Organizations that report successful AI initiatives invest up to 4 times more (as a percentage of revenue) in foundational areas, such as data quality, governance, AI-ready people, and change management, compared with those that experience poor AI outcomes, according to Gartner.
These organizations recognize that an accurate model, useful prediction or compelling dashboard does not automatically produce a better business outcome. Too often, the insight never reaches the workflow, person or system that needs to act on it.
That disconnect is the last mile of data and AI makes a difference. The last mile is where an insight must become a timely, consistent operational decision. If the connection breaks down, organizations remain in the decision gap between knowing what to do and acting when it matters.

We sat down with Diana Maris, Senior Product Manager for SAS Intelligent Decisioning, who works with organizations seeking to translate data and AI into governed decisions and trusted production. In this Q&A, she explains why the last mile matters, how decision intelligence helps close the gap and why the rapidly advancing nature of autonomous AI makes it even more important.
Organizations have more data, models and AI capabilities than ever. Why is the last mile still a challenge?
Maris: Generating an insight and acting on it are two different disciplines. An AI and analytics team may build a strong model that predicts fraud, credit risk or customer churn. But the organization still has to determine what to do with that prediction, when to do it, which policies apply and whether a person needs to review the decision.
That is where many initiatives slow down. The model may be ready, but it is not connected to the operational process. Someone may still need to interpret a score manually, consult a spreadsheet, apply a policy and send the result to another team. By the time that happens, the moment to act may have passed.
The last mile is not simply a deployment issue. It is the work of connecting data and AI to business rules, policies, workflows and accountability so that insights can consistently influence real decisions. If that connection is missing, it’s hard to see business value.
Why isn’t AI alone enough to close that gap?
Maris: AI is a powerful technology, but it generally only provides one part of the answer. A model can estimate what is likely to happen. A language model can interpret information or generate a recommendation. Neither automatically accounts for every policy, regulatory requirement, business constraint or operational step surrounding the decision.
A probabilistic answer may be valuable, but an organization still needs to know whether the recommendation is allowed, appropriate and aligned with its objectives. That determination may require deterministic business rules, eligibility requirements, approval thresholds, historical context and human judgment.
The problem becomes more difficult when these elements are fragmented. Models may be managed by one team, rules by another and workflows within separate applications. That fragmentation can lead to different decisions being made about the same customer or situation. Closing the last mile requires those elements to work together rather than handing the final decision in isolation.
Can you walk us through a decision flow?
Maris: Decisioning is how that discipline becomes operational. It applies the models, rules and decision logic within the workflow where an approval, fraud response, credit evaluation, claims decision or customer interaction actually occurs.
A useful way to think about a decision flow is in four parts. First the inputs like the data, signals and context are available at the decision moment, making it connected to your organization. Followed by intelligence, which are the analytical models, AI, scores and business rules used to interpret those inputs. Then you have guardrails, which are the policies, compliance requirements, thresholds and human checkpoints. And finally, the outcomes. These are the recommendations or actions that can be executed, explained and audited.
A model predicts or recommends. A decision flow makes explicit how that intelligence should be used. That is what connects business intent to operational action.
Why is the last mile becoming a trust issue for organizations?
Maris: Organizations are not only asking whether AI can act quickly. They are asking whether they can trust the action. If decision logic is opaque, people may not know which data, rule, model or threshold produced an outcome. That creates uncertainty for employees, customers, risk teams and regulators. Even an accurate recommendation may be difficult to adopt if the organization cannot explain why it was made.
Trust requires decision logic to be visible and governed throughout its lifecycle. Organizations need to know which version of a rule or model was active, who approved a change, what data was used and how the decision performed after deployment. Explainability shows why an outcome was produced. Auditability makes it possible to reconstruct the decision. Governance establishes who owns the logic and how it may change.
These controls do not have to slow innovation. When they are designed into the decision process, they give teams a controlled path from testing to production. That can make it easier to update decision logic while preserving transparency, compliance and accountability.
Does that mean organizations should manage decisions as assets?
Maris: Yes. Many organizations carefully manage their data and models, but the decision itself often remains hidden across applications, spreadsheets, policies and institutional knowledge.
Treating decisions as strategic assets means making the complete decision logic explicit. That includes the data inputs, business rules, analytical models, AI components, workflows and human checkpoints that determine an outcome.
Once that logic is visible, it can be versioned, tested, reused and monitored. The organization can see how a decision supports a business objective, assess the effect of proposed changes and apply consistent logic across different channels or departments. It also becomes easier to identify what needs to improve when outcomes change.
This shifts the focus from managing isolated technical artifacts to governing the decision the organization is ultimately accountable for.
Explore the white paper, Decision intelligence: A leader’s guide to better decisions and trust with data and AI, to see how organizations can close the decision gap and turn analytics and AI into governed action.What role should people continue to play as more decisions are automated with AI?
Maris: Human judgment should be designed into the decision rather than added as an afterthought. The appropriate level of involvement depends on the impact, complexity and reversibility of the decision.
Routine, lower-risk decisions may be suitable for automation. Higher-risk or consequential cases may require a recommendation to be routed to a person with the context needed to make the final call. The important point is that the handoff should be deliberate. The person should see the relevant information, applicable policy and reasoning behind the recommendation.
Decision intelligence is not about removing accountability from people. It is about helping their judgment go further by providing consistent information and clearer reasoning. Humans still establish the intent, policies, thresholds and escalation paths. They also remain responsible for determining where automation is appropriate and where oversight is required.
How does agentic AI change the last-mile conversation?
Maris: We’ve actually been here before, in a smaller way. When predictive models first went into production, organizations had to figure out who was accountable for a model-driven recommendation, who owned the threshold and who could explain it to a regulator. Agentic AI doesn’t raise that question for the first time. It raises the stakes on it. A model-era recommendation still had a built-in pause. Meaning a person decided whether to act on it. An agent closes that pause. The action happens and only afterward can you ask who authorized it, on what data and under what policy. That’s why accountability can’t just follow the agent anymore. It has to be built into the decision layer the agent calls, before the action happens. Not reconstructed after the fact.
An agent should not have to recreate an organization’s credit policy, fraud thresholds or eligibility rules every time it performs a task. It should be able to call governed decision logic that already reflects the organization’s policies, risk tolerance and compliance requirements.
That separation is important. The agent can interact with users, gather context and orchestrate work. The governed decision layer determines which actions are permitted and records how the outcome was reached. This helps prevent different agents from applying conflicting interpretations of the same policy.
Agentic AI makes decision intelligence more urgent because the organization needs confidence that an agent’s actions remain explainable, auditable and aligned with business intent.
Where should an organization start to adopt decision intelligence?
Maris: I advise starting with one decision and not an enterprise-wide technology program. Choose a decision that is frequent enough to matter, has a recognizable business outcome and currently creates friction because of manual steps, inconsistent logic or delayed action.
Map the decision from beginning to end. Identify the required data, models, business rules, policies, workflow and human judgment. Establish who owns the outcome, who owns the risk and who has authority to approve changes. Then determine how success will be measured in operational and business terms.
That first use case should demonstrate that the decision can be executed consistently and governed throughout its lifecycle. Once the organization has a repeatable approach, it can reuse the components and operating model for additional decisions.
Closing the last mile does not require handing every decision to AI. It requires being intentional about how insights become actions and ensuring those actions remain aligned with the organization’s goals, policies and responsibilities.
Closing the decision gap
The value of data and AI is not realized when an insight is generated. It is realized when that insight helps produce a better outcome.
Decision intelligence gives organizations a practical way to connect models and recommendations to business rules, workflows, governance and human judgment. By managing decisions as transparent, reusable and governed assets, organizations can move faster with greater trust in AI.