People Before Platforms: Make AI Serve the Mission

Stop asking "What's our AI strategy?" Start with "How does AI help our mission?"

Since the administration released an AI Action Plan in July 2025, many agencies have rushed to craft standalone AI roadmaps. Results have been uneven. One IBM analysis found enterprise AI initiatives returned about 5.9% ROI despite roughly 10% budget allocation - a poor trade if you measure outputs, not outcomes.

The better question is simple: How does AI help us hit existing objectives faster, cheaper, and with higher quality? Treat AI as an accelerant to current priorities, not a separate agenda. That shift turns abstract initiatives into concrete wins your teams can see and support.

Start with people, not platforms

Past software deployments locked in the process once the tool went live. Today, outcomes vary by how people prompt, review, and iterate. Two employees can use the same model and get different results. That means behavior change is the work.

• Translate AI to mission value: For each priority, state the before/after: "Case reviews in 10 days vs. 30," "Grant errors down 40%," "Citizen response time under 2 minutes."
• Set a clear job-impact narrative: Specify which tasks shrink, which skills matter more, and what growth paths exist. Reduce anxiety with clarity and commitments.
• Create safe-to-try spaces: Launch sandboxes and office hours where teams can test use cases without fear of blame. Reward useful experiments and share what's learned.
• Codify good habits: Publish prompt libraries, review checklists, and data-use guardrails so quality scales with adoption.

Adopt a multi-channel approach

Top-down direction and bottom-up discovery must work together. Leaders tie AI to strategy, budgets, and risk posture. Frontline teams surface bottlenecks, repetitive work, and quick wins you can't see from the org chart.

• Build a small portfolio of use cases: Mix quick saves (weeks) and strategic bets (quarters). Kill weak ideas early; double down on traction.
• Link to goals, not tools: Add AI outcomes to OKRs and performance reviews. If it's not in the plan, it won't scale.
• Run discovery sprints: 2-3 week cycles where frontline staff test workflows, document impact, and propose next steps.
• Create a champions network: Identify one point person per division to coach peers, collect feedback, and spread standards.
• Set minimal viable policies: Start with pragmatic guardrails on data, privacy, and review. Expand as you learn.

Practice adaptive leadership

AI progress comes from testing, learning, and adjusting in real time. Traditional governance slows that down. You'll need fewer approval layers, faster cycles, and visible transparency.

• Slim the path to pilot: Pre-approve low-risk trials under clear thresholds for data, spend, and scope.
• Time-box everything: 30-60 day pilots with crisp success criteria - cycle time, error rate, satisfaction, cost-to-serve.
• Broadcast what's working: Share demo videos, metrics, and playbooks internally so teams copy the wins.
• Adopt proven guardrails: Use frameworks like the NIST AI Risk Management Framework as your baseline.

Make ROI inevitable by measuring outcomes, not activity

Funding proofs-of-concept without business metrics creates busywork. Tie every initiative to measurable operational improvements. Start with baselines, then report deltas weekly.

• Operational: Cycle time, throughput, error rate, backlog cleared, first-contact resolution.
• Service quality: Citizen satisfaction, accessibility compliance, response time.
• Financial: Cost per case/claim/inspection, rework costs avoided, contractor spend reduced.
• Workforce: Hours redirected from admin tasks to mission-critical work, training completion, tool adoption.

Partner early with finance and audit so savings and risk controls are credible. Publish dashboards and hold monthly reviews that decide: scale, fix, or stop.

A practical 90-day playbook

• Weeks 1-2: Pick 3-5 mission priorities. Map pain points with frontline interviews. Define target outcomes and baseline metrics.
• Weeks 3-4: Stand up a secure sandbox, minimal guardrails, and short training. Draft standard prompts and review checklists. If your teams need structured upskilling by role, explore role-based AI courses.
• Weeks 5-8: Launch three pilots. Instrument metrics from day one. Hold weekly show-and-tell sessions and capture playbooks.
• Weeks 9-12: Scale the winner, fix the "almost," and sunset the laggard. Update policies, templates, and funding based on actual results.

Common risks and how to avoid them

• Tool-first buying: Require a business case and owner for every license. No orphan software.
• Shadow AI: Provide approved tools and clear do/don't rules to reduce risky workarounds.
• Model drift and quality: Set human-in-the-loop reviews for sensitive decisions. Log prompts, sources, and outputs.
• Overhype: Share misses as openly as wins. Credibility builds adoption.

The bottom line

AI is a means to mission outcomes - faster decisions, fewer errors, better service, lower cost. Start with people and workflow design, not platforms. Tie every initiative to an existing objective, measure relentlessly, and keep your governance light enough to let learning happen.

If your workforce understands the why, feels safe to experiment, and sees proof in the numbers, adoption follows - and so does real value.