What makes this hard
Handing a hundred offices to one company sounds like signing an outsourcing contract. In practice it all comes down to three questions: who owns it, how fast you turn up, and how you prove it.
No single counterparty
A local integrator runs part of it, the vendor another part, head-office IT owns change. When something breaks, responsibility is settled before diagnosis starts — and half the fixing time goes into meetings.
Nobody at night
Europe's daytime is China's small hours. Forwarding to one duty phone puts the whole P1 on a single person — who has to work again tomorrow.
A service you cannot measure
Responsive is an adjective. Without tiers, deadlines and a monthly score, quality rests on impressions and renewal talks have nothing to point at.
Documents always behind the network
Circuit lists, topology, high- and low-level designs — leave them six months and the next incident starts with rediscovery. Most root causes are documents, not hardware.
Delivery model: six things at once
Operations is not a role, it is a mechanism. Drop any one of the six below and within six months the service is a fire brigade again.
Three shifts, following the sun
Ten people across three shifts cover Europe, the Americas and Asia-Pacific. Nights go to SD-WAN and network management, days to change and project work.
L2 handles, L3 digs in
The first line works standard incidents from the knowledge base; complex ones escalate on a defined path to L3. Every ticket names who gets it next — no heroics.
A monthly SLA scorecard
Response, recovery, change success rate and documentation currency are all quantified. Anything missed gets an improvement plan that month and a review the next.
Through the customer's CAB
Request, technical design, review, implementation and quality check are all recorded. Emergency changes take a fast path, with the paperwork completed within 24 hours.
One ledger for circuits, topology and certificates
Cross-border circuit schedules, design documents, topology, SSL certificates and backups — reconciled quarterly, physically counted once a year.
Named contacts at vendors and carriers
Fixed contacts and escalation paths at Equinix, Fortinet and the carriers, so time is not lost waiting for a reply.
Commitments: times written into the tiers
The commitment is split into four tiers, each with a stated response time, and attainment is reported monthly.
| Severity | Response time | When it applies |
|---|---|---|
| P1 | 30 minutes | Core service down or the whole network unusable — emergency response starts at once and management is notified |
| P2 | 8 hours | A critical link or device degraded, affecting part of the business or some sites |
| P3 | 24 hours | A single failure or performance issue with no impact on continuity |
| P4 | 48 hours | Advisory, improvement requests, routine changes |
Missing a time is handled as the contract sets out. We do not promise that nothing fails — that is vendor language. We promise that when it does, someone is on it in a stated time and you get a conclusion in a stated time.
Reporting: a quality record you can measure
Weekly
This week's incidents, changes carried out, backup status and open items. One page, sent on Friday.
Monthly report and SLA scorecard
SLA attainment, capacity trends, IDS and authentication events, asset changes and next month's plan, walked through at the service review.
Annual review
A year of incidents and their root-cause distribution, architecture recommendations, next year's capacity and budget, and the physical asset audit result.
Major incident report
For P1 / P2, issued within 48 hours of recovery: timeline, root cause, blast radius, remediation and prevention.
A live project
A diversified international group headquartered in Hong Kong, China, spanning logistics, cosmetics, pharmaceuticals and trading: a hundred offices and some thirty thousand staff across several timezones. A team of ten took on its network operations — three consecutive years without a major service interruption, SLA met throughout, renewed annually.
The full post-mortem — seven kinds of site problem, how ten mechanisms were put in place, and where it stood after three years — is on the case page.
How we start
Inventory the current state
We count devices, circuits, vendor contracts and open issues, and put the judgement in writing: what still has life, where the risks are, where the operational load actually sits. Scope and duration follow the project.
Parallel running (1–2 months)
Both providers run in parallel while the incumbent is still in place: knowledge transfer, document rebuild, monitoring onboarding and an emergency drill.
Steady state
Delivered to the SLA, scored monthly. Each quarter a service review turns the question of expanding to other sites into a decision backed by data.