From Questions to Action - Tackling Value Leakage in the Era of AI

Introduction

As large language models (LLMs) begin to reshape many aspects of business, procurement teams and commercial leaders are asking probing questions, such as, “What does this mean for us?” How do we make it safe? Can we trust the outputs? And how do we act on them - responsibly and at scale? 

At Digital-Mirror, these questions have shaped our approach to Contract Performance Management (CPM). At its heart, CPM is about reconnecting contracts to the business purpose behind them—the “why” they were signed—and ensuring that purpose is actually fulfilled. That’s what drives our solution: helping you identify what matters in your contracts, pinpoint where value is being lost and take confident steps to rectify it. 

To make that happen and to address the rest of those probing questions, we rely on a framework we call QAQA, which stands for Question, Answer, Quality, Action. It’s a structured approach that supports reliable decision-making by making quality the foundation, not an afterthought. 

This article builds on our earlier piece, where we explored the shift from Contract Analytics to CPM. Now, we go deeper: how do we ensure that insights translate into meaningful, quality-assured action? 

Dealing with AI Reality: Multiple Models, Variable Costs

The LLM landscape is evolving rapidly, with new models, changing pricing and shifting performance profiles. The rapid pace of change is one of the concerns that prompts questions about the safety and trustworthiness of AI-based solutions. 

To address this, we designed our system to work reliably across different AI models, ensuring that you receive consistent answers regardless of the model used (GPT, Claude, or another). We achieve this by carefully controlling what the AI sees, how it’s asked, and how the answers are verified, ensuring the results remain stable and trustworthy as the technology evolves. 

Automation with Guardrails

Automation holds real promise for contract processes, but without proper checks and assurances, it can introduce new risks as it attempts to resolve existing ones. We only automate actions once three conditions are met: 

• The question is well-formed (diagnostic or prescriptive).

• The answer is complete and consistent across models.

• The answer passes quality thresholds we’ve built into our system. 

These thresholds are calibrated not just on model confidence scores but also on benchmarking against industry norms and business expectations. This combination gives the green light for action or pauses it for human review.  

Trust, Quality and Performance

This focus on trust, quality and performance is echoed in McKinsey’s recent report, Mitigating Procurement Value Leakage with Generative AI, which highlights contract optimization and contract compliance as two of the biggest drivers of unrealized value.  

McKinsey also points to the role of AI in turning this around, not by replacing people, but by scaling tasks that humans struggle to manage at volume: monitoring compliance, reconciling invoices to contracts, and surfacing errors. These are precisely the kinds of issues our QAQA framework was designed to tackle - ensuring AI is used not just for insight but for action, backed by the assurance that quality and consistency are in place from the start. 

QAQA: The Route to Quality Answers You Can Act On

QAQA is the loop that makes all of this work: 

• Question: Sourced from business stakeholders, legal libraries, or inferred from risk/value events.
• Answer: Generated using LLMs + structured data fusion from ERP, CLM, and financial systems.
• Quality: Assessed via model agreement, business relevance, and explainability checks.
• Action: Only taken once the answer is proven reliable, whether it’s surfacing a missed early-payment discount or flagging a clause misalignment. 

This is automation in service of performance, delivering not just answers but actions that protect margin and help unlock contract value. 

Industry Knowledge as a Living Layer

Solutions that act as a wrapper to generic AI won’t succeed. The quality of insight, and ultimately action, comes from embedding domain-specific knowledge into the QAQA process.

We work with industry experts to define meaningful questions and acceptable answer patterns. This means the system can self-learn over time, getting smarter as more contracts, performance data, and user interactions are ingested. It’s a way of capturing institutional experience (also known as tribal knowledge), allowing the system to improve over time and making decisions more consistent. 

Connecting the Dots Across Systems

Contract performance is reflected in payment terms, Service Level Agreements (SLAs), usage data, and supplier obligations, which are spread across ERP, finance, and procurement systems.

Our approach integrates contract metadata, structured performance data, and contextual diagnostics into a single view. By combining internal and external data (e.g., ERP terms versus contract commitments), we provide a comprehensive view of contract health, enabling more proactive management.

This also means we’re not just answering “What’s wrong?” but “Why is it happening?” and, crucially, “What should we do next?” 

Conclusion: Why This Matters

Contract Performance Management must extend beyond measurement to assurance, action, and accountability. QAQA is our way to ensure this, even in a world where models are updated monthly, and costs are difficult to predict.