How AI can improve an EPL implementation before go-live

Use AI on the implementation work itself

An Enterprise Permitting and Licensing (EPL) implementation produces a substantial amount of working material: requirements, fee inventories, workflow specifications, data mappings, test scripts, and training instructions. AI could help draft and compare these deliverables, giving specialists more time to resolve the decisions they expose.

Tyler describes its Enterprise Permitting & Licensing Community Development offering as including embedded AI. Confirm the native features available in the proposed version and contract. The approaches in this guide use separately approved AI tools and reviewed project documents. Check any connector or automated configuration capability against the tools and permissions available to your implementation team.

Sources: Tyler Technologies — Enterprise Community Development

Turn approved rules into a configuration worksheet

Provide an approved tool with the relevant fee schedule, application instructions, and adopted workflow rules. Ask it to extract a structured worksheet: source reference, case type, trigger, calculation inputs, exemptions, effective date, responsible role, and unresolved questions.

Require each entry to cite the exact source passage. If a rule is absent or ambiguous, the output should flag a decision needed. A fee with an unclear effective date should reach the authorized agency owner before anyone configures it.

For example, the tool could compare the adopted business license renewal rules with the proposed configuration worksheet and identify a missing late-fee condition. A specialist verifies the finding against the ordinance, records the decision, and approves the specification.

Use approved formulas and deterministic calculations to establish expected monetary results. Generative AI can propose test cases and help explain differences; it should not be the authority for what a resident or business owes.

Build a traceable requirements and test inventory

With appropriate recording and data approvals, use workshop notes or transcripts to draft requirements and workflow steps. Have department owners verify the result. Ask the tool to compare the approved requirements with configuration decisions and test scripts.

The useful output is a set of gaps: a requirement with no test, a test with no expected result, an unresolved policy question, or a workflow step with no owner. AI can suggest those gaps, while the project team confirms them and maintains the authoritative inventory.

Keep stable identifiers for requirements, source rules, configuration items, and tests. That gives leadership a way to ask whether a fee or permit rule has been specified, configured, demonstrated, and accepted.

Help staff review legacy data and attachments

Use schema descriptions and approved samples to draft mapping suggestions between legacy fields and the target data model. Ask for alternative interpretations and missing information rather than one confident answer. A data specialist reviews the proposal before creating conversion rules.

AI could also suggest document categories or flag records that appear incomplete or duplicated. Test the suggestions against a set labeled by staff. Keep case identifiers and source links so reviewers can return to the original record.

Use controlled queries and conversion routines to perform counts, balance checks, relationship checks, and approved changes. Do not let an AI classification automatically merge people or businesses, discard an attachment, or rewrite an active case. Apply the agency's retention and access rules to every dataset used.

Expand test coverage around exceptions

Give the tool approved workflows and ask it to propose variations: an incomplete submission, a fee change during processing, a withdrawn application, a failed inspection, a reassigned reviewer, or a business license renewal with changed ownership. The business owner selects the relevant cases and checks their expected results.

AI could draft repeatable scripts for the staff application and Civic Access, or help prepare automation code for a test environment. Use supported interfaces and the agency's access rules; an AI-generated script needs the same review as other project code.

Separate generation from verification. If the same tool invents both the rule and the expected answer, a consistent-looking test can still be wrong. Finance verifies fee expectations, departments verify process outcomes, and technical staff verify the automation.

Draft training from accepted workflows

After staff accept a process, use its approved instructions and test results to draft role-based job aids. Ask for separate versions for counter staff, reviewers, inspectors, and finance where their tasks differ. Include the exception paths they will actually encounter.

An internal assistant could retrieve answers from approved project documents and link to the source. Test its behavior when a document is outdated, a question has no approved answer, or the user lacks access. It should direct unresolved questions to the responsible person rather than invent an agency rule.

Worked example: catching an unsupported late-fee rule

Illustrative review exercise, not a client result or an adopted fee schedule. Suppose a deliberately hypothetical source passage says: “A late fee applies after the renewal deadline.” A separately approved AI tool drafts a test that applies the fee on the deadline. A reviewer compares the draft with the passage and catches the difference between “after” and “on.” The passage does not settle the time zone, cutoff time, or whether a grace period applies; the authorized agency owner must decide those points from the adopted rule.

Keep the source passage alongside the draft and do not pass unapproved records to a tool. Test before the deadline, on the deadline, and after the deadline. Compute any monetary expectation independently from the agency’s actual approved schedule, not from the draft. Record accepted cases, rejected suggestions, review effort, and rework as pilot measures; no result is assumed here.

Start with one deliverable and measure the result

An illustrative pilot could cover one license type and its approved fee rules. Produce a reviewed worksheet, a requirements inventory, and a test pack using controlled copies of source material. Compare the effort and quality with the team's current method.

Record accepted entries, incorrect or unsupported suggestions, missed exceptions, staff review time, rework, tool cost, and whether the result was usable by the implementation team. An apparent time saving disappears if specialists must repeatedly reconstruct the output.

NIST's voluntary AI Risk Management Framework provides a basis for assigning ownership, assessing risks, and evaluating results. Apply it to the implementation task and approved data, then decide whether to expand, revise, or stop.

• Approve the tool, source material, access, and retention arrangements.
• Assign a business reviewer and a technical reviewer for the deliverables.
• Require source references and a decision log for unresolved rules.
• Expand only after accepted outputs and total staff effort have been measured.

Sources: NIST — AI Risk Management Framework