Six Things AI Can Do in Your Process Right Now

A note on terminology: some of what gets called "AI" is really automation - rules-based scripts that could have been built without AI.

True AI involves pattern recognition, generation, or judgment under uncertainty. The highest-value applications often combine both: AI handles the judgment calls, and automation handles the repeatable execution.

1. Document Processing

The "killer app" for government. Summarizing reports, extracting data from documents, classifying incoming requests, translating between English and French, for example. If your team processes large volumes of program applications, grant submissions, or correspondence, this category pays for itself quickly. Example: a policy unit that used to spend 40 staff hours per quarter summarizing stakeholder submissions now does it in 6, using the other 34 hours to analyze important themes and draft high-quality recommendations.

2. Information Management

Searching across databases, retrieving files, matching records, cleaning messy data, verifying information against source. Example: a procurement office matches vendor submissions against historical data to flag inconsistencies - a task that used to happen after an award was already made, now happens before.

3. Communication and Service

Answering common inquiries, generating first drafts of routine responses, providing guidance to clients for simple transactions, triaging incoming cases - all to improve the Citizen Experience (CX). Example: a benefits unit uses an AI assistant to draft responses to common questions, with human officers reviewing and sending. Response time dropped from 5 days to same-day on roughly 60% of inquiries. Shouldn't they prevent many of these questions in the first place by making the client journey clearer and simpler? Yes, but until they have solved that larger systemic issue, drafting responses to these routine questions is a time-saving countermeasure.

4. Analysis and Decision Support

Trend analysis, anomaly detection, risk assessment, application scoring, case prioritization. Example: a health program office uses AI to analyze regional incident data and flag patterns that human reviewers had been missing because they analyzed the data region-by-region rather than nationally.

5. Regulatory and Compliance

Monitoring compliance, auditing records, detecting potential fraud, screening applications, validating eligibility. Example: a grant program uses AI to screen applications against eligibility criteria before they reach human reviewers. Reviewers focus only on eligible applications. Turnaround time goes down. Appeal rate drops.

6. Financial Operations

Calculating benefits, reconciling accounts, generating invoices, assessing taxes, allocating resources. Example: an internal services branch reconciles departmental chargebacks with AI in a few hours rather than the three days it used to take - freeing the team to focus on the important exceptions and disputes.

Beware: I'm not saying AI is able to run any of these unattended. In every example, there needs to be a human in the loop doing validation, judgment, and exception handling. The point is that AI can do much of the repetitive, rules-based, pattern-matching work that consumes a lot of a team's time - freeing people to do the work that actually needs a human involved.

Which of these six categories describes best the bottleneck in your process? Don't pick the category that sounds the most exciting or technically sophisticated - choose the one that is actually holding up your throughput. That is where you start. (If more than one truly describes a bottleneck in your work, start with the one that is having the most negative impact on your clients).

If you don't know where your bottleneck is, AI will not be of much help. It's a problem that requires Lean analysis to solve.

7. AI Orchestration (an emerging seventh category)

One AI takes a request, decides what subtasks are needed, hands them off to another AI or automation routine, and then verifies the assembled output. An HR practitioner from the federal government that I work with has built a pipeline that does exactly this: 8 AI calls per job description to extract elements, translate them into standardized language, and identify similarity across hundreds of job descriptions. The end product helps the organization see whether it has the right amount of people in the right roles to deliver its mandate. That's orchestration. That's where AI in 2026-2027 is heading.

Key takeaways:

• AI does six categories of work well in knowledge and admin environments.
• Document Processing, Information Management, and Communication & Service are usually the easiest starting points.
• Analysis & Decision Support, Regulatory & Compliance, and Financial Operations often deliver the biggest savings.
• Every example still requires human judgment, validation, and exception handling.
• Start with the category that describes your bottleneck work - not the one that sounds most impressive.