Try It Live
Click Analyze Device Health to auto-scan your browser-accessible device info, optionally supplement it with metrics you can pull directly from your OS using the built-in guides, and receive an AI health assessment with a score, risk factors, and replacement recommendation. If replacement is warranted, fill in the procurement form and submit — a structured request is emailed to IT instantly.
How It Works
-
Auto-scan — The agent reads what your browser exposes: operating system, CPU core count, available RAM (Chrome/Edge), battery level, screen resolution, and browser storage quota. These are displayed instantly, no input required.
-
Asset database callout — In a live internal deployment, this step is skipped entirely: device age, model, disk usage, and utilization history are pulled directly from your asset management system (Jamf, Microsoft Intune, ServiceNow CMDB, or a custom HRIS). The demo surfaces this integration point prominently so stakeholders understand the production architecture.
-
Manual supplement — For metrics the browser can't read (device age, disk usage %, average CPU and RAM utilization, reported symptoms), the form provides optional fields. Each field includes an expandable "How to find this" guide with step-by-step instructions tailored to your detected OS (Windows or macOS).
-
AI health assessment — GPT-4o-mini scores the device across five dimensions (Age, Performance, Storage, Reliability, Security), each out of 20 points. The overall score drives a recommendation: REPLACE (0–40), MONITOR (41–65), or KEEP (66–100). The AI returns a plain-English reasoning summary, a list of specific risk factors, and a ready-to-send procurement justification.
-
Procurement request — If replacement is recommended, the agent pre-fills a procurement form: suggested replacement model (OS-aware), priority level, shipping preference, and an editable AI-drafted justification. Review and submit — the request is emailed to your IT team immediately.
What This Demonstrates
| Capability | Description |
|---|---|
| Browser hardware detection | Reads OS, CPU, RAM, battery, screen, storage via Web APIs |
| AI scoring & reasoning | Multi-dimension health assessment with structured JSON output |
| OS-aware contextual help | Platform-specific guides appear based on detected OS |
| Asset database integration point | Clear callout showing where Jamf/Intune/CMDB would connect |
| Human-in-the-loop | Manager reviews all data before anything is submitted |
| AI-drafted communications | Procurement justification drafted from structured device data |
| Email delivery | Structured HTML procurement request routed to IT |
From Demo to Production
This demo evaluates one device at a time using browser-available data. A production deployment would connect to your asset management platform and evaluate your entire fleet on a schedule.
Real-World Challenges
| Challenge | Why It Matters |
|---|---|
| Asset inventory accuracy | Most CMDB/asset databases are 60–80% accurate — devices go untracked, creating blind spots |
| Heterogeneous fleet | Windows, Mac, Linux, and mobile each have different health indicators and lifecycle expectations |
| Budget cycle alignment | Replacement recommendations need to align with annual IT budget planning, not just technical need |
| Vendor management | Preferred vendors, volume discounts, and lead times affect procurement decisions |
| Employee disruption | Replacing a device mid-project has a productivity cost — timing matters |
| Sustainability / disposal | Lifecycle management includes secure data wiping, recycling compliance, and asset recovery |
Cost Estimates (Monthly)
| Component | Starter | Growth | Enterprise |
|---|---|---|---|
| AI API | $10–40 | $40–150 | $150–600 |
| MDM/CMDB integration (Jamf, Intune, SCCM) | $100–400 | $400–1,500 | $1,500–5,000 |
| Procurement system integration | $50–200 | $200–600 | $600–2,000 |
| Asset tracking infrastructure | $0–100 | $100–400 | $400–1,500 |
| Total monthly | ~$150–600 | $600–2,500 | $2,500–8,000 |
ROI Definition
- Primary metric: Reduction in unplanned device failures — target 40–60% fewer emergency replacements
- Secondary metrics: Extended useful life for devices that don't need replacing, standardized procurement
- Break-even: Typically 3–6 months
- Concrete example: An unplanned laptop failure costs $800–1,200 in emergency procurement premium + 1–3 days of lost productivity ($400–$1,200). For a 200-device fleet experiencing 15 failures/year, proactive replacement saves $18,000–$36,000/year vs ~$8,000/year tool cost.
Build It or Buy It?
The asset register is free. Nobody sells the part you actually want. — priced comparison of the three routes
Knowing what you own is a solved problem, and it has been free for years. Knowing what is about to break is the thing this demo does, and it is not a product you can buy on its own.
| Route | What you pay | 200 devices, 3 IT staff |
|---|---|---|
| Self-host the register — Snipe-IT open source | Nothing. Free, open source, unlimited assets | $0 licence |
| Buy the register hosted — Snipe-IT cloud | Basic $39.99/mo ($399.99/yr), Small Business $99.99/mo ($999.99/yr) | $400–1,000/year |
| Buy an ITSM suite — Freshservice | $19/agent/mo Starter, $49 Growth, $99 Pro, billed annually. Asset Units for ITAM are licensed separately in packs of 500. Freddy AI Copilot is a further $29/agent/mo | $1,764/yr for 3 agents on Growth, before asset units and before AI |
Freshservice bills on three independent axes — agents, asset units, and AI — and only the first is on the headline price. A quote built from the per-agent figure alone will be wrong, and wrong in the direction that shows up after you have committed.
Meanwhile Snipe-IT self-hosted is genuinely free for unlimited assets. If the goal is an accurate inventory with check-in/check-out and warranty dates, that is a solved problem with a mature open-source answer and the licence cost is zero.
So what are you actually building? Not the database. The prediction. None of these tools tell you which device is likely to fail next quarter, because none of them hold the data that would say so. Battery cycle counts, SMART attributes, thermal events, crash frequency and disk health live in your MDM — Jamf, Intune, SCCM — not in the asset register. The register knows a laptop is three years old. The MDM knows its battery is at 68% design capacity and it has thermally throttled two hundred times this month.
That is the honest framing of this build: you are joining procurement data to telemetry that already exists but has never been put in the same place. The AI is a small part of it and, as with most of these demos, close to free.
What building it actually requires — skills, systems and the ongoing work
Skills you need on hand
| Area | Why it is needed |
|---|---|
| MDM integration | Jamf, Intune and SCCM each expose device health differently, and the useful signals are not the ones on the dashboard. This is where the predictive value lives, and it is the real work. |
| Asset reconciliation | The register says you own 200 devices, the MDM sees 187, and finance depreciates 214. Reconciling those three is unglamorous, tedious, and the precondition for everything else. |
| Basic failure modelling | Battery cycles and design-capacity decline are genuinely predictive and simple to reason about. Resist reaching for anything sophisticated before you have a year of your own failure history to test against. |
| Procurement integration | A prediction that does not become a purchase order is a report nobody reads. |
| Warranty and lease tracking | Knowing a device fails is worth much less than knowing it fails while still under warranty, which changes who pays. |
The part that is easy to underestimate: your failure history is probably not written down. Devices get swapped, the old one goes in a drawer, and nothing records why. Without that, there is nothing to learn from and nothing to validate a prediction against. Start capturing replacement reasons before you build the model — that ordering is not optional, and it costs a year if you get it backwards.
The second thing: a refresh recommendation is a budget request. It will be argued with. It needs to show its reasoning — this device, this evidence, this cost of doing nothing — or it will be overruled by whoever holds the budget, every time.
Choose self-hosting the register when you need accurate inventory and nothing more. It is free, mature, and you will not beat it.
Choose an ITSM suite when asset management is one requirement among many and you are buying ticketing, service catalogue and change management anyway. Price the asset units and the AI add-on explicitly before comparing.
Choose build when you already have both an asset register and an MDM, and the gap is that neither one is telling you what to replace next.
What to expect if you go ahead — timeline, accuracy, and where it goes wrong
Timeline. Joining an asset register to MDM telemetry takes days. Getting the reconciliation clean enough to trust takes weeks, because it surfaces every device nobody could account for.
Accuracy, stated honestly. Battery degradation is genuinely predictable — it is a physical process with a well-understood curve. Storage failure has useful leading indicators. Almost everything else, including the failures that actually hurt like liquid damage and drops, is not predictable at all and never will be. A system that claims to forecast those is fitting noise.
Set the expectation accordingly. The win is not a crystal ball. It is converting a subset of emergency replacements — the predictable subset — into planned ones, so you buy at list price on your schedule rather than at a premium on somebody else's.
Where it actually goes wrong
- Ghost assets. Devices in the register that no longer exist, inflating both the fleet count and the budget built from it.
- Shared and shelved devices look like healthy low-usage machines, so the model recommends keeping equipment nobody has switched on in a year.
- Recommendations arrive out of sync with the budget cycle. Technically right, financially unactionable, and it teaches people to ignore the system.
- Warranty status is stale, so a covered failure is paid for out of pocket. This is a pure, avoidable loss and it is common.
- The alert fires and nothing happens because no one owns the resulting purchase. A prediction with no owner is a notification.
The honest question to ask first: do you have MDM telemetry and a year of honest replacement history? With both, this is a genuinely valuable build and the data is already yours. With neither, you are not ready to predict anything — start with a free Snipe-IT register, record why every device gets replaced, and revisit in twelve months with something worth modelling.
Sourcing note. Prices above are list prices read from vendor documentation in August 2026: Snipe-IT, Freshservice. Freshservice Enterprise is quote-only, and its pricing page does not state how many Asset Units each plan includes or what a 500-unit pack costs — that figure is deliberately absent rather than estimated, and it should be asked for explicitly in any quote. Snipe-IT also lists dedicated hosting at $249.99/mo and larger dedicated servers at $5,000–$7,500/year. Jamf, Intune and SCCM licensing is not included in the figures above. Verify current pricing before committing.
Technology Stack
- AI Model: OpenAI GPT-4o-mini
- Email Delivery: Google Workspace SMTP (Nodemailer)
- Browser APIs:
navigator.userAgent,navigator.hardwareConcurrency,navigator.deviceMemory,navigator.getBattery(),navigator.storage.estimate(),window.screen - Frontend: React client component with multi-phase flow, auto-detect on mount, collapsible OS-aware guides
- Backend: Next.js API route (serverless) — two actions:
analyze(AI scoring) andprocure(email send)
Want This for Your Business?
A production deployment connects to your asset management platform (Jamf, Microsoft Intune, ServiceNow CMDB, or Lansweeper) to pull device records automatically. Hardware age, model, purchase date, incident history, and utilization data are evaluated on a schedule — flagging at-risk assets before they fail. Procurement requests route to your existing ticketing system (ServiceNow, Jira, Freshservice) with full audit trail. A full deployment typically takes 1–2 weeks and starts at $2,500.