Six months from kickoff to "the team trusts the inventory". This is the timeline of a real GLPI asset-tracking deployment at a 200-person Slovak manufacturing company. Anonymised per editorial policy — no client name, no specific products, no identifying details. What stays in is the sequence of decisions, what broke, what stuck, and what would be done differently next time.
The starting state
The customer ran a precision-machining operation across two sites. IT footprint: ~180 office workstations (1 – 7 years old), ~40 production-floor PCs running CAM and QC software, ~15 servers, two Cisco core switches with ~30 access switches and dispersed APs. Software: M365, AutoCAD, Solidworks, ERP, antivirus, ~20 specialised tools.
The "asset register" was a 600-row Excel maintained by one senior IT specialist. Last bulk audit: 14 months prior. Licences in a separate spreadsheet, contracts in a Google Calendar. When something broke, locating it relied on tribal knowledge.
Month 0 — kickoff
Two-day workshop with IT and operations. Output:
- Decision: GLPI self-hosted on existing internal VM (8 vCPU, 16GB RAM, MariaDB on the same box).
- Decision: phased rollout — workstations first, then servers, then production-floor PCs (those needed locked-down agent profiles).
- Decision: Excel master would NOT be used as initial import — it was too out-of-date. Use GLPI Agent automatic discovery as the source of truth, validate against Excel afterwards.
- Taxonomy decided: Notebook, Desktop, Server, Network device, Phone (mobile), Phone (desk), Printer, Production-PC. Custom fields: "Site" (Bratislava / Trnava), "Department", "Production zone" (for production-PCs).
Month 1 — install + agent rollout
GLPI 10.0 installed on Wednesday. By Friday it was up with admin profiles created. Following Monday the agent rollout began via Active Directory GPO push:
# GPO startup script (PowerShell)
$installer = "\\fileserver\install\glpi-agent-1.7-x64.msi"
$server = "https://glpi.internal.example.sk/marketplace/glpiinventory/"
msiexec /i $installer SERVER=$server RUNNOW=1 /quietBy end of week 4, GLPI Agent was running on 165 of 180 workstations. The 15 misses were laptops belonging to people who hadn't been in the office (sales team, holidays). Discovery had identified 12 machines that weren't in the original Excel — three were ghost entries (decommissioned but still in the spreadsheet), nine were genuinely undocumented.
Month 2 — first reality checks
What broke this month:
- Production-floor PCs refused agent install. They ran a hardened image with Windows Defender Application Control blocking the agent service. Resolution: separate code-signed install package and an exception in the AppControl policy. Took 8 working days.
- Software inventory was overwhelming. Each workstation reported 200 – 400 installed packages. Most were Windows components, drivers, runtimes — noise, not assets to track. Resolution: built an exclusion list in Setup → General → Inventory → Software exclusions, dropped the noise to ~30 tracked items per machine.
- The first attempt at importing licences from Excel failed (column mismatch, dates in DD/MM/YYYY vs YYYY-MM-DD). Resolution: cleaned the source spreadsheet first, reimported successfully.
Month 3 — contracts and SLAs
This is where the soft savings started showing up. Loading the supplier and contract data into Management → Contracts revealed:
- Two server-maintenance contracts that had auto-renewed in the prior year despite the servers being decommissioned. Total wasted: €3,200/year.
- Antivirus enterprise licence was billed for 220 seats; actual installs: 178. Renegotiated at next renewal: −42 seats × €38 = €1,596/year.
- Three workstations identified as "in repair for 6+ months" — actually scrapped, but never marked as such. Operations manager hadn't had the data to know.
Month 4 — physical audit + barcode labels
First physical audit using GLPI's barcode export (Code 128 labels printed on a Zebra ZD230). Two pairs of technicians walked both sites with phones over three days. Discrepancies: 11 monitors moved without IT knowing, 4 laptops belonging to leavers (two recovered, two written off), production-zone allocations 23% wrong after a 2024 reorganisation.
Month 5 — helpdesk integration
With assets tracked, tickets started linking back to specific machines. New incidents on workstations got the asset linked automatically (because the requester's primary machine was now known). Patterns emerged:
- One specific laptop model (5-year-old line) was responsible for 38% of hardware incidents in the period. Conclusion: prioritise that model in the next refresh cycle.
- Three production-floor PCs were responsible for 60% of production-floor IT tickets. Investigation revealed they ran an older OS that was incompatible with a recent ERP update. Replacement scheduled.
Month 6 — the inventory the team trusts
By month 6, the inventory had stopped being "the IT manager's project" and started being "the system the operations manager checks before quarterly capex planning". Annual savings identified to date:
- Cancelled redundant maintenance contracts: €3,200/year.
- Renegotiated antivirus seats: €1,596/year.
- Avoided one-off 2-laptop-replacement cost from recovering written-off equipment: €1,800.
- Estimated downtime reduction from preemptive replacement of the bad-laptop model: not quantified but observable in the helpdesk trend.
What we'd do differently
- Skip the Excel-first import attempt — agent rollout produces a cleaner starting state. Excel reconciliation belongs in step 2, not step 1.
- Plan for hardened-PC quirks earlier — code-signing the agent for AppControl adds 1 – 2 weeks. Schedule it week 1, not week 5.
- Build the software exclusion list before rollout — avoids weeks of "why is GLPI tracking 80,000 items?" panic.
- Operations manager in the kickoff workshop — capex and contract decisions are the highest-leverage outcomes.
Six months is a realistic horizon for "deployed and trusted". Year 2 onwards the maintenance overhead drops while accumulated value compounds.