Skip to main content

Prompt Architecture

AI Win Strategy System

A 5-phase structured prompt architecture that compresses weeks of competitive research and win strategy development into hours. Built for strategic pursuits at a global IT services company, now being built into standard process for large deals worldwide.

Jonathan Lasley

Jonathan Lasley

Fractional AI Director

5-Phase ArchitectureState ManagementAnti-HallucinationWin Theme Validation

At a Glance

Weeks → hours

Time Compression

Full competitive research and win strategy in a single workflow

$20M–$500M

Deal Size Range

Built for strategic pursuits with 5–15 person teams

3 global teams

SVP-Presented

Now being built into standard process for strategic pursuits worldwide


The Challenge

Strategic Deals Deserve Better Intelligence

As a Solution Director for strategic engagements at Atos, a global IT services company, pursuit teams of 5–15 people would spend weeks assembling competitive intelligence and win strategy for $20M–$500M+ deals. The intelligence existed, scattered across analysts’ laptops, tribal knowledge in senior people’s heads, competitor data in various systems. The data existed. Assembling it consistently was the bottleneck.

Manual research produced different results depending on who did it. Two analysts working the same competitive set came back with different conclusions, not because one was wrong, but because human attention is selective. Win themes got debated in 2-hour meetings without structured evaluation. By the time a strategy made it into a proposal, the original research was weeks old and partially forgotten.

The real cost wasn’t the research hours themselves. It was the strategic decisions made on incomplete, inconsistent data on deals worth $20M to $500M or more.


The Approach

5-Phase Prompt Architecture with State Management

Decision-makers: the business results are in the next section. This section is the technical detail your team will want to evaluate.

I built a 5-phase prompt architecture where each phase has defined inputs, outputs, and validation criteria. The system turns unstructured deal information into a pressure-tested win strategy with full traceability from source data to final recommendation.

Phase 0: Deep Research Foundation

Before running the prompt sequence, generate a comprehensive deep research report on the client using AI-assisted search. Corporate strategy, leadership changes, financial signals, technology stack, industry position. Grounded facts up front prevent hallucinations downstream.

Phase 1: Intake & Qualification

Extract structured intelligence from RFx documents: a requirements snapshot with win strategy implications, evaluation criteria and weights, early risks, and known unknowns with resolution actions. The system uses an “intake gate” pattern, refusing to proceed without required inputs.

Phase 2: Client & Stakeholder Research

Research decision-makers, pain points, and delivery constraints. Build a stakeholder map with influence levels. Every stakeholder requires evidence: LinkedIn, official bio, press release, SEC filing. No evidence means no stakeholder in the strategy.

Phase 3: Competitor Analysis

7-section battlecards per competitor: positioning, strengths with evidence, vulnerabilities with evidence, proof points to neutralize, attack angles, defensive plays, and likely pricing posture. Every claim gets a confidence tag: [HIGH] multiple sources, [MEDIUM] single source, [LOW] limited data, [TBD] needs validation.

Phase 4: Win Strategy Synthesis

Win themes (3–6) with full traceability. Every theme must map to at least one RFx requirement, one stakeholder priority, and one verifiable proof asset. Discriminators name the specific competitor they beat. Ghosting angles cite specific vulnerabilities from Phase 3.

Phase 5: Document Compilation

Assemble everything into final deliverables with anti-summarization controls. AI’s natural tendency is to “helpfully” compress research. Explicit controls prevent that, preserving the analytical depth that makes the output valuable.

State Management

Structured JSON files accumulate data across phases. Each phase reads the previous state, adds its findings, and outputs an enriched file. AI has no memory between sessions. This pattern engineers memory through structure.

Anti-Hallucination Framework

Hard prohibition on inventing names, statistics, dates, or quotes. Inference tagging throughout: direct evidence is treated as fact, reasonable inference gets tagged, and hypothesis gets flagged for validation. The system prefers silence over speculation.

Win Theme Validation Stress Testing

Every win theme gets stress-tested across 4 dimensions so leaders can be confident in the strategy. Resonance: does it matter to the client? Differentiation: can competitors match it? Evidence: can we prove it? Consistency: does it fit what they already know about us? Themes that fail get reshaped or discarded.

5-phase win strategy architecture showing deep research foundation, intake qualification, stakeholder research, competitor analysis, win strategy synthesis, and document compilation with state management

The Results

From Research to Strategy in Hours, Not Weeks

I presented this system to peers and leaders of sales and presales across large global teams. It’s now being built into standard process for large strategic deals worldwide.

Pursuit teams now focus on strategy refinement and stakeholder engagement instead of data gathering. Win themes are pressure-tested across 4 dimensions, eliminating the 2-hour debates over unvalidated claims that used to slow down every major pursuit.

Output quality is consistent: the tenth run produces the same analytical depth as the first. With manual research, quality depended on who did the work and how much time they had that week. I’ve run the system across multiple markets and industries with the same results.

Comparison of manual competitive research process taking weeks with inconsistent outputs versus the AI win strategy system producing structured, traceable strategy documents in hours

Why This Matters for Your Business

Any Knowledge-Intensive Workflow

The architecture isn’t specific to sales or presales. The same patterns apply to M&A due diligence, market entry analysis, vendor evaluation, RFP response, and regulatory compliance research. Really any workflow where teams spend more time gathering and organizing information than acting on it.

Mid-market companies don’t have a dedicated competitive intelligence team. Leadership makes strategic decisions based on whatever research someone had time to do between other responsibilities. A structured AI system doesn’t replace human judgment, but it ensures that judgment operates on complete, consistent data instead of whatever was convenient to gather.

The prompt architecture behind this system was developed and refined using PromptAssay, a prompt engineering workbench I built specifically for this kind of multi-phase AI workflow development.


Key Takeaways

5-phase architecture with defined interfaces

Each phase has specified inputs, outputs, and validation criteria. This structure is what separates production AI workflows from one-off prompting.

Anti-hallucination is engineering, not hope

Confidence tagging, verification checklists, and source requirements are built into every phase. A shorter, high-confidence output is always preferable to a longer, speculative one.

State management solves AI’s memory problem

Structured JSON files carry context across sessions. This pattern applies to any multi-session AI workflow, not just competitive research.

The methodology is transferable

The specific prompts matter less than the architecture. The same 5-phase pattern with validation and anti-hallucination works for any structured research workflow.


Spending More Time Gathering Data Than Acting on It?

I build structured AI systems that compress weeks of research into hours. The same architecture that powers strategic deal intelligence works for market research, vendor evaluation, and competitive analysis.

Take the Free Assessment