How to Use AI in IT Asset Management

Introduction

AI tools can now help get tasks like these done quickly:

• Resolving help desk tickets
• Preparing Deployments
• Cleaning and recognizing data
• Software inventories
• Hardening firewalls, desktops and servers
• Preparing for audits
• Answering questions and writing reports based on your data

This article is intended to be a practical guide for IT asset managers, service desk teams, and IT operations staff who want to use AI to improve productivity and solve problems that previously might have been difficult or impossible to solve quickly. The tools referenced are general-purpose AI assistants like Claude and ChatGPT. A vendor-specific AI module or an enterprise platform upgrade isn't needed.

Now you can ask an AI questions like this:

    Write a PowerShell script that lists every 
    machine on the network with less than 10% disk space
            

In seconds it generates a working script.

These are some of the concepts covered in this article:

• Generate PowerShell scripts — scan subnets, audit software, query remote machines
• Clean messy asset data — normalise vendors, deduplicate records, classify assets
• Detect unauthorised software — compare installed software against an approved list
• Diagnose help desk tickets — cross-reference tickets against asset data and suggest fixes
• Generate SQL queries from plain English — ask questions, get reports, no SQL knowledge needed
• Identify SaaS sprawl — find duplicate tools, unused licences, and shadow IT
• Prepare audit evidence — reconcile entitlements against deployments and flag gaps
• Detect security risks — find missing patches, outdated firmware, expiring certificates
• Automate documentation — generate docs from emails, processes, and application knowledge
• Monitor asset health — track disk space, event logs, reboot history, and patch status

Keep Personal Data Safe

If you give data to AI, you can never undo that action. Most AI providers handle data ethically, but the safest approach is simple: don't share sensitive data at all. Companies often develop policies covering GDPR, HIPAA, SOC2, etc, which forbid the disclosure of personal information, and this may extend to other types of information, so check your local policies before proceeding, and even in the absence of such policies, it would still be bad practice to share sensitive data with an AI.

Start With Data

AI can use or even generate a CSV export from your discovery tool, your CMDB, other IT apps, or a spreadsheet. It can also use powershell to create a CSV list of assets or installed software titles by running a simple discovery script. An example of this has been included in this article.

If permissions allow, powershell can get almost any data that a discovery tool can get, and with an AI tool to write it for you, all you need is a basic knowledge of how to run scripts. So use AI to get your data, and then use AI to analyse it.

Some example prompts:

                Write a powershell script to show which machines have 
                unapproved software installed. 
                Use this PC as a template for allowed software.
            

If you want to find PCs which are low on disk space, you could ask:

                Write a powershell script to show which machines have 
                less than 10% free disk space on the windows drive.
            

AI will write the script for you, and you can run it immediately. If it does not work first time, paste the error message back into the AI engine and ask it to fix the script. If you have agent tools like Codex or Claude Code — you can ask it to test the script itself and keep fixing it until it works.

Cleaning Up Messy Data

According to a recent survey, complete visibility across the technology stack has dropped to just 43%. SaaS visibility sits around 50%. Bring Your Own License visibility is around 27%. The data exists, but it is scattered across tools, inconsistent, and full of duplicates.

AI can help with this. Here are tasks you could give to an AI assistant:

• Normalize vendor names between data sources ("Microsoft Corp", "Microsoft Corporation", "MSFT" become one entry)
• Deduplicate asset records by serial number, MAC address, or hostname
• Identify ghost assets: devices that appear in your CMDB or Excel Spreadsheet but are not responding to ping
• Find mismatches between your discovery data and your purchase records
• Classify uncategorised assets from their description fields
• Build a Sql Server Express repository of the data you are collecting with WMI scripts, then build queries on it

Software Recognition

Discovery tools find installed software, but the raw data is often cryptic. A discovery scan might report "Adobe Acrobat DC 24.002.20933" on one machine and "AcroPro.msi" on another. Both are the same product. AI can take a list of discovered software titles and normalize them into a clean, recognisable catalog.

Try pasting a list of Add/Remove Programs entries into Claude (Claude can pull these from WMI using Powershell) and asking it to group them by product family. The results are surprisingly accurate, and where they are not, you can correct them interactively.

License Reconciliation

Once you know what is installed, you need to reconcile it against what you are entitled to. AI can help here by comparing your entitlement data against your deployment data and highlighting gaps. Feed it your license spreadsheet and your discovery export, and ask: "where are we potentially non-compliant?"

It will not replace a formal SAM tool for ongoing reconciliation, allocations, prioritisation, etc, but it is excellent for ad-hoc analysis, audit preparation, and catching problems before an audit catches them for you.

Understand Your Data in Plain English

Network and Computer Data

One of the most useful things AI can do is translate plain English questions into database queries. This matters because most ITAM data lives in databases, and most ITAM professionals are not SQL experts.

Instead of learning query syntax, you can ask:

• "Show me all laptops not seen on the network in the last 90 days"
• "Which departments have the most software installations per user?"
• "List all assets still under warranty that are assigned to the London office"
• "What software is installed on more than 50 machines but does not appear in our license records?"

Some of these might need the underlying data, such as a warranty spreadsheet, or which departments are in which IP range, alongside the network device data. However, AI can easily merge data from multiple sources so these queries are definitely possible.

If you give the AI your database schema (or even just your column names), it can generate accurate SQL that you can run directly. This means that custom reports can now be created by anyone, without any knowledge of SQL.

Some ITAM tools are starting to build in this capability. xAssets, for example, provides an agent-based command-line interface that lets AI assistants query the asset database directly, so the AI can answer questions about your live data rather than just exported snapshots.

Database Data

If you have large databases of data, in Sql Server, or MySql, MariaDB, etc, AI can write a training file for itself based on the data model of this database. AIs with agents can do this automatically given a connection string (which the AI can also help you generate), and then that training file is used to construct queries and answer questions about it without needing to export it or share it with the AI directly. One thing that's really great about this approach — with agents, your data stays in your environment, and the AI is just building queries for you to run locally. No data needs to be sent to the AI engine in the cloud.

Help Desk Applications

Senior managers think about compliance whereas people on the ground are thinking about Joe's disk space, Anne's Microsoft Office not working, Server B overheating.

IT support in practice is a stream of problems that are individually small but collectively large. Anne has broken her Office installation, Joe has installed software that is not on the approved list, Jack in accounts is locked out, the Salesforce single-sign-on fails regularly for people working from home, and Fred's laptop is running out of disk space, a printer has gone offline, a monitor is not being detected.

AI can help with these in two ways:

1. Diagnosis

When a help desk ticket comes in, AI can read the ticket text and cross-reference it against the asset database. It can pull up the user's assigned devices, check their hardware specs, see what software is installed, review recent changes, and suggest a diagnosis before a technician even looks at it.

ITAM tools that integrate with service desks like JIRA can already auto-detect which assets a ticket relates to based on the reporter's name, computer names, and IP addresses mentioned in the ticket text. This is already a native capability within the xAssets system.

2. Agentic Fixes

Beyond diagnosis, AI can take action. An AI agent with appropriate permissions could:

• Generate and run a PowerShell script to clear temp files, downloads, and free up disk space
• Check group policy and Active Directory to diagnose an account lockout
• Query WMI to pull the current state of a remote machine
• Identify why an SSO connection is failing by checking certificate expiry and federation metadata
• Draft a remediation script that a technician can review and approve before execution

AI agents should diagnose freely but act only with human approval, at least until your organization has confidence in the guardrails. Start with read-only access and only allow write access for low-risk operations.

PowerShell and AI

PowerShell allows automation of almost all Windows IT operations. Many ITAM teams use it for gathering hardware data, checking software installations, querying Active Directory, scheduled processing, and automating repetitive tasks.

AI is usually accurate when writing PowerShell scripts. You describe what you want in plain English, and the AI generates a working script.

Example: Discovering Hardware Details

Ask an AI assistant to :

    Write a PowerShell script that queries WMI to get the make, model, serial number,
    CPU, RAM and disk space for a list of computer names
                

and you will get a script you can run immediately.

Here is a real-world example. This script discovers Dell Docking Stations connected to a laptop using the Dell Command Monitor WMI namespace:

    $computerName = "JOESPC"
    $namespace = "root/dcim/sysman"
    $query = "SELECT CreationClassName, ElementName, Manufacturer,
              Model, Name, SerialNumber, Tag, Version
              FROM DCIM_chassis"

    Get-WmiObject -Namespace $namespace `
        -Query $query `
        -ComputerName $computerName

            

That script was written by a human. But the point is that AI could have written it just as easily, and could adapt it for Lenovo, HP, or any other manufacturer with similar WMI providers.

Example: Finding Unauthorised Software

Ask your AI to "write a PowerShell script that scans a list of computers and reports any installed software not on an approved list" and it will generate something usable. You provide the approved list, the AI writes the logic, and you have a compliance scanner built in minutes rather than days.

Example: Scanning a Subnet

We asked an AI assistant to "write a PowerShell script to ping every address in a subnet, query WMI for hardware details on each machine that responds, and export the results to CSV." It produced this:

    param([string]$Range = "192.168.1.1-254")

    $parts = $Range -split "\."
    $base = ($parts[0..2]) -join "."
    $bounds = $parts[3] -split "-"
    $start = [int]$bounds[0]
    $end = if ($bounds.Length -gt 1) { [int]$bounds[1] }
           else { $start }

    $results = @()
    $start..$end | ForEach-Object {
        $ip = "$base.$_"
        if (Test-Connection $ip -Count 1 -Quiet -ErrorAction SilentlyContinue) {
            try {
                $cs   = Get-WmiObject Win32_ComputerSystem -ComputerName $ip
                $bios = Get-WmiObject Win32_BIOS -ComputerName $ip
                $os   = Get-WmiObject Win32_OperatingSystem -ComputerName $ip
                $disk = Get-WmiObject Win32_LogicalDisk -ComputerName $ip `
                            -Filter "DriveType=3" | Select-Object -First 1
                $results += [PSCustomObject]@{
                    IP           = $ip
                    ComputerName = $cs.Name
                    Manufacturer = $cs.Manufacturer
                    Model        = $cs.Model
                    SerialNumber = $bios.SerialNumber
                    RAM_GB       = [math]::Round(
                                     $cs.TotalPhysicalMemory / 1GB, 1)
                    OS           = $os.Caption
                    DiskFree_GB  = if ($disk) {
                                     [math]::Round($disk.FreeSpace / 1GB, 1)
                                  } else { "" }
                }
            } catch { }
        }
    }
    $results | Export-Csv "subnet-scan.csv" -NoTypeInformation

            

We ran this against a network segment. Three minutes later it outputted this data in CSV format:

This data was in a CSV file so we can open it with, or paste into, Excel. The script is about 15 lines of code and took a few minutes to write. Any ITAM analyst without scripting experience could produce this by describing what they need to an AI assistant.

Centralizing Your Data First

AI works better when it has a single normalised data source. If your asset data is spread across Intune, SCCM, a spreadsheet, a procurement system, and three different discovery tools, asking AI to make sense of it means combining exports.

A centralized ITAM platform like xAssets ITAM pulls data from multiple sources into one place giving AI a single, comprehensive dataset to work with. Instead of stitching together CSVs, you point the AI at one database, or one API, and ask your questions. xAssets provides an AI plugin so your xAssets instance can be queried directly from the chat interface, without needing to export any data.

A survey from 2025 found that around 65% of IT managers use six or more ITAM tools. That is an operational headache and also a barrier to using AI effectively, because AI would need access to every data source.

xAssets is built around this principle. It centralizes data by integration with Intune, SCCM, Meraki, JAMF, Azure, AWS, and other sources to create a single asset inventory. Once data is centralized, AI tools can query it directly through APIs, Sql queries, or command-line interfaces. This means the AI can answer your questions by translating your prompts into data queries with no knowledge of Sql or scripting required.

Preparation for Software Audits

License compliance is an important issue for IT leaders. According to industry data, 45% of organizations spent over $1M on software audits in the past three years. 23% spent over $5M.

AI can help with audit preparation in several ways:

• Match software entitlements against deployment data and flag gaps
• Find software installed on machines where it should not be
• Convert complex license agreements to plain language
• Generate evidence packs: lists of installations matched to POs and license keys
• Estimate financial exposure for identified compliance gaps
• If you know some machines are non-compliant, AI can help find others using pattern matching

Feed your AI assistant your Microsoft Enterprise Agreement alongside your deployment data and ask "are we compliant?" You will not get a definitive legal answer, but you will get a picture of where to focus your attention.

SaaS - A Growing Problem

Large enterprises run over 2,000 applications, and 61% of them are not formally approved by IT. SaaS waste is increasing year-on-year for 35% of organizations. This problem did not exist ten years ago, and most ITAM processes have not caught up.

AI can help identify and rationalise SaaS sprawl:

• Use SSO logs and browser data to find SaaS tools
• Expenses and credit card data to find SaaS subscriptions
• Identify duplicate functionality across tools
• Calculate per-user cost and usage for each SaaS tool to identify candidates for consolidation

Around half of Microsoft 365 users do not utilise all their license tier. Analysis of usage data can identify which users could be downgraded from E5 to E3, saving money with minimum impact.

More ideas of how you could use AI in your environment

The best place to start isn't with "what can I ask it", instead describe what problems you need help solving, and ask AI which ones it can help with without much time investment to get the most return.

Going Deeper with Agentic AI

So far this article has only touched the tip of the iceberg. With agent-based tools like Codex and Claude Code, AI more than doubles in its ability to do real work. You can set AI running an overnight job to perform network tasks, documentation, upgrades, audits, data analysis and reporting, discovery, data cleansing, and looking for issues in databases. You can allow it to log into your help desk and let it find incidents it thinks it can solve, and when you get in the next morning, the solution is there waiting for you to apply.

Some agents stop to prompt for permissions, and the one way to prevent that is to develop an MCP server as described below.

If you give AI a big task, it can get confused, context windows fill up, and it can start going wrong. But this is not necessary:

• Keep AI tasks small
• When a task is large, instruct the agent to break up the task into jobs and perform each job as a unit of work
• Skills files help AI know your environment — it's better to have several smaller skills files rather than one big one
• You can tell the AI which file to use for a particular type of task

MCP (Model Context Protocol) servers act as a bridge between AI assistants and your internal systems.

You can write your own MCP server, and AI can help with this. MCP servers allow AI engines to access data from your network that agents like Claude Code and Codex do not understand. For example:

• Complex APIs
• Natively formatted or encrypted files
• Searches where you don't want to send the whole file to the AI cloud engine
• Databases

So MCP servers understand how to read your data, can break up tasks with the help of the AI, and can launch AI tasks agentically. Well-written MCP servers send more focussed data to the AI engine, so you consume less tokens, that equates to lower cost. With AI assistance, an IT department could build an MCP server capable of tasks like these:

• Read help desk incidents and suggest diagnoses or fixes
• Query your CMDB or asset database without exporting data
• Run PowerShell commands against remote machines (with approval controls)
• Check AD/Entra for account status, group membership, and policy assignments
• Get license entitlement data and compare against installed software
• Monitor disk space, CPU, memory, and event logs
• Read and summarise documentation for specific applications
• Query Intune, SCCM, or other management APIs to check device compliance
• Search network logs and firewall rules for security anomalies
• Generate and file change requests or incident updates in your service desk
• Scan for expiring certificates, warranties, and contract renewals

Because the MCP server runs inside your network, only the results reach the AI — not the raw data. That means lower token costs, faster responses, and your sensitive data stays local.

This may seem like something only a developer could take on, but AI can write the MCP server for you.

Conclusion

Many ITAM teams are still either not using AI, or simply use it through a chat-bot. These tools now have the right level of maturity to help solve real problems quickly by adding in knowledge of your environment and local agents which have access to your data and environment.

Recommendation: Start with an agent-based AI tool such as Claude-Code or Codex. Using this as an agent allows you to generate scripts which act on your live data and your IT assets directly. To limit costs, you don't always need to use the cleverest model — the lower level models can perform many tasks perfectly well, so an account costing $20 a month can be enough to get a lot of value if you choose models carefully. For long and complex tasks, you can also ask the agent to launch sub-tasks using lower level models.

Agent based AI tools allow you to store files to train the AI — these files are used in every question, and can be simple text or MD files. These are worth creating early on, they should be dense and concise, with no repeated information. The AI tool can help you write them.

What is xAssets Doing with AI?

Everything in this article can be done today with mainstream AI tools and your data. xAssets is building AI directly into it's platform, with the first release planned for mid-2026:

• Ask questions about your asset data in plain English and get answers immediately
• Describe a report you want and the AI generates it in real time, with the correct columns, filters, and grouping
• AI configuration — describe a requirement in a sentence or paste in a requirements email, and the system builds the forms, queries, and workflows to match. Even a full Statement Of Work can be pasted in
• 24 MCP tools already in use — we have been using AI agents internally across development, testing, documentation, and security assessments since 2025, and this experience is feeding directly into the product
• Customer instances will be automatically documented with their specific configuration covered