Definition
A parallel run is a planned period during which the old and new versions of something operate at the same time on the same work, the same traffic or the same budget, so their results can be compared before the old version is switched off. In the site's guides the something is a process, such as the rule that hands leads or accounts to sales; a platform, such as the system that buys display advertising; or a tag set, such as the tags that send purchases to analytics and ad platforms.
The aim is to learn, while the old version still protects the business, whether the new one does at least as well on the measure that matters, and to find faults before any target or budget depends on the new figures.
What makes a parallel run readable
Learning something from a parallel run takes four decisions made before it starts.
- One measure, from one source. Both versions are judged on the same agreed figure, read from the same place over the same dates. Each system's own reports help diagnose problems but do not settle the verdict.
- A baseline. What the old version produced before the run starts, recorded, so the comparison is with a known figure rather than with memory.
- An end date. Agreed in advance with the people who will read the results, so that doubled or shifting numbers in a dashboard surprise nobody, and so the run does not drift on. If the answer is still unclear on that date, extending the run is better than switching on a hunch.
- A pass that lasts. The new version has to hold up across the whole run, not on its first good day.
One caution applies: the versions can interfere. Two ad platforms reaching the same people raise frequency, and a report that reads both tag routes counts conversions twice.
In practice: lead hand-offs, platform moves and server-side tags
Lead hand-offs. The tutorial on replacing MQLs with account qualification begins in shadow, with the CRM flagging the accounts the new rule would hand over while routing still follows the old threshold, then moves one segment to the live rule while the rest carry on as before. The account-based marketing guide points to that sequence for any team moving from lead-by-lead hand-offs to account-level ones.
Platform moves. The tutorial on moving display campaigns out of Google Ads keeps the existing campaigns spending while the new platform's line items start, judges both on the measure agreed at the outset and shifts budget one tactic at a time at review points written down beforehand.
Server-side tags. The server-side tagging go-live checklist treats the run as part of the cutover gate: destinations are compared during the run, the old setup stays ready to restore, and the cutover checks are ticked only after the destination checks have passed for its full length.
Behind all three sits the reasoning of the framework of one-way and two-way doors: keeping the old version running while the new one proves itself turns a switch that would be hard to undo into one that can still be reversed, at the cost of some duplicated work.
A shoe retailer catches a double count before it matters
Heldeien Shoes, an online shoe retailer, moved its purchase tags to a server container and kept the old browser tags running beside them for 14 days agreed with its marketing lead. The one measure was purchases each destination received, set against orders in the shop's back office. Over the fortnight the back office recorded 1,180 orders. Analytics matched within 1% on both routes, but one ad platform received 2,240 purchases from the server route, close to two for every order: a trigger in the new container fired on the order confirmation page and again on the account page that lists past orders. No budget had yet moved on the new figures. The trigger was fixed, a fresh 14 days were run, and the browser purchase tags were removed only when that platform's count stayed within 2% of the back office.
Related terms
A Staged RolloutReleasing a change to a growing share of visitors, pages or regions in steps and reading each before the next; a test only if groups are random and concurrent. moves a change to a growing share of traffic or regions in steps, while a parallel run keeps both versions whole for a period; the display platform move combines them, running both platforms and shifting budget in stages. Server-Side TaggingRunning tags on a server the business controls instead of in the visitor's browser. is the tag move whose go-live gate asks for a parallel run before the browser tags go. An MQLA lead that marketing judges ready for sales follow-up against agreed criteria. count keeps being calculated during the hand-off run, so the old stage can still be read until it is retired. A RollbackReturning a changed system, such as a tag container, workflow or page template, to its last working version, planned and rehearsed before the change goes live. is simplest while the old version is still running; once it has been switched off, going back depends on a saved previous version. ABMA B2B approach that targets a chosen list of companies with coordinated marketing and sales. is the account-level approach the hand-off run moves a team towards.
Common confusions
- Anything happening at the same time is a parallel run. Concurrent workstreams, reviews or ad sets are simply parallel work. A parallel run compares an old and a new version of the same thing before one replaces the other.
- A parallel run is a controlled test. Unless the groups were assigned at random, a difference between them is an estimate, not proof.
- Each system's own report can decide. Two platforms counting in their own ways cannot settle a comparison; one agreed measure from one source can.
