Automation vs AI: Rules-Based Bots in U.S. Agencies

How the U.S. federal RPA playbook separates rules-based automation from machine learning, with figures from a 2020 report covering 23 programs.

Guide: AI Tools and IndustryLast verified: October 20264 min read

Automation carries out a fixed set of steps that a person defined in advance, while AI, in the machine-learning sense, builds its own decision rules from examples. The U.S. federal government's own guidance on robotic process automation (RPA) draws the line this way: RPA is low-code commercial software that automates repetitive, rules-based tasks by emulating human actions, and machine learning is a separate layer that agencies may add later.

Key takeaways

  • The Federal RPA Community of Practice's playbook, version 1.1 of January 17, 2020, defines RPA as low to no-code commercial software for repetitive, rules-based tasks.
  • The playbook advises starting with mature, standardized processes that have few decision points, and it treats machine learning as a later step.
  • A November 1, 2020 community report covering 23 programs counted 460 automations in fiscal year 2020, up from 219 a year earlier.
  • The same report lists natural language processing, machine learning, image analysis, cognitive agents, and chatbots as the advanced end of its automation scale.
  • The hours-saved figures are self-reported estimates of annualized capacity, not audited savings.

How the federal playbook defines the line

The RPA Program Playbook was published by the Federal RPA Community of Practice, a group of agency staff supported by the General Services Administration. It describes RPA as a low to no-code commercial off-the-shelf technology for repetitive, rules-based tasks. Like an Excel macro, RPA can record actions across a computer, reach systems, and perform delineated tasks for human users. The document says all RPA products emulate human actions and lists typical uses: data entry, data reconciliation, spreadsheet work, systems integration, automated reporting, and customer communications. It also states that it should not be read as official agency policy.

That description contains no learning step. If the form changes or an input falls outside the rules, a bot of this kind stops or follows the wrong rule. A machine-learning model works differently: it is trained on examples, so its behavior comes from data and not from a script. The playbook acknowledges the second category by noting that programs often combine RPA with process mining, business process management, and machine learning as demands grow.

Why the playbook starts with rules

For a pilot, the playbook recommends processes that are manual, repetitive, mature, and impactful, with few decision points, few connected systems, and preferably no personally identifiable information. Its scoring criteria include repetitiveness, frequency, exception handling, stability of requirements, structure of the data, and the degree of rules-based standardization. In effect the selection test asks whether a process can already be written down as rules.

The document also answers the claim that RPA is obsolete because AI is coming. It argues that RPA is a stepping stone, since it builds the technology structures, skills, and standardized business processes that later AI adoption needs, and that moving from manual work straight to machine learning is often a leap too far. That is the community's position and not a measured result.

What the 2020 federal report counted

The Federal RPA Community of Practice published The State of Federal RPA on November 1, 2020. It assessed 23 programs at two points, the end of fiscal year 2019 and August of fiscal year 2020.

MeasureEnd of FY 2019FY 2020
Automations deployed219460
Annualized hours of capacity created285,651848,336
Average annualized hours per automation1,3351,708

The report states the gains as 110 percent for automations and 195 percent for hours. Hours here are the programs' own estimates of capacity, and a freed hour is not the same as a dollar saved or a job removed. The report's maturity scale also separates basic uses of "native sensory tools" from an advanced level that includes natural language processing, machine learning, image analysis, cognitive agents, or chatbots. Rules-based automation and AI therefore appear in the same federal documents as points on one scale, not as rivals.

A quick way to sort a task (illustrative example)

The following sorting is illustrative and reflects the playbook's criteria, not a ruling by any agency.

TaskBetter fitReason
Copy figures from a spreadsheet into a form every weekRules-based automationFixed steps, structured data, no judgment
Flag invoices that look unusual compared with past onesMachine learningNo fixed rule; the pattern comes from examples
Route emails by free-form textMachine learning, with rules as backupWording varies too much for a script
Rename and file reports by dateRules-based automationA script can do it with no data to learn from

FAQ

Is robotic process automation a form of AI?

The federal playbook treats RPA as rules-based software and describes machine learning as a complementary technology that can be added. Some products combine the two, so the label depends on whether a learned model is involved.

Did federal RPA save jobs or money?

The 2020 report measures annualized hours of capacity that programs estimate they created. It does not report audited dollar savings or job changes.

Why start with rules-based automation before AI?

The playbook argues that standardized processes and technical foundations make later AI adoption easier. It offers this as guidance from practitioners, not as the result of a controlled comparison.

Are the 2020 figures current?

No. They describe fiscal years 2019 and 2020. The community has since been renamed the Federal Automation Community of Practice, and its newer reports are available only to members.

Last verified: October 2026

Sources

  • RPA Program Playbook, Version 1.1, Federal RPA Community of Practice, January 17, 2020
  • The State of Federal RPA: An Analysis of Governmentwide RPA Impact, Deployment, and Best Practices, Version 1.0, Federal RPA Community of Practice, November 1, 2020
  • Federal Automation Community of Practice, U.S. General Services Administration, 2026

New to artificial intelligence?

Start with the basics, then follow the guides to tools, rules, and effects on society.

Start with AI Basics