Tutorial ID | T-24 |
Section | AI & Analysis |
Title | Creating and Running Automated Workflows |
Subtitle | Build the right workflow for each case type and run it only on the documents that matter. |
Before You Begin
You have a case with documents uploaded and reviewed, and you know its case type.
You have access to Workflows, Custom Run, and AI Prompt Hub for this case.
What You Will Accomplish
By the end of this tutorial, you will know how to choose or build the right workflow for a case type, run it only on the documents that matter, and avoid rerunning it without a reason to expect a different result. You will also know how to prepare a workflow for sensitive topics before it runs.
Why Does this Tutorial Matter
Different case types need different documents, processing steps, and report structures. A workflow built for one case type, run on exactly the records that matter, produces a focused report that meets what the attorney or reviewer actually needs — without spending budget on documents or reruns that don't add anything new.
Applied without that care, a one-size-fits-all workflow skips steps that matter for the case at hand, processing every document equally lets irrelevant records dilute the output, and rerunning a workflow unchanged just returns the same report at additional cost. Left unprepared, a workflow can also stop partway through when it reaches a sensitive topic, leaving the report incomplete at the worst possible time.
What to Do if Something Goes Wrong
The list below covers the most common problems when creating and running workflows. Each entry follows the same pattern: what you'll notice, why it likely happened, a real example, and how to fix it.
Problem: A workflow produces a report that's incomplete or misaligned with what the case actually needs.
Likely cause: The same workflow was applied to the case regardless of its type, even though personal injury, workers' compensation, and other case types each need different documents, steps, and report structures.
Example: A workflow built for a personal injury case is run on a workers' compensation case and skips a wage-loss section the report actually needed.
Fix:
Open the case and review the Case Summary and Expected Outcome fields.
Open the file batches and review the documents present, noting whether they're hospital records, diagnostic reports, or legal documents.
Based on the case details and documents, identify the case type — for example personal injury, workers' compensation, or car accident.
Go to Workflows and review the available workflows, reading each name and description carefully.
If an existing workflow matches the case type and needs no adjustments, select it and proceed.
If no existing workflow fits, click Create Workflow and give it a name that reflects the case type.
Configure the workflow to include only the steps, document types, and report structure relevant to that case type.
Once the workflow runs, confirm the structure and content match what's expected for that case type.
Problem: A workflow takes longer than expected to run, or its output includes noise from documents that don't matter to the case.
Likely cause: The workflow was run across every document in the case instead of only the ones relevant to the case type and expected outcome.
Example: A hundred-document case includes thirty administrative and duplicate files. Running the workflow across all of them adds processing time and lets irrelevant details creep into the report.
Fix:
Before running the workflow, review the file batches to understand what types of records are present and how they're organized.
Evaluate each batch against the case's purpose — does it contribute directly to the expected outcome, or is it administrative, duplicative, or unrelated?
Go to Workflows and select the workflow that matches the case type.
Open Custom Run and select only the batches that are directly relevant to the case type and legal purpose.
Run the workflow with that selection.
Review the output and confirm it reflects only the selected records; if irrelevant information appears, refine the selection and rerun.
Problem: Two or more workflow outputs in the same case are identical or nearly identical, and the team has to validate all of them.
Likely cause: The same workflow was run again on the same case without changing the prompt or document selection, so the AI processed the same records with the same instructions.
Example: A workflow is rerun on a case with no changes, and the second report matches the first almost word for word, doubling the review work for no new information.
Fix:
Before running a workflow, go to Workflows and check whether it has already been run on this case.
Review the status, date, and time of any previous run of the same workflow.
If a previous run exists, open its output and review the generated report before doing anything else.
If the existing report is sufficient, don't rerun the workflow — use the existing output and proceed with review and editing.
If the report needs improvement, identify specifically what needs to change, such as the prompt, added documents, or the document selection.
Make the necessary changes in AI Prompt Hub or Custom Run before rerunning the workflow.
If the new output isn't an improvement, review the prompt and document selection more carefully rather than rerunning again.
Problem: A generated report looks complete but is missing entire sections or case-specific content once you look closely.
Likely cause: No dedicated workflow exists for this case type, so the report was generated from a standalone prompt that couldn't cover the full scope of the case on its own.
Example: A complex multi-provider case is run through a single prompt instead of a workflow, and the resulting report never covers a required causation section.
Fix:
Open the case and review the Case Summary and Expected Outcome fields to confirm the case type and the scope required.
Go to Workflows and check whether a workflow exists for this case type.
If one exists, review its configuration and confirm the steps, document types, and report structure still match the case.
If none exists, click Create Workflow before proceeding.
Configure the new workflow to include all required document types, processing steps, and report sections for this case type.
Run the workflow using Custom Run so only the relevant documents are processed.
Review the report section by section once the output is ready.
If systematic gaps remain, return to the workflow configuration, identify which steps produced insufficient output, and adjust before rerunning.
Problem: A workflow produces incomplete output, returns an error, or stops entirely partway through.
Likely cause: The case involves a topic — such as content related to children or a rare diagnosis — that triggers a built-in limitation in AI processing, and the prompt wasn't prepared for it in advance.
Example: A workflow stops partway through a section describing a pediatric injury, leaving the rest of the report unfinished.
Fix:
Before running the workflow, review the Case Summary to check whether the case involves a sensitive topic, such as a pediatric injury, a congenital condition, or a rare genetic disorder.
Open AI Prompt Hub and open the prompt associated with the workflow.
Reframe sensitive phrases using neutral, clinical terminology drawn from the source documents — for example, “a minor sustained a traumatic injury” instead of “a child was hurt.”
Run the workflow and monitor the output as it generates.
If an error occurs or the workflow stops, note which topic triggered the interruption.
For the affected section, try a more narrowly scoped prompt focused on specific facts rather than the sensitive topic broadly.
Once the report is complete, review it as a whole to confirm no sections were left incomplete.
Tip A workflow is a one-time investment — once it's configured correctly for a case type, it protects the quality of every future report of that type, not just the one in front of you. |
Summary
Every workflow now matches its case type, runs only on the documents that matter, and only reruns when there's a real reason to expect a different result — with sensitive topics reframed in the prompt before the workflow ever starts.