Voice assignment tracking and company technology inventory.
IT case study 11 / Operations engineering · AI orchestration · Data governance
From scattered records
From scattered records
to coordinated operations.
I created the company's device inventory and IT assignment workbooks when these records had not previously been organized in a centralized system. After building the Excel and SharePoint foundation, I trained a team of specialist Grok bots to interpret conversational requests, propose and validate changes, and—within approved scope—update the live workbooks.
Direct assistant, coordinator, three operational specialists, and testing lab.
Suggestions only → reviewable copies → scoped live edits.
01 / The foundationStructure first.
Structure first.
Automation second.
The AI capabilities came after the operational data had a consistent place to live. I built and initially maintained the workbooks myself, created asset-tag conventions, and established how assignments could be found, checked, and updated. In an earlier documented inventory-automation trial, I then directed a detailed examination of the Excel workbook before permitting any edits.
WORKBOOK ATelephony operations
Voice for Pest assignments
Organizes phone users, extensions, departments, queue membership, and call-queue configuration. A central assignment sheet uses priority values and feeds individual queue views; a comparison field highlights differences between recorded and current queue-agent counts.
UsersDirectory of phone assignments
Call QueuesQueue-level configuration and reconciliation
AssignmentsAgent-to-queue membership and priority
Queue viewsSupporting detail by call queue
Based on structure inspected in the provided workbook, not an export of company phone assignments.
WORKBOOK BAsset administration
Office technology inventory
A company device register built around categories, device details, serial and asset identifiers, assignment, status, and employee associations. The workbook includes a summary dashboard, configurable lists, guidance, and a reusable employee profile template.
Device InventoryEquipment, tags, status, and assignment
DashboardCategory and status summaries
EmployeesEquipment ownership reference
Reusable templateConsistent staff profile layout
Based on structure inspected in the provided workbook. Names, asset tags, serial numbers, emails, and workbook files are not published.
Beyond IT: I made the asset-tracking workbook template available to other departments and helped adapt it for warehouse management of pest-control equipment. That broader adoption is described from my work, without publishing operational warehouse records or claiming an independently measured utilization outcome.
02 / Specialist-bot organizationA team of agents.
A team of agents.
One human authority.
The earlier inventory engineering work began with two specialists: IT Director coordinating plans, approvals, verification and reporting; and Data Manager analyzing records and performing specifically authorized test operations. I subsequently expanded that design into the broader specialist team shown below. The hierarchy is a conceptual reconstruction, not a live view of permissions, bot-to-bot messages, or production access.
Human IT AdministratorFinal decision-maker · Authorizes consequential changes
IT DirectorMaintains bot registry · Routes work · Reviews evidence · Reports status
03 / Moving from manual to conversational updatesTrust was tested.
Trust was tested.
Access was earned.
I initially updated the Excel workbooks manually. I then trained specialist bots to turn natural-language requests into structured workbook changes, increasing the level of autonomy only after iterative testing and validation.
Suggestion only
The bot interpreted a request and proposed what should change. No changes were made until I reviewed the recommendation.
Human-reviewed proposalsReviewable workbook copies
The bot prepared a new version for comparison. I validated the revised workbook and uploaded the accepted version myself.
Version comparison before productionScoped direct updates
After extensive trials reached an error-free result in my observed testing, I permitted the bots to update the relevant production workbooks within their authorized scope, with coordinator review and human oversight.
Controlled write access—not unrestricted authorityTwo different milestones: The earlier, rigorously documented inventory module trial was isolated from production and ended with four of six planned modules passing Excel Desktop verification. The later account of scoped live workbook updates describes a subsequent stage of the broader bot workflow. Neither account independently proves which publishing mechanism was ultimately enabled, or that every earlier module was released to production. The reported error-free trial experience is not a guarantee of future error-free operation.
04 / Workbook engineering & isolated testingPrecision matters.
Precision matters.
Especially in Excel.
The earlier technical project concentrated on the technology inventory workbook. Before attempting production changes, I directed the IT Director and Data Manager agents through read-only analysis, isolated workbook modifications, structural comparisons, and final human verification in Microsoft Excel Desktop.
A routine save was not safe enough.
A conventional Python Excel save operation could remove three specialized dropdown-validation rules from this interconnected workbook. That meant a cell edit could appear successful while silently damaging unrelated features.
Preserving formulas alone was insufficient: employee-specific profiles, dashboard calculations, hidden worksheet elements, formatting, and dropdown behavior also had to remain intact.
Change only what was authorized.
I worked with the bots to develop a targeted Office Open XML editing approach rather than rewriting the entire workbook. The agents first validated the source, located the intended record, checked expected values, and prepared the modified cells in an isolated test copy.
SHA-256 fingerprints, structural and formula comparisons, validation-rule checks, and audit logging were part of the trial verification process.
01Inspect read-onlyWorkbook dependencies and target record
02Validate inputOriginal values, scope and file fingerprint
03Edit OOXMLAuthorized cells in a test copy only
04Compare resultsFormulas, dropdowns, structure, unrelated records
05Review in ExcelHuman final check before declaring a pass
Reconstructed workflow from the user's earlier project report. Not a copy of actual source code, an agent execution log, or a live SharePoint publishing diagram.
05 / Inventory automation test resultsFour workflows.
Four workflows.
Verified in isolation.
Four of the six planned inventory automation modules passed the documented trial and the administrator's final Microsoft Excel Desktop verification. The report does not identify the other two modules as completed or establish later production deployment of these specific implementations.
Add equipment
Create a device record while checking that inventory summary calculations remained correct.
Assign equipment
Update asset ownership and confirm the item appeared in the intended employee profile.
Transfer equipment
Move a device between employee profiles without damaging formulas and record the transfer history.
Repair profile formulas
Restore missing formulas and verify the dependent calculations in Excel.
What passed review: Source-versus-output comparisons, preserved validation controls, formula and dashboard checks, hidden elements, unrelated record integrity, and human verification in Excel Desktop. These are reported results from isolated testing, not an independent external audit or a blanket claim that the workbook can never be corrupted.
06 / Secure integration designPublishing was the
Publishing was the
next trust boundary.
At the time of the earlier report, SharePoint publication for this particular targeted-OOXML workflow was still in architecture research and testing preparation.
Microsoft Graph, Power Automate and Office Scripts
I evaluated these approaches alongside a dedicated publish-only execution mechanism. The proposed design separated workbook preparation from the credentials and privileges needed to publish an approved result.
Verified identity, version and result
Design requirements included a narrowly scoped publishing identity, verified file fingerprints, concurrency protection, SharePoint version history and post-publication checks.
Chronology: This architecture remained at design/test-preparation stage in the earlier technical report. The broader bot system was subsequently allowed to update authorized SharePoint workbooks, as described elsewhere in this case study; that later deployment does not, by itself, establish that the specific Microsoft Graph or publish-only architecture above was implemented.
07 / Task routing & verificationA request is more than
A request is more than
a command.
The governance model gives each task a responsible specialist, a review stage, and a path back to the administrator. These are the operating rules the bots were instructed to follow; prompt rules alone are not represented as technically enforced access controls.
01RequestHuman provides a conversational task
02RouteIT Director selects responsible specialist
03PrepareData Manager, Onboarding, or IT Operations handles its assigned stage
04VerifyCompare result to the request and available evidence
05Approve / updateRequire human approval for consequential actions; use preauthorized scope for routine edits
06ReportTask status, actions, outcome, and exceptions
08 / Guardrails by designAutomation with
Automation with
clear responsibility.
Approval remains with the administrator
The IT Director cannot approve its own work or another bot's work on my behalf. Creating accounts, changing access, deleting records, or other consequential actions require my explicit approval.
No self-issued permissions
Bots must not grant each other credentials or new permissions. Routine updates to approved workbooks are distinguished from changes to account access or sensitive production controls. In the earlier trial, agents operated against isolated workbook copies, not the live SharePoint master.
One responsible editor per stage
Task ownership is assigned to one bot at a time, with instructions to prevent simultaneous edits to the same workbook and to investigate failures before retrying delegation.
Verify, record, and report
The coordinator maintains a bot registry and central task log, checks completion claims, captures errors or missing information, and sends consolidated updates. The earlier inventory trial also used SHA-256 fingerprints, test files, and documented results. My final approval remained separate from both agents.
Privacy & security practices: The earlier technical work included source-level data masking, sanitized reports, restricted sensitive-content handling, automated output scanning, and documented incident-response procedures. Implementation boundary: Bot prompts specify separation of duties, unique task IDs, approvals, single-editor coordination and stop-on-error behavior. Isolated inventory tests provide evidence of file-level verification checks, but prompt instructions alone do not establish enforced permissions, concurrent-edit locking, complete production approval-log coverage, or guaranteed autonomous correctness.
09 / What changedFrom personally maintained files
From personally maintained files
to repeatable IT processes.
The most meaningful result is not the number of bots—it is that operational information now has a structure, ownership, review process, and more than one way to be maintained.
Previously undocumented assets became trackable
Established an internal technology asset system with tags, device classifications, assignment records, employee references, and an overview dashboard.
Inventory methods extended beyond IT
Shared the workbook approach with other departments and helped implement a warehouse equipment-tracking adaptation.
Routine updates became conversational
Trained bots to process requests, prepare structured edits, and update authorized SharePoint workbooks after progressive testing.
Specialist roles and review rules were defined
Established a coordinator and specialist responsibilities, approval boundaries, task logs, verification expectations, and escalation rules. Earlier inventory automation skill instructions were maintained in a version-controlled series of structured Markdown revisions.
No time-savings percentage, audit-grade accuracy rate, uninterrupted-write guarantee, or quantified labor-cost reduction is inferred from the materials provided.
Related case studiesBetter records.
Better records.
Better support.
The multi-agent system complements a separately deployed Microsoft Forms, SharePoint and Power Automate service desk, including Teams and email-based AI intake.