{"slug": "designing-a-tool-that-uses-llms-to-negotiate-saas-prices-looking-for-feedback-on", "title": "Designing a tool that uses LLMs to negotiate SaaS prices. Looking for feedback on the architecture.", "summary": "A developer behind DealMind described an architecture for an LLM-powered SaaS price negotiation tool that organizes memory, evidence, strategies, what-if analysis, counteroffer guidance, and learning around a single object: the current negotiation. The design surfaces recalled Hindsight memories with evidence identifiers and explicitly flags when historical memory or the LLM is unavailable, using a deterministic fallback rather than pretending recall occurred. The workflow runs from analysis and recommendation through recorded outcome and learning, positioning the tool as a workspace that supports a salesperson's decision rather than an autonomous chatbot.", "body_md": "One of the easiest ways to make an AI application difficult to use is to turn every capability into another screen.\n\nWhile building DealMind, we wanted to avoid that.\n\nThe product has memory, evidence, strategies, what-if analysis, counteroffer guidance, customer history, and learning.\n\nBut the user should not feel like they are navigating seven different applications.\n\nThe solution was to organize everything around one object:\n\nthe current negotiation.\n\nThe main workflow is:\n\nDashboard\n\n↓\n\nNew Negotiation\n\n↓\n\nAnalyze\n\n↓\n\nRecommendation\n\n↓\n\nStrategies / What-if / Counteroffer / Evidence / Customer\n\n↓\n\nRecord Outcome\n\n↓\n\nLearning\n\nThis keeps the user's mental model simple.\n\nThe user starts with a deal.\n\nEverything else exists to help with that deal.\n\nThe New Negotiation screen captures information that matters to the negotiation:\n\n• Customer\n\n• Industry\n\n• Segment\n\n• Deal value\n\n• Initial offer\n\n• Customer counteroffer\n\n• Requested discount\n\n• Objection\n\n• Competitor pressure\n\n• Contract length\n\nThe user can enter their own negotiation.\n\nThere is also a demo scenario for quickly understanding the workflow.\n\nThe demo customer is not hardcoded into the product's identity.\n\nThe system should work with arbitrary customers and negotiation values.\n\nAfter analysis, the user enters the deal workspace.\n\nInstead of creating separate permanent pages for every feature, the current negotiation contains tabs such as:\n\n**Overview Strategies\nWhat-if\nCounteroffer\nEvidence\nCustomer**\n\nHindsight should not be hidden behind the backend.\n\nOne of the UX goals was to make memory visible when it matters.\n\nThe user should be able to see that the recommendation is based on historical experience.\n\nEvidence identifiers help connect the recommendation to recalled memories.\n\nThis gives the user a way to understand:\n\n**What did the system remember?**\n\nand:\n\n**Why did that memory matter?**\n\nA professional AI interface also needs to communicate when it does not have enough information.\n\nIf Hindsight is temporarily unavailable, the system should not pretend that it recalled historical negotiations.\n\nInstead, the interface should indicate that historical memory is unavailable.\n\nSimilarly, if the LLM is unavailable and the application uses a deterministic fallback, that should be clear.\n\nWe considered this an important UX principle:\n\n**Uncertainty should be visible rather than hidden.**\n\nThe same design philosophy applies to the short product demo.\n\nThe demo should tell one story:\n\nLoad a negotiation.\n\nAnalyze it.\n\nRecall Hindsight memory.\n\nShow evidence.\n\nShow confidence and economics.\n\nExplore strategies.\n\nRecord the outcome.\n\nShow the learning update. The user should be able to understand the entire memory loop without navigating through unrelated screens.\n\nThe learning feature also changes how we think about the product.\n\nIf the application only produced a recommendation, the workflow would end at the recommendation.\n\nBut DealMind continues:\n\n**Recommendation → Outcome → Memory → Future recommendation**\n\nThat means the UX needs to make the outcome step visible.\n\nThe user should understand that recording the result is not just administrative work.\n\nIt is how the system gains another piece of organizational experience.\n\nAnother design decision was keeping technical system information separate.\n\nAPI health, memory configuration, and system status belong in Settings.\n\nThey should not dominate the main negotiation experience.\n\nThe salesperson's main concern is the deal.\n\nThe system's technical implementation should support that workflow without getting in its way.\n\nDealMind is designed to support the salesperson rather than replace them.\n\nThe interface presents evidence, calculations, strategies, and customer context.\n\nThe user decides what to do.\n\nThis makes the product feel more like a negotiation workspace than an autonomous chatbot.\n\nThe UX challenge in DealMind was not simply making the interface look good.\n\nIt was making the memory loop understandable.\n\nThe user should be able to see:\n\n**What the system remembered. Why it mattered.\nWhat the system recommends.\nWhat happened afterward.\nHow that outcome can become future memory.__**", "url": "https://wpnews.pro/news/designing-a-tool-that-uses-llms-to-negotiate-saas-prices-looking-for-feedback-on", "canonical_source": "https://dev.to/karunya18/designing-a-tool-that-uses-llms-to-negotiate-saas-prices-looking-for-feedback-on-the-architecture-35kf", "published_at": "2026-09-28 17:08:04+00:00", "updated_at": "2026-09-28 17:22:15.015651+00:00", "lang": "en", "topics": ["ai-products", "ai-agents", "large-language-models", "ai-tools"], "entities": ["DealMind", "Hindsight"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/designing-a-tool-that-uses-llms-to-negotiate-saas-prices-looking-for-feedback-on", "markdown": "https://wpnews.pro/news/designing-a-tool-that-uses-llms-to-negotiate-saas-prices-looking-for-feedback-on.md", "text": "https://wpnews.pro/news/designing-a-tool-that-uses-llms-to-negotiate-saas-prices-looking-for-feedback-on.txt", "jsonld": "https://wpnews.pro/news/designing-a-tool-that-uses-llms-to-negotiate-saas-prices-looking-for-feedback-on.jsonld"}}