When to Switch From MPLS to SD-WAN (and When Not To)

Last updated: 7 July 2026

Here is when to switch from MPLS to SD-WAN: when most of a branch’s traffic is SaaS or internet-bound and decent broadband exists there. Here is when not to: when it doesn’t. The rest of this is how one Indian company actually worked out which was which.

The renewal notice was sitting on top of the patch panel in the server room, not on anyone’s desk. That is usually where paperwork goes to be ignored for a week before somebody notices it is actually a deadline.

I was at a 46-branch logistics and courier company headquartered in Pune, hubs running out to Nashik, Nagpur, Surat and Indore, there to look at why branch staff kept complaining that the parcel-tracking app “hangs” every afternoon. The network admin, Prashant, had the MPLS renewal letter open on his laptop. The carrier wanted a fresh 3-year lock-in, at a price about 14% higher than last time, and the fine print said if nobody responded in writing, it would auto-renew on its own.

Ninety days. That was the actual clock running, not the afternoon slowdown. I have seen this same letter arrive in a dozen server rooms by now, always with a different company name on the letterhead and the same 90-day window buried on page two.

The real cost behind the MPLS invoice

Prashant’s HQ ran two 20 Mbps MPLS circuits for redundancy. The 44 branches ran on single MPLS links between 2 and 8 Mbps depending on branch size. Add it up and the company was paying carrier-grade prices for links that mostly carried a warehouse app, email, and a parcel-tracking SaaS tool that lives on the public internet anyway. Per Cisco’s own WAN architecture guidance on carrier MPLS pricing, that premium is the cost of a guaranteed, carrier-backed SLA, not a markup for no reason.

MPLS earns its price with a carrier-backed SLA: guaranteed latency, guaranteed jitter, and someone else’s problem if a link goes down. That is worth paying for on the handful of links carrying something that cannot tolerate a rough patch. It is not worth paying for on a link whose busiest job all day is loading a browser tab.

When to switch from MPLS to SD-WAN: the short answer

Direct answer: switch from MPLS to SD-WAN when most of a site’s traffic is SaaS or internet-bound and a decent broadband option exists there. Keep MPLS, at least for now, on sites with no reliable broadband or with genuinely latency-sensitive traffic. Everything in between runs hybrid for a while. That is the honest answer for most Indian branch networks in 2026, not a hedge.

The plan that lasted about eleven minutes

My first instinct, and I will say this plainly because it was wrong, was to recommend cutting all 46 sites to broadband SD-WAN in one project. Cato’s cloud backbone would give every branch a single tunnel into the nearest point of presence, the warehouse app and the tracking SaaS would both improve, and the MPLS bill would drop by more than half.

That plan lasted about as long as it took Prashant to pull up the branch list. Six of the tier-3 town branches did not have a broadband option with anything close to a real SLA. One branch’s backup internet was a mobile hotspot the branch manager paid for out of his own pocket during monsoon outages. Cutting MPLS there would not have saved money. It would have traded a known, boring problem for a new, exciting one, right before peak festive shipping season.

I had walked in ready to retire the whole MPLS footprint on principle. The principle was right for maybe 38 of the 46 branches, and wrong for the other 8. That correction, arre, is the actual point of this piece, not the part where SD-WAN wins.

The branch-by-branch traffic map

Once we stopped asking “MPLS or SD-WAN” as one question for the whole company, and started asking it per branch and per traffic type, the decision more or less made itself. For Prashant’s 46 branches, the map came out to 38 on SD-WAN within four months, 6 tier-3 branches staying on MPLS with SD-WAN as a broadband overlay whenever local internet improved, and 2 branches kept exactly as they were because their MPLS contracts still had 26 months left and the exit fee made switching pointless this year.

Cato, Viptela, or building it yourself

The other decision Prashant had to make was which SD-WAN model to run, and this is where a lot of the DIY cost comparisons quietly go wrong. A self-managed stack, Cisco Viptela or VMware Velocloud style, needs its own controller infrastructure, its own orchestrator licensing, and a team that knows how to run it. Cisco documents this controller layer (vManage, vSmart) as a core, separately licensed part of the architecture. That stack alone can eat 15 to 20% of the savings you were counting on, before you have touched a single branch router.

None of that makes Viptela or Velocloud bad products. They are good products built for a different kind of buyer, one with a dedicated network engineering team and an existing Cisco or VMware estate to fold them into. For a 46-branch logistics company whose network team is Prashant and one junior engineer, a single-vendor platform without its own controller stack to run was the honest recommendation. Cato also carries ISO/IEC 27001 certification on its platform security, which mattered to Prashant’s compliance team once security and WAN sat on the same vendor instead of three.

Consolidating WAN and security onto one platform has a quieter benefit too: fewer separate logs to hand over if CERT-In’s incident-reporting rules or MeitY’s DPDP framework ever require an audit trail across the network. That was not the reason Prashant made the switch, but it made the compliance review afterward considerably shorter.

Where we landed, ninety days later

The renewal letter got answered on day 82, with a one-year extension on just the 2 branches that genuinely needed to stay put, instead of the full 3-year lock-in on all 46. That single move, buying one year instead of three, is the part I would tell any network head to copy even before they touch SD-WAN.

Six months in, the tracking app complaints had mostly stopped, the WAN bill was down by a little over 40%, not the 60% the optimistic slide promised, because the 8 branches on MPLS still cost what MPLS costs. That gap between the slide and the invoice is normal. Anyone promising the full number on day one has not looked at your branch list yet.

Label everything twice. That includes the decision itself: label which branches are SD-WAN-ready today, and label which ones are not yet, and revisit the second list every quarter instead of once every three years when a renewal letter forces the issue.

Key takeaways

• The trigger to act is usually your MPLS renewal notice, not a general sense that SD-WAN is trendy. Indian carrier contracts typically need about 90 days’ notice before they auto-renew.
• Decide branch by branch and traffic type by traffic type. A blanket “cut it all” plan breaks on the branches with no real broadband alternative.
• Expect a hybrid phase of 6 to 18 months. That is the normal outcome for most Indian networks, not a sign the project stalled.
• A self-managed SD-WAN stack has real controller and licensing costs that eat into the promised savings. Factor that in before comparing sticker prices.
• Buy the shortest renewal term you can on any circuits you are keeping. It protects the timeline for the rest of the migration.

Talk to us about a branch-by-branch WAN cost review

Frequently asked questions

Can I run SD-WAN over my existing MPLS circuit instead of replacing it?
Yes. This is the most common first step for Indian enterprises. The SD-WAN overlay runs on top of the MPLS link alongside broadband, so latency-sensitive traffic can keep using MPLS while everything else shifts to the cheaper path. See what SASE actually means for an Indian business buyer for the underlying architecture.

How long does a real MPLS to SD-WAN migration take?
For a multi-branch Indian network, plan for 4 to 9 months depending on branch count and how many sites need fresh broadband circuits ordered from scratch, which can itself take 30 to 45 days in smaller towns.

What is the payback period?
Most mid-size Indian deployments see a break-even between 10 and 18 months once controller, licensing and migration costs are counted honestly, not just the headline circuit saving. Our own Cato deployment bake-off walks through one such rollout end to end.

When should a business NOT move off MPLS yet?
If most branches lack a reliable broadband option, if the existing MPLS contract has years left with steep exit fees, or if a core application genuinely cannot tolerate variable latency, staying on MPLS a while longer is the correct call, not a failure to modernise.

Does moving to SD-WAN mean giving up a carrier SLA entirely?
Not necessarily. A hybrid design keeps MPLS or a premium leased line under the SD-WAN overlay for the sites and applications that need a guarantee, while everything else runs on standard broadband. If data residency and audit trail matter for your DPDP compliance posture, a single-vendor platform simplifies that review too.

Get my free quote for a WAN circuit audit

Explore the full Cato Networks SASE in India platform, part of our wider cloud and security portfolio. It runs SD-WAN and security off one cloud platform, without a separate controller stack to license and run. If you are staring at a renewal letter with a deadline on it, that is usually the right week to have this conversation, not the week after you have signed for another three years.

Talk to us before your renewal notice expires

P.S. Naveen here. A logistics client’s renewal letter had a 90-day window on it and nobody opened it for three weeks. If yours is sitting on a patch panel somewhere too, open it today. The clock started the day it arrived, not the day you read it.