From data to decisions: Closing the AI value gap

You've launched your AI Customer Agent. You're tracking KPIs. You have data. But here's the uncomfortable truth: only 5% of generative AI pilots deliver sustained value at scale¹ – and only 20% of organizations have even appointed someone responsible for value realization². The gap between measurement and action is where most AI investments go to die. This article closes that gap with a structured framework for translating data into decisions to secure value from your AI Customer Agent. 

Closing the circle: Making investment decisions based on the value of your AI Customer Agents

After developing your go-to-market strategy, defining the right analytics, and launching your AI Customer Agent, the critical question becomes: Is it delivering value? 

Stakeholders across your organization will inevitably ask: Are customers adopting our AI Agent? Are we realizing the expected returns? Is this investment justified?

In our previous articles, we explored how to set up modern analytics for effective insights; establishing KPIs, tracking performance across multiple dimensions, and building dashboards that give you visibility into your AI Customer Agent's performance. This article builds on that foundation by answering a more complex question: How do we translate success criteria and KPIs in the Value Blueprint into actionable insights that inform investment decisions?

We will walk through a structured decision-making framework, that enables you to confidently answer whether to enhance, de-prioritize, stop, or scale your AI Customer Agent investment, or parts of it.

Why value prioritization matters

Reason #1 – AI Customer Agents are multi-faceted

Unlike traditional software deployments, AI Customer Agents operate across multiple dimensions simultaneously. They span multiple use cases, involve interconnected design and technology choices, and must satisfy several stakeholder expectations.

This complexity creates a critical challenge for investment decision-making: you cannot optimize all dimensions at once. When you have multiple use cases running in parallel – each with different adoption curves, technical maturity levels, and customer satisfaction profiles – a single aggregate metric (e.g., "overall satisfaction at 72%,") can mask the real story: One use case might be delivering 85% satisfaction while another languishes at 58%. 

Value prioritization forces you to evaluate each use case individually and make differentiated investment decisions: scale the winners, enhance the promising ones, and stop the weak performers.

Reason #2 – Humans make investment decisions, which are not always objective

The second challenge is that without a data- and value-led prioritization approach, organizations default to suboptimal strategies.

The first trap: Investing equally across all use cases. When stakeholders are unable to agree on which use cases matter most, the default is to spread resources evenly. This dilutes impact. You end up with three mediocre use cases instead of one excellent one and one paused one. Resources are wasted on low-value initiatives because no one wants to make the hard call to stop them.

The second trap: Prioritizing based on performance dips or emotional reactions. A CFO sees a weak 3-week performance snapshot and demands the solution be stopped. A product leader falls in love with a technically sophisticated use case that customers do not actually want. Without a clear, documented framework for decision-making, emotional reactions can override data-driven reasoning, which can result in suboptimal investment decisions, whiplash changes in direction, and loss of stakeholder confidence.

Reason #3 – The gap between data and value

Tracking on data does not mean knowing how to make decisions. According to Deloitte's 2026 State of the Enterprise research, 60% of Nordic organizations track non-financial benefits from AI investments². Yet, the gap between desired and achieved benefits is 18pp. for cost reductions and 57pp. for revenue increases². 

Moreover, only 1 out of 5 organizations appointing a Value Lead for AI solutions². These findings point to a fundamental issue: organizations are tracking multiple KPIs (adoption, satisfaction, cost, revenue), but do not seem to have a systematic way to translate those KPIs into value. Should you scale a use case with 40% adoption and 75% satisfaction? Should you invest in enhancing a technically immature use case that customers love, or kill a technically sophisticated one that customers ignore?
 

The solution: Value prioritization as an enabler of data-driven decisions

Value prioritization is about leveraging performance data to answer the few – but critical – questions for decision-making:

• Which use cases are delivering the highest value relative to investment?
• Which use cases have the greatest upside potential?
• Which use cases are consuming disproportionate resources for minimal return?
• Where should you double down, and where should you pause or simplify?

But then the key question is: how do you translate multiple KPI measures into an informed decision without drowning in complexity?

Four steps to value prioritization 

In Deloitte, we take a structured, four-step approach that moves you from measurement to decision-making.

Step 1: Define the prioritization framework 

Before you can make decisions, you must establish a clear framework for how you will assess use case performance based on the KPIs defined in your Value Blueprint.

With your KPIs and success criteria in place, we suggest the Value Prioritization Matrix as a 2x2 framework that plots each use case across two dimensions:

Technical sophistication  (low to high), informed for example by:

• How mature is the underlying technology?
• How complex is the implementation?
• How reliable is the solution in production?
• How well does it perform against technical KPIs? (accuracy, latency, uptime, error rates)

User desirability  (low to high), informed for example by:

• How much do customers want this use case?
• How frequently is it being used?
• How critical is it to the customer journey?
• How well does it perform against desirability KPIs? (adoption, satisfaction, NPS, usage frequency)
   

This matrix produces four decisions, each suggesting a different action:

1. Scale what customers love and the technology supports,

2. Enhance what customers want and technology can further support

3. De-prioritize what has room for technical improvement but low desirability, and 

4. Stop what is technically sophisticated but unwanted.

Step 2: Design measurement methods

From our first Article in our AI Customer Agent series, you have established KPIs across multiple lenses: customer desirability, technical feasibility, and commercial viability. From our sixth article on Analytics, you have set up analytical tracking methods to track quantitative KPIs. 

Now you need to ensure your measurement approach also covers qualitative input, for example through surveys and interviews, and that it captures performance at both solution and micro-level.

2a: Solution-level metrics

These are the primary metrics you track for each use case, giving you visibility into overall performance:

• Adoption rate (% of target audience using this specific use case)
• Customer satisfaction (CSAT, NPS, CES for this use case)
• Cost savings or revenue impact (attributed to this use case)
• System performance and reliability (uptime, latency, error rates for this use case)

2b: Use case-level micro metrics

These granular metrics reveal the drivers of performance and enable root cause analysis, for example:

• Performance by customer segment (SMB vs. enterprise, new vs. existing customers).
• Performance by geography.
• Root cause analysis for underperformance (Why is adoption low? Why is satisfaction declining? Which customer segments are driving or dragging performance?). This will be informed by a comprehensive Value Blueprint (see article 1). For example, aggregate customer satisfaction is informed by share of customers who feel like it is easy to talk to the agent and share of languages supported by the agent.

Why is it important with a dual-layer approach? 

A single underperforming use case can mask overall success. Conversely, a weak macro metric for a use case might hide strong performance in specific segments.

For example, your "Virtual Agent for Product Recommendations" use case shows 58% satisfaction overall, which looks weak. But when you break it down by customer segment, you realize new customers show 72% satisfaction and existing customers show 48% satisfaction. This granular view enables targeted improvements (e.g., simplify recommendations for existing customers) rather than killing the entire use case.

Step 3: Create Decision-Making Processes & Governance

With only one out of five organizations appointing a value lead for AI solutions², this is perhaps the most overlooked step. Without a clear value realization process and ownership, a weak 3-week performance snapshot can trigger a "stop" decision, regardless of maturity stage.

3a: Establish decision gates and timelines

When will you evaluate performance? Different use cases may have different timelines based on expected adoption curves, for example:

• A 90-day gate can provide an initial assessment of customer reach and trial.
• A 6-month gate can provide an evaluation of adoption trajectory and early satisfaction signals.
• A 12-month gate can provide a ROI assessment and strategic decision on scale/optimize/pause/stop.

Document these gates upfront, to prevent stakeholders from demanding decisions before you have sufficient data and discuss the performance results in appropriate operational and strategic review forums.

Step 4: Make data-driven decisions on which use cases to scale, refine, or de-prioritize

With a clear framework, measurement approach, and governance process in place, you can now make confident decisions for each use case.

4a: The review process

Here is how to apply this framework in practice:

1. Gather data on all active use cases across solution and micro-level metrics.
2. Plot each use case on the Value Prioritization Matrix (technical sophistication vs. user desirability) – with technical sophistication typically informed via interviews with your technical experts.
3. Assess financial attractiveness for each use case.
4. Apply the decision logic:
   • Scale quadrant + Financially attractive → Scale investment.
   • Enhance quadrant + Financially attractive → Invest in improvement.
   • De-prioritize quadrant + Financially attractive → Invest in development.
   • Stop quadrant OR Not financially attractive → De-prioritize or stop.
5. Align stakeholders on decisions using the documented governance process.
6. Communicate decisions with clear rationale and next steps.

4b: Example decision scenario

Your AI Customer Agent has three active use cases:

• Product recommendation: High desirability, high sophistication, financially attractive → Decision: Scale. Expand to new customer segments, invest in personalization, measure ROI expansion.
• Multilingual customer support: High desirability, low sophistication, financially attractive → Decision: Enhance. Invest in improving accuracy, expand to new customer support requests, measure adoption growth.
• Quote generator: Low desirability, high sophistication, not financially attractive (ROI is negative) → Decision: Stop. Reallocate resources to higher-value use cases.

This framework removes emotion from the decision-making process and ensures resources are allocated to the highest-value opportunities.

Putting it all together: value realization

You have now completed the full value realization cycle for your AI Customer Agent:

• Article 1 established the Value Blueprint: Defining business objectives, success criteria, and KPIs across customer desirability, technical feasibility, and commercial viability.
• Article 2-5 revealed key elements to support the business objectives of your AI Customer Agent: crafting meaningful customer experiences, applying conversational design principles, intentionally shaping bot personality, and developing a strategic go-to-market approach.
• Article 6 showed you how to measure and track quantitative KPIs, building dashboards and reporting mechanisms to monitor performance across multiple dimensions.
• This article provides the decision framework: Translating measurement into strategic choices about which use cases to scale, enhance, de-prioritize, or stop.

The path forward – from measurement to action

The framework is clear, and the governance is in place. Now comes the most important step: execution.

During your strategic review forums, plot each use case on the Value Prioritization Matrix. Assess financial attractiveness. Make your first scale, enhance, de-prioritize, stop decisions. Communicate them clearly to stakeholders using the framework you have established. Then repeat the process as defined by your governance, to continuously refine your AI Customer Agent portfolio based on data, not emotion.

This is how you close the circle from investment to measurement to decision to action, and back again – and how you realize the true value of your AI Customer Agent.

Click here to explore our AI Customer Agents Series: a series on crafting modular AI Customer Agents that deliver tangible value. 

Sources:

1. The GenAI Divide: State of AI in Business 2025, MIT NANDA, 2025

2. State of AI in the Nordics, Deloitte, 2026