed around one idea:
A completed negotiation should become useful experience for the next one.
The problem with treating every deal as new
Consider a salesperson negotiating a $100,000 deal.
The customer says the price is too high and asks for a significant discount. There is also competitor pressure.
The salesperson now needs to answer several questions:
• Have we negotiated with this customer before?
• What happened in those negotiations?
• What objections appeared?
• Which strategies were used?
• Did large concessions actually help?
• What happened with similar customers?
• How much confidence should we have in the available history?
A normal AI assistant can provide general negotiation advice.
But general advice is not the same as organizational experience.
We wanted DealMind to answer a more useful question:
“What can our previous negotiations teach us about this negotiation?”
Making memory part of the workflow
The current negotiation begins with structured information such as the customer, industry, segment, deal value, initial offer, counteroffer, requested discount, objection, competitor pressure, and contract length.
DealMind then uses Hindsight to recall relevant historical negotiation experiences.
The recall can consider the customer, normalized customer identity, segment, industry, objection, discount context, competitor pressure, and contract context.
This gives the current negotiation access to experiences that happened before it.
That distinction is important.
Without organizational memory, the system has to reason mainly from the current input.
With memory, the current input can be connected to previous outcomes.
Hindsight as organizational memory
One of the architectural decisions we made early was separating structured application state from long-term memory.
SQLite handles the structured application state.
Hindsight handles the long-term negotiation experience.
When a negotiation is completed, DealMind can retain information such as:
• Deal ID
• Customer
• Segment
• Industry
• Objection
• Initial offer
• Counteroffer
• Strategy
• Concession
• Competitor pressure
• Contract length
• Outcome
• Outcome reason
This means the system does not simply store what happened for reporting purposes.
The experience can become something that future negotiations can retrieve.
From memory to decision support
After Hindsight returns relevant experiences, DealMind processes them through deterministic analysis.
The system calculates the economics of the current negotiation and determines confidence using explicit rules.
Groq then helps synthesize the available evidence into understandable negotiation guidance.
We deliberately separated these responsibilities.
Hindsight provides historical experience.
The application provides deterministic calculations and confidence.
Groq turns the supplied information into readable guidance.
The salesperson makes the final decision.
This also gives us an important safety boundary: the language model should not invent historical deals, statistics, confidence values, or evidence IDs.
The recommendation should be based on information that the application actually retrieved and calculated.
The learning loop
The most important part of DealMind happens after the recommendation.
Once a negotiation is completed, the salesperson can record the actual outcome.
DealMind can then retain that outcome in Hindsight.
The workflow becomes:
Current negotiation
↓
Recall historical experience
↓
Analyze
↓
Recommend
↓
Salesperson chooses a strategy
↓
Negotiation outcome
↓
Retain outcome
↓
Use the new experience later
This creates a feedback loop between past and future negotiations.
Learning from failures too
We also did not want memory to mean “remember only successful deals.”
A failed negotiation can be useful.
If a large concession was made but the deal was still lost, that is important experience.
If a smaller concession combined with additional value resulted in a successful outcome, that is also useful.
The objective is to build a useful representation of organizational experience rather than a collection of success stories.
More than a chatbot
This is where DealMind differs from a generic chatbot.
A chatbot can answer:
“How should I negotiate this customer?”
DealMind is designed to answer something closer to:
“Given this customer, this negotiation context, and what happened in comparable negotiations before, what does our organizational experience suggest?”
That changes the workflow from:
Question → AI answer
to:
Current deal → Memory → Evidence → Analysis → Guidance → Outcome → New memory
Keeping humans in control
DealMind is decision support, not autonomous decision-making.
The salesperson can inspect the historical evidence, understand the recommendation, compare possible strategies, and decide what action makes sense.
That was important to us because organizational memory should help people make better-informed decisions without hiding the reasoning behind an AI-generated answer.
Conclusion
We built DealMind around a simple principle:
A negotiation should not end when the deal ends. Its experience should remain useful.
Hindsight provides the long-term memory layer that connects previous negotiations with future ones.
The application handles structured data and deterministic analysis.
Groq helps synthesize the supplied evidence.
And the salesperson remains in control.
The result is a negotiation workflow where experience can accumulate over time.
DealMind doesn't just remember what happened. It changes what it recommends next.

DEV Community: nikhil sathelli

We Built DealMind to Remember What Actually Works in Negotiations