About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.
Split tunnelling sends only selected traffic through the corporate VPN; other traffic uses the local internet path. Full tunnel sends essentially all device traffic through the VPN. When staff on Windows clients reach hosted line-of-business applications — by RDP, a browser client or a private hostname — the tunnel mode affects latency, security inspection and helpdesk load. The decision is yours; it is not implied by choosing managed hosting.
Why it matters
Hosted business apps compete with Microsoft 365, CRM SaaS and videoconferencing for bandwidth. Forcing full tunnel on a home NBN link can make hosted sessions feel slow even when the hosting environment is healthy. Conversely, permissive split tunnelling can expose broader internal reach if VPN routes are too wide — a risk your security team must judge.
Staff may work from office fibre, regional 4G or hotel Wi-Fi. One global VPN profile rarely fits all without explicit exceptions and monitoring — editorial view, not vendor guidance.
How it works
Microsoft's VPN routing decisions page (for Windows VPN profiles) states that the force-tunnel versus split-tunnel choice impacts configuration, capacity planning and security expectations, and that in split-tunnel mode only listed routes use the VPN. On that page, exclusion routes are the mechanism for traffic that should not traverse the tunnel — in standard practice, named ranges are sent outside the tunnel (author reading of Microsoft's wording; confirm against the VPNv2 CSP if you deploy via policy).
Intune's Windows VPN settings describe enabling split tunnelling and adding routes with destination prefix and prefix size (inclusion lists for split tunnel). That page also covers DNS suffix search lists and NRPT rules — pair those with hosted hostnames so split profiles do not cause "works in office, fails on VPN" failures.
Mechanism summary an engineer can act on:
- Classify destinations. List hosted app entry points (gateway, private IP or hostname), internal file shares still on-prem, and SaaS that must be inspected.
- Build route lists. In split tunnel, only prefixes on the inclusion list traverse the VPN; use exclusion routes on the routing page (above) when named ranges must bypass the tunnel. Document FQDNs and CIDRs; the force-versus-split choice itself drives capacity planning and security expectations per Microsoft — not individual prefix sizing alone.
- Match tunnel mode to path. If hosted apps are reachable only on a private network attached to hosting, those prefixes must be on the VPN regardless of split versus full for everything else.
- Measure, do not guess. From representative sites, trace which path packets take to the hosted name and to major SaaS. Editorial: latency and MTU friction often show up here before cutover night — validate in your environment.
The pattern table below is editorial judgement for hosted-app migration planning, not Microsoft or vendor prescription.
| Pattern | Split tunnel often fits when… (editorial) | Full tunnel often fits when… (editorial) |
|---|---|---|
| Hosted LOB via internet gateway | Strong app auth, narrow inclusion routes, acceptable local egress risk | Centralised inspection, DLP or conditional access require all egress through one control point |
| Hosted LOB via private link | Only private prefixes need VPN; SaaS can bypass | Policy requires all traffic through one firewall stack |
| Branch with site-to-site link | Branch uses site-to-site for LOB; VPN for travellers only | Policy requires all egress through a centrally inspected firewall path |
Opposing view (editorial): teams that mandate full tunnel often cite malware egress inspection, uniform logging and simpler helpdesk triage ("one path to inspect"), even when hosted entry points are already public HTTPS or RDP with strong authentication. Split tunnel trades some central visibility for latency and user adoption.
This article is staff VPN design, not repair of a site-to-site tunnel that is up with no useful traffic — see Site-to-site IPsec VPN up but no traffic.
For access path choice (gateway versus site-to-site versus user VPN), see RD Gateway or site-to-site VPN for staff reaching hosted applications.
What to choose and why
Choose narrow split routes when performance complaints block adoption and your hosted entry points are individually authenticated. Document the exact prefixes and FQDNs included; avoid "split except everything" configs that become full tunnel in disguise.
Choose full tunnel when regulatory or insurer requirements demand centralised inspection — editorial recommendation; confirm against your own policy. Full tunnel does not, by itself, make an untrusted or unpatched endpoint safe.
Revisit the choice after migration, when traffic patterns change: fewer on-prem servers may mean fewer private routes, but new hosted integrations may need added exceptions.
Read next
Related articles you may want next.
- HostingRD Gateway or site-to-site VPN for staff reaching hosted applicationsChoose a per-user RDP gateway or routed site connection by the resources staff need, with separate identity, segmentation and MFA decisions.Read article
- TroubleshootingPort Forward Works From the Internet but Not Inside: NAT Hairpin or Split DNSPublic name works outside but not on the LAN? Prove the NAT hairpin fault with captures, then choose hairpin NAT (DNAT plus SNAT) or split DNS with the TLS name intact.Read article
- TroubleshootingIPsec VPN Tunnel Up but No Traffic: Read-Only DiagnosisAn IKEv2 tunnel can be up while no data flows. Check child SAs, selectors, NAT, routing and packet size on both strongSwan gateways before any change.Read article
Risks and limits
VPN route selection is not authorisation or segmentation. Narrow inclusion routes only steer which destinations use the tunnel; they do not stop malicious traffic from a compromised endpoint on other paths. Pair tunnel design with VPN access controls, firewalls, application authentication and endpoint management — documented separately from route lists.
Over-broad split tunnel routes can expose more internal reach than intended on a compromised laptop.
Under-tested MTU and DNS (editorial). Split profiles that omit hosted DNS suffixes or NRPT coverage cause name-resolution failures on VPN; align Intune DNS settings with Port forward works from the internet but not inside: NAT hairpin or split DNS and your VPN vendor's assignment rules.
Provider scope. Private connectivity into hosting depends on what your provider offers — see VPC networking.
Not lab-validated. Profiles described conceptually; test in your MDM or firewall lab before production.
Sources
- Microsoft Learn — VPN routing decisions — force versus split tunnel; impacts configuration, capacity planning and security expectations; exclusion routes.
- Microsoft Learn — Windows VPN settings in Microsoft Intune — split tunnel routes (prefix, prefix size); DNS suffix and NRPT.
- Site-to-site IPsec VPN up but no traffic — related tunnel troubleshooting.
- VPC networking — related service link.
Learning notes
- Split tunnel steers some traffic outside the VPN; full tunnel steers essentially all on Windows clients.
- Editorial: hosted-app performance on home links may depend on tunnel mode — measure on your paths.
- List hosted destinations explicitly in VPN inclusion routes or exclusion routes.
- Full tunnel has security arguments beyond hosted-app convenience — weigh them explicitly.
- Re-test after migration when on-prem dependencies shrink.
This article was originally published on .