Helpful tools, but people still move the data
Open HaloPSA to review a ticket, search the Wiki, then pass the information to AI for analysis. Afterwards, remove lengthy text, adjust the format and copy it back into the system. Familiar issues require another search through past tickets. Purple and Criss of Team 5 saw that although AI was already part of the work, moving data between tools still took effort.
Describe the work before connecting features
The team first explained the actual sequence to AI: mark the ticket as in progress, keep records, check its category, obtain final confirmation from the customer or sales team, then close the ticket and decide whether to send a notification. After obtaining the necessary API permissions, they implemented the steps one by one.
People supplied requirements, demonstrations, screenshots and error messages, while Codex and Claude Code helped modify and debug the code. Each iteration returned to real tickets to check whether the features matched daily work. This process knowledge and test feedback were essential development inputs.
Submitting an action is only the start; confirm it succeeded
Status changes, final confirmation, closure and closure emails all failed repeatedly during development. The team worked through permissions, action order and system wait times. They also found that without a clear response after clicking a button, users could not tell whether to wait or retry.
Through demonstrations and repeated revisions, Workbench added waiting indicators, processing progress, result checks and human confirmation. Features progressed from submitting requests to showing users what happened. People still review important content and actions.
Past tickets become useful references for today's work
Workbench can now find similar or related tickets for the same customer so AI can refer to past resolutions. Based on actual usage estimates, finding a useful historical ticket takes about 3 minutes instead of 10. This estimate covers the search step, not total ticket handling time or a formal performance measurement. Ticket records are first organized into completed work, current status, key findings, recommendations and items to confirm, before people edit, verify and write them back.
These features share a workflow with status changes, final confirmation and closure notifications, reducing repeated searching and preparation. Regular customer email exchanges are still handled in HaloPSA.
Before and after: how the work changed
Compare the same work steps to see the roles of tools and people.
| Work step | Before | After |
|---|---|---|
| Finding historical tickets | Manually search past tickets and prepare information for AI. Finding one useful reference took about 10 minutes. | Workbench finds similar or related tickets for the same customer and provides them to AI for analysis, taking about 3 minutes. Both figures are usage estimates. |
| Analyzing tickets | Search, copy and switch between HaloPSA, the Wiki and different AI tools. | Ticket information, historical references and AI analysis are brought together in the Workbench workflow. |
| Preparing records | AI responses varied in format, requiring manual trimming and adjustments before copying them back into the ticket system. | Records are organized into completed work, current status, key findings, recommendations and items to confirm, then checked by people before being written back. |
| Status changes and closure | Return to HaloPSA to change status, perform final confirmation, close the ticket and send a closure notification. | These actions are integrated into Workbench, with waiting indicators, result checks and human confirmation. Regular customer email exchanges remain in HaloPSA. |
Applying this experience to your own work
The estimated search time covers only that step; overall handling time and long-term quality still need measurement. Start by identifying data you repeatedly move and actions you often perform. Clarify their sequence, waiting periods and confirmation points, then connect and verify them one by one.