How to: Switch Survey Platforms Without Breaking Your Trend Data

Introduction

To switch survey platforms without breaking trend data, keep every question, scale and invitation identical, then run the old and new platforms side by side on the same population for each tracked metric. Compare the two results against a tolerance you set in advance. If the gap stays inside it, connect the trend. If it doesn’t, flag the break or restart the series. Archive all historical responses outside both platforms.

Last updated November 2026.

Why Switching Platforms Changes Your Numbers

A survey platform isn’t a neutral pipe. It decides how a question renders on a phone, which order the answers appear in, when the invitation arrives, who gets reminded, and what happens when someone skips a question. Change the platform and you usually change several of those at once.

The tracked number moves, and the movement looks exactly like a change in customer sentiment. A research lead who reports that movement to an executive team as a finding has published an artifact of the migration. That’s the failure this guide exists to prevent.

Two products have announced retirements this year. SurveyMonkey has said GetFeedback Direct "will no longer be available after Dec. 31, 2026." Delighted, which Qualtrics acquired in 2018, ended customer access on July 1, 2026, so its customers have already moved. GetFeedback Direct teams are moving trackers now, whether they planned to or not.

What a Platform Switch Actually Involves

A tracker migration is four projects with different owners and deadlines, and running them as one is how teams end up with a broken series. The instrument rebuild recreates every survey that keeps running. The data archive moves history somewhere durable. The measurement bridge estimates how much the platform itself moves your numbers. The administrative cutover repoints accounts, integrations and dashboards.

The archive runs on your contract’s termination date, which is often the hardest deadline in the program. The bridge runs on your reporting calendar, and it has to start before cutover, not after. Get that order wrong and there’s no old arm left to bridge against.

What Breaks Comparability When You Change Platforms

Most of a tracker survives a platform change with care. A specific set of changes breaks comparability on its own, and the dangerous ones are the changes nobody decided to make. They arrive with the new platform’s defaults.

The table below lists the changes that most often move a tracked score, what each one does, and what to do instead. Read it before you set a cutover date, because several rows can’t be fixed after launch.

The Comparability-Breakers Inventory

| What changes | Does it break the trend | What you do instead | Source | |:---------------------------------------------:|:-----------------------------------------------------------------------------------------------------------------------------------------------------:|:---------------------------------------------------------------------------------------------:|:-----------------------------------------------------:| | Number of scale points | Yes. A 10-point scale scored 0.3 points lower than rescaled 5- and 7-point scales in one controlled test | Keep the exact scale length. Never rescale an old series to fit a new default | Dawes, International Journal of Market Research, 2008 | | Scale labels and anchors | Yes. A changed label changes what the number means | Copy every anchor word for word, including endpoints and midpoints | Forsta, tracking study migration guidance, 2021 | | Answer option order | Can shift results. Test it | Match the old order, or match the old randomization rule exactly | Check both platforms’ settings | | Answer codes and values | Yes, silently. Reports join on codes, not labels | Map every old value to the new one before launch, and test the export | Forsta, tracking study migration guidance, 2021 | | Rendering (slider, radio buttons, star icons) | Can shift results. Test it | Use the same input type, and check it on a phone | Test on both platforms | | Interview mode or channel | Yes. Mode alone moved 60 Pew items by 5.5 points on average | Hold the channel constant through the bridge. Change channel later, as its own bridged change | Pew Research Center, 2015 | | Invitation copy and sender | Sometimes. A new sender name changes who opens the email | Reuse the subject line, body and sender display name for the bridge period | Platform setting | | Trigger and timing | Yes, for in-product surveys. A survey shown after a task measures a different moment than one shown on login | Write down the old trigger event and delay, then match both | Your old configuration | | Sampling frame or panel | Yes. A different population is a selection effect, not a measurement effect | Keep the same contact list or user segment. Treat a panel change as a new series | OMB bridge study definition, 2006 | | Reminder cadence | Can shift who answers. Test it | Match the number and spacing of reminders | Your old configuration | | Forced response and "don't know" handling | Yes. Pew found phone respondents say "don't know" while web respondents skip | Match the required-answer setting and keep or omit "don't know" exactly as before | Pew Research Center, 2019 | | Weighting | Yes. After a 2011 weighting and sampling change, New Jersey's health department warned that 2011 results compare with earlier years only with caution | Apply the old weighting scheme to new data until you've bridged a new one | NJ Department of Health on BRFSS, 2011 | | Metric formula | Yes. Net Promoter Score (NPS) bands and top-2-box definitions vary by setup | Recompute the metric from raw responses using the old formula, written down | Your old reporting logic | | Historical responses | Doesn't port by default. Most destinations document no response import | Archive the full history outside both platforms (see The Historical Data Decision) | Destination platform docs |

Who Should Break the Trend Instead of Bridging It

Some teams shouldn’t bridge at all. A bridge study costs a parallel run, an analyst’s time and a delayed shutdown, and for these readers it buys very little.

Teams Whose Old Number Was Already Unreliable

If your old tracker had a falling response rate, a changing invitation list or a questionnaire that was edited mid-series, its trend already contained breaks. Bridging to a series that was never continuous spends money protecting a line that didn’t mean what the chart implied. Restart the series on the new platform and say so.

Teams Changing the Population on Purpose

Some switches exist to change who answers. Moving from an emailed customer list to a research panel, or from email to in-product, means the new number describes a different group. A bridge can show the size of the gap, but it can’t tell you which population’s number was right, because they’re different questions. Start a new series and keep the old one archived beside it.

Teams With No Executive Audience for the Trend

If nobody makes a decision from the line chart, the line chart doesn’t need protecting. An operational survey that drives a support queue or a product backlog needs consistent questions going forward, not a reconciled past.

When the Old Mode Was the Problem

Pew Research Center dropped prior years' data for at least one question when it moved online, because the online version measured the attitude with less error. Pew wrote that "breaking the trend may break a researcher’s heart," and broke it anyway. If your migration fixes a known measurement problem, keeping the old trend preserves the problem.

Decision rule: if the old series was unstable, the population is changing on purpose, or no one decides anything from the trend, break the trend, label the break and skip the bridge.

The Instrument Triage

Run every live instrument through the same short checklist before you plan the parallel run. The goal is to shrink the bridge to the few instruments that need one.

The Triage Checklist

List every instrument that fielded in the last 12 months, including ones nobody remembers owning.
Mark each one that feeds a metric with a target, a board slide or a compensation plan.
Mark each one compared year over year in any report.
Mark each one whose metric has a known seasonal pattern.
Name the downstream consumers of each export, like a dashboard, a warehouse table or a quarterly deck.
Find the contract clause that sets your export deadline, and name who owns it.
Retire every instrument that fails all of the above. It doesn’t need migrating at all.

Sorting the Results

Instruments with a target, a year-over-year comparison or an executive audience get a bridge. In many programs that’s a short list, like a relationship NPS survey, a core customer satisfaction (CSAT) measure or a brand tracker. Everything else cuts over directly, with its questions frozen and its old data archived.

This is the step that makes the parallel run affordable. A program with dozens of live surveys may end up bridging only a handful of them, which is a very different conversation with whoever owns the budget.

Decision rule: bridge an instrument only if someone outside the research team compares its numbers across time.

The Historical Data Decision

Historical responses are the one part of a migration with a contractual clock on it. Decide where they’ll live before you decide anything else, because once the old contract ends, the export window may end with it.

Most Destinations Don't Import History

Many destinations don’t document any way to import responses collected elsewhere. Check your destination’s documentation yourself, and treat a missing page as "not documented" until the vendor confirms otherwise in writing.

The platform you’re leaving may be the one that can import. Qualtrics documents response import from CSV or TSV files up to 100MB, says formulas and bucketed fields can’t be imported, sets the recorded date to the time of import, and asks customers to contact Account Services or their account executive before importing historical responses.

Where History Should Live

The answer is usually neither platform. Make the archive a named step with an owner and a finish date.

A data warehouse. The strongest home, because history stays queryable and joinable with product and revenue data after both survey platforms are gone.
A business intelligence layer. Next best, because the dashboards that read the old platform can be repointed at the archive instead of retired.
A frozen file archive. The floor. It holds labeled and coded exports of every instrument, the rendered questionnaires, the codebooks and a written index, with an owner and a retention policy.

The Archive Is a Privacy Object

Response files often carry email addresses, user IDs, free-text answers that name people, and attributes nobody remembers adding. Inventory the personal data before you load anything, and move only what trend analysis needs. That’s usually responses and segment attributes, rarely contact identifiers.

Where Sprig fits. Sprig’s export documentation, retrieved November 2, 2026, covers CSV export of up to 950,000 rows with response text, numeric values or both, and describes no response import. Sprig can feed the archive going forward, but your pre-switch history stays in the archive you build.

Decision rule: export everything, in both labeled and numeric form, and verify the row counts before the old account closes.

Freezing the Instrument Before You Rebuild It

The rebuild is where comparability is won or lost, and most of the damage happens through defaults. Freeze the old instrument as a written specification first, then rebuild from the specification, not from memory or screenshots.

What the Specification Holds

The specification holds the exact wording of every question and answer, the scale length and every anchor, the answer codes, the order and randomization rule, every required-answer setting, the trigger or send rule, the reminder schedule and the metric formula. It’s a document a second person could rebuild the survey from without asking you anything.

Forsta’s own guidance on tracker migration makes the same point. It says questionnaire labels, data labels, values and positions need to match so the old datamap lines up with the new dataset, and it warns that adding or removing answer options or changing question types alters results.

Where Sprig fits. The Design Agent can draft a study from a written specification, which speeds up the rebuild. But an agent tuned to improve question wording is the wrong tool for a frozen tracker, so tell it not to edit the questions and check its output against the specification line by line.

Decision rule: if a second person can’t rebuild the survey from your specification alone, the specification isn’t finished.

The Measurement Discontinuity

Changing platform changes how a question is measured, and it often changes who answers. Those are two different effects, and a migration plan has to separate them.

Selection Effects and Measurement Effects

A selection effect means different people answered. A measurement effect means the same kind of person answered differently because the question was asked differently. Both move the number, and both look identical on a dashboard.

You can’t separate them by judgment after the fact. Separating them takes a comparable dataset collected alongside the old one, from the same population, during the same period. That dataset is the bridge study.

How Large Mode Effects Get

Pew Research Center’s 2014 experiment randomly assigned 3,003 panelists to answer by web or by phone. Across 60 questions, the mean difference was 5.5 percentage points and the median was five points. The largest gap was 18 points, on satisfaction with family life, where 62% of phone respondents said they were very satisfied against 44% on the web.

Satisfaction items were among the most affected. That matters for anyone migrating NPS or CSAT, because those are satisfaction measures. Pew’s caveat applies, which is that a mode gap tells you the two modes disagree, not which one was right.

Why Satisfaction Scores Move Most

Satisfaction ratings are sensitive to who’s asking and how. Pew Research Center reports that phone respondents more often choose the most positive option on a rating scale, though not the most negative one, which it ties to giving a socially desirable answer to a live interviewer. A self-administered survey removes the interviewer, so the same person reports a little less satisfaction.

A platform switch between two self-administered tools doesn’t remove an interviewer, but it can change the cues around the question, like branding, page layout and where the survey appears. Those cues are smaller than an interviewer, and no published study we found measures them for software platforms, so treat their size as unknown until your own bridge measures it.

The House Effect

A house effect means the same instrument, run on the same population, produces a different number when a different organization fields it. It’s the closest published analog to a survey platform switch, because the mode stays the same and only the operator changes.

Gallup tested this directly.

Before moving its U.S. phone interviewing to the vendor ReconMR in January 2024, Gallup ran paired surveys in July, September and November 2023, one in-house and one with ReconMR each month. Gallup reported that "for most of the results, we observed no statistically significant differences," worked with ReconMR to adjust procedures where gaps remained, and said it may suppress a data point or report it without comparison if needed.

How Large Organizations Handled Transitions

The Bureau of Labor Statistics redesigned the Current Population Survey in January 1994. Before the switch, it ran a parallel survey of 12,000 households, against roughly 60,000 in the main survey, from July 1992 through December 1993, so it could estimate how the new methods would change labor force measures.

The CDC’s Behavioral Risk Factor Surveillance System added a cell phone sample and a new weighting method in 2011. The New Jersey Department of Health, publishing state results, warned that "comparisons between 2011 and prior years should therefore be made with caution." Pew, for its part, ran randomized mode experiments in 2014 and 2017 before deciding trend by trend.

Is There a Standard for Overlap Length?

There’s a standard that names the obligation and no published rule for how long an overlap should run. The U.S. Office of Management and Budget’s 2006 Standards and Guidelines for Statistical Surveys advises agencies to "develop adjustment methods, such as crosswalks and bridge studies that will be used to preserve trend analyses."

OMB defines a bridge study as one that "continues an existing methodology concurrent with a new methodology for the purpose of examining the relationship between the new and old estimates."

No duration appears. The precedents range from three paired months at Gallup to 18 months at the Bureau of Labor Statistics. Vendor guidance is vaguer still. Forsta’s 2021 tracker guidance says to run in parallel "for at least a few months (or waves)." Treat any fixed number of months as advice, not a standard.

The Commercial Tension

The methodologically correct answer is to run both platforms at once. You’re migrating partly to stop paying for two platforms. Both of those are true, and a plan that ignores the second one gets ignored itself. The way through is to bridge less, which the triage step already did, not to skip the bridge for the instruments that need it.

Decision rule: if a tracked number moves after cutover and you have no bridge, report the move as unexplained, not as a change in sentiment.

Designing the Bridge Study

A bridge study runs the same measure on both platforms, over the same period, to the same population, and compares the results. It’s the one instrument that tells you whether your series is continuous.

What to Hold Constant and What to Vary

Vary the platform and only the platform. Every other difference you allow becomes part of the estimated effect and can’t be separated from it later.

The Bridge-Study Design Spec

The Frame Problem

Splitting one population between two platforms is easy to write and often impossible to do.

An in-product survey can only run where the platform’s code is installed, and two platforms rarely run side by side in the same app. If you can’t split the frame, run both platforms on adjacent groups that you can show are similar on the attributes that drive the score, and call the result an estimate with a known weakness.

Setting the Tolerance Before You Look

Decide in advance how large a gap you’d accept as noise, and write it down before the first response arrives. A tolerance chosen after seeing the result will be chosen to fit the result.

Base the tolerance on the smallest change in the metric that would change a decision. If a three-point NPS move would trigger action at your company, a platform gap of three points isn’t acceptable, however small it looks. Then size each arm so the confidence interval on the gap is narrower than that tolerance, using a standard power calculation for a difference in proportions or means.

When You Can't Afford a Full Bridge

If budget or list size rules out a full split, run a smaller bridge on the new platform alongside the old one’s normal wave, and report its result as an estimate with its interval stated. A small bridge that comes back inconclusive is still useful, because it tells you to flag the trend, not connect it. What it can’t do is license a quiet connection.

Reading the Result

There are three outcomes. If the gap and its whole interval sit inside the tolerance, the series is continuous for this metric. If the gap sits outside the tolerance, the series isn’t continuous and you choose a trend treatment. If the interval straddles the tolerance, the bridge was too small to decide, and you either extend it or treat the result as a break.

Where Sprig fits. Run the comparison yourself, from raw exports, in your statistics tool. The Analyze Agent page lists crosstabs, t-tests and ANOVA, but the agent is marked as in preview, so check any result it gives against your own calculation before it goes in a report.

Decision rule: never declare a series continuous on a gap that’s merely "not significant." Declare it continuous only when the whole interval sits inside a tolerance you set before fielding.

Worked Example: Bridging a Quarterly NPS Tracker

This example is illustrative. The numbers are invented to show the arithmetic, and your own tolerance and sample will differ.

The Setup

A software company runs a quarterly relationship NPS survey by email. Its last four waves scored between 24 and 27, and its leadership reviews any quarter-on-quarter move of four points or more. That review threshold is the tolerance, so the team writes down ±4 NPS points before the bridge starts.

The team splits its usual mailing list at random into two halves. One half gets the old platform’s invitation and the other gets the new platform’s, with identical wording, sender name, send date and reminder schedule.

Sizing Each Arm

NPS is noisier than it looks, because each response falls into one of three groups. With about 45% promoters and 20% detractors, the variance of a single response is 0.65 minus 0.25 squared, or about 0.59. To get a 95% interval on the gap no wider than ±3 points, each arm needs roughly 5,000 responses.

The same precision on a top-2-box customer satisfaction share of about 80% needs roughly 1,400 responses per arm. So an NPS bridge needs about three and a half times the sample of a CSAT bridge at the same precision.

The First Result

After one wave, each arm has about 5,100 responses. The old platform scores 25.0 and the new one 22.6, a gap of minus 2.4 points. The 95% interval runs from minus 5.4 to plus 0.6.

The gap itself sits inside the tolerance, but the interval crosses minus four. The bridge hasn’t ruled out a platform effect large enough to trigger an executive review, so the result is inconclusive, not continuous.

The Extended Result

The team runs a second wave on both platforms and pools the two. With about 10,000 responses per arm, the gap is minus 1.8 points and the 95% interval runs from minus 3.9 to plus 0.3. The whole interval now sits inside ±4, so the series is continuous for this metric, and the team connects the trend with a footnote at the cutover point.

What the Example Shows

Two waves of overlap came from the sample the tolerance required, not from a calendar rule. A team with a tighter tolerance, or a smaller list, would need more waves or would have to accept a flagged break. Both are honest outcomes, and neither is visible until the arithmetic is written down.

Running the Parallel Run

The parallel run is the period when both platforms field the bridged instruments at once. It’s the most expensive part of the migration, so keep it as short as the bridge allows and no shorter.

What the Overlap Costs

The overlap costs two licenses for its length, two sets of instruments to keep identical, two result streams someone has to reconcile, and a team running the old program while building the new one. That last cost doesn’t show up on an invoice, so budget it as staff time.

How Long to Run It

Run it until each bridged arm reaches the sample size your tolerance calculation required. If the metric is seasonal and the season matters to the decision, extend the run to cover it. For a monthly tracker with a modest tolerance, that can mean several waves. For a continuous in-product measure, it depends on daily response volume, so do the calculation before you pick an end date.

Forsta’s 2021 guidance suggests starting with the old platform carrying most of the sample, about 75%, and shifting as confidence grows. That’s a reasonable way to limit risk, but a lopsided split needs a larger total sample to reach the same precision on the smaller arm.

Where Sprig fits. For emailed trackers, Sprig’s email surveys can send from your own domain and run multiple survey runs from one study, which keeps the new arm’s invitation close to the old one.

The Field Agent helps manage targeting and response monitoring. Neither one splits a population across two vendors for you, so do the random assignment before either platform sees the list, and check the Field Agent’s targeting against that assignment list.

Decision rule: end the parallel run when the bridge reaches its planned precision, not when the budget runs out. If the budget runs out first, record the result as inconclusive.

Choosing a Trend Treatment

Every bridged metric ends the migration with one of three treatments. Pew Research Center described the same three in a 2019 post on its move from phone to online polling, and it chooses among them case by case.

The Trend-Treatment Decision Tree

Did the bridge put the gap and its interval inside your tolerance? If yes, connect the series. Add a footnote with the cutover date, and move it to the methods section once several waves of new data are established.
Is the gap outside the tolerance, but stable and in one direction? If yes, connect the series with a visible flag: a marked cutover point, a labeled change in line style and a sentence in the report text stating the estimated platform effect.
Is there evidence the new measurement is more accurate than the old one? If yes, break the trend. Show only new-platform data for this metric and archive the old series with an explanation.
Is the result inconclusive? Treat it as a break until a later bridge or a longer run settles it.

Adjusting the Old Numbers

Some teams adjust the old series by the measured gap to splice one continuous line. Do this only when the gap was estimated with an interval narrow enough to trust, and keep the unadjusted series beside it. An adjustment built on a small bridge adds false precision to every past data point at once.

Decision rule: a connected trend needs a bridge that passed. Anything else gets a visible flag or a break.

Reporting a Platform Switch to Stakeholders

The trend treatment only works if the people reading the chart understand it. Trend damage can happen in a slide deck, when a flagged line gets copied without its flag.

What to Tell Executives Before Cutover

Tell the people who own the tracked metrics that the platform is changing, which metrics are being bridged, and what tolerance each bridge uses. Do it before the first post-switch number exists. A tolerance announced after a bad quarter looks like an excuse, and one announced before looks like a method.

How to Label the Chart

Mark the cutover point on every chart of a bridged metric. If the series is connected, a footnote naming the cutover date and the bridge result is enough. If it’s flagged, change the line style at the cutover point, label both sides, and state the estimated platform effect in the report text, as Pew Research Center does for its own flagged trends.

What to Say When the Number Moves

If the number moves after cutover by less than the bridge’s measured platform effect, say the move is within the expected effect of the switch. If it moves by more, report the excess as a change, with the platform effect stated beside it. Don’t report a post-switch move as sentiment until you’ve subtracted what the bridge says the platform did.

Decision rule: if a chart of a bridged metric can be screenshotted without its cutover label, the label isn’t on the chart itself, and it needs to be.

Sequencing the Migration Against the Old Contract

Lay the work out backward from the date the old contract ends. The export has to finish before the contract does, and the parallel run has to finish before the export closes.

The Sequenced Migration Plan

When the Dates Don't Fit

If the arithmetic doesn’t fit, you have three honest options: extend the old contract for the overlap, shorten the overlap and accept a weaker bridge, or cut over without a bridge and break the series deliberately. All three are legitimate. Choosing one by accident, because nobody laid out the dates, isn’t.

Decision rule: if the dates don’t leave room for both the export and the bridge, start with the archive, not the rebuild.

Changing the Channel After the Platform Switch

Many teams switch platforms partly to change channel, usually to add in-product or mobile app surveys alongside email. Stage those as two moves. Bridge the platform first on the old channel, then treat the channel change as its own break or its own bridge.

Why Staging Matters

A channel change shifts who answers and how they answer at the same time. An emailed relationship survey reaches customers who open email, while an in-product survey reaches users who are in the product during the window, often at a specific moment in a task. If you change platform and channel together, a single gap carries both effects, and no analysis can split it afterward.

Running Old and New Channels Side by Side

If the in-product measure is meant to replace the email one, run both on the new platform until each arm reaches the sample your tolerance calls for, and compare them the same way you compared platforms. Expect a gap. It’s a difference in population and moment, not an error, so the usual outcome is two series reported side by side, not one spliced line.

Decision rule: never change platform and channel in the same wave for a tracked metric.

Cutover and Validation

Cutover takes four steps in a fixed order, and shutting down the old platform is the last of them.

The Cutover Sequence

Stop new fielding on the old platform for each rebuilt instrument, but keep the account live and the data reachable. Complete the export, then verify it by opening the files and checking row counts and headers against expectations. A finished download isn’t a verified export.

Repoint every downstream consumer, including alerts and sends triggered from customer relationship management tools, and test each one with a live response. Then close the account.

Validating the New Platform

Validate the first few waves after cutover with the same discipline as the bridge. Check response rates by segment against the old platform’s last waves, since a drop concentrated in one segment is a selection effect in progress. Check the share of skipped and "don’t know" answers. Recompute every metric from raw exports with the written formula and compare it to what the new platform’s dashboard shows.

Validating Open-Text Coding

If the old platform coded open-text answers into themes, the new one will code them differently. Theme shares from two coding systems aren’t a trend. Keep the old theme series closed, and start a new one with a human-checked sample.

The Synthesize Agent groups open-text answers into themes. Like any automated coding, its output should be spot-checked against a human-coded sample before its theme shares go into a trend.

Decision rule: don’t close the old account until the archive is verified and every downstream report has run at least once against the new source.

Where Your Current Platform Wins

Leaving a platform means giving up things it does well. Name them before you start, because discovering them mid-migration, with the contract running out, is far more expensive.

Historical Response Import

Qualtrics documents a way to import historical responses into a survey, with stated file types and limits. Sprig’s documentation describes no equivalent. If keeping all history inside one survey tool matters to your team, the platform you’re leaving may be the one that can do it, and the destination may not.

Published Statistical Methodology

Qualtrics publishes the tests and assumptions behind its Stats iQ analysis, including Welch’s t-test, Mann-Whitney U, Fisher’s exact test and Pearson’s chi-squared. A platform that names its tests can be defended in a room with a methodologist.

Sprig’s Analyze Agent lists crosstabs, t-tests and ANOVA, is marked as in preview, and comes with no published weighting approach or power guidance. For a bridge study, that means the reader runs the analysis outside the survey tool.

Compliance and Hosting Reach

Qualtrics announced FedRAMP authorization in July 2018, with the Department of Health and Human Services as sponsoring agency. Sprig’s security standards page, retrieved November 2, 2026, lists SOC 2 Type II, HIPAA, GDPR and CCPA, names AWS facilities in the U.S. as its hosting location, and doesn’t list FedRAMP or regional data residency. If your program needs either, check that list against your requirements before any trend work starts.

Decision rule: if any of these three is a requirement rather than a preference, resolve it before you plan the bridge.

Common Mistakes When Switching Survey Platforms

Starting the bridge after cutover. By then there’s no old arm to compare against.
Improving the questions during the move. Every edit becomes part of the platform effect and can’t be separated from it.
Treating "not significant" as "the same." A small bridge often lacks the power to detect a real gap.
Bridging everything. A full parallel run of every survey is why teams give up on bridging entirely.
Trusting the new dashboard’s metric. Recompute it from raw responses with the old formula.
Changing channel and platform at once. Each change needs its own bridge, so stage them.
Letting the new platform’s defaults decide. Check every setting in the frozen specification against the rebuilt survey, because defaults are where unplanned changes come from.
Bridging on a different population. A bridge on a convenience sample estimates the effect for that sample, not for your tracker.
Deleting before verifying. Open the export files and count rows before the old account closes.

How Sprig Supports a Tracker Migration

Sprig is an enterprise survey platform powered by AI agents, and it fields trackers through email, links, panels, in-product surveys on websites and web apps, and native mobile apps. Sprig can field the same NPS or CSAT question as part of its experience measurement work and attach user attributes to each response in the export.

For a migration, the useful parts are multiple survey runs from one study for longitudinal tracking, user attributes that can be included in the export, and CSV export with both response text and numeric values.

The limits matter as much. Sprig documents no response import, so history lives in your archive. Its export is CSV only, and a study over 950,000 rows has to be split with date filters before it’ll download. The Analyze Agent page doesn’t mention comparing results across survey runs, so the bridge comparison and any trend analysis happen in your own tools.

Frequently Asked Questions

Will My NPS or CSAT Score Change When I Switch Survey Platforms?

Your Net Promoter Score (NPS) or customer satisfaction (CSAT) score will probably move when you switch survey platforms, usually not because customers changed. Pew Research Center found a mean gap of 5.5 percentage points across 60 questions when only the interview mode changed, from phone to web, with satisfaction items among the largest. Run a bridge study to estimate your own metric’s effect first.

How Long Should I Run Two Survey Platforms in Parallel?

No statistical authority publishes a required length for running two survey platforms in parallel. Gallup used three paired months before changing phone vendors, and the Bureau of Labor Statistics ran 18 months before its 1994 redesign. Size your overlap from the sample each platform needs to measure the score gap within the largest gap you’d accept as noise, and extend it for seasonal metrics.

What Is a Bridge Study in Survey Research?

A bridge study in survey research runs an old method and a new method at the same time, on the same population, so the two estimates can be compared. The U.S. Office of Management and Budget defines it as a study that "continues an existing methodology concurrent with a new methodology." For a platform switch, it estimates how much the platform itself moves a tracked metric.

Can I Import My Historical Survey Responses into a New Platform?

Importing historical survey responses depends on the destination, and many document no import at all. Where the destination documents none, history belongs in a warehouse, business intelligence tool or frozen file archive. Qualtrics, often the platform teams are leaving, does document response import, with a 100MB file limit and a request to contact its account team first.

Should I Change My Survey Questions When I Switch Platforms?

You shouldn’t change survey questions for any tracked metric during a platform switch. Any edit to wording, scale or answer order becomes part of the measured platform effect and can’t be separated from it afterward. Freeze the questions, migrate, bridge, and make improvements later as a separate, separately bridged change.

Can I Switch from Email Surveys to In-Product Surveys Without Breaking My Trend?

Switching a tracker from email surveys to in-product surveys changes both how people answer and who answers, so it usually breaks the trend. A bridge can measure the gap but can’t decide which population’s number is right. Treat the in-product measure as a new series, and keep the email series running or archived beside it.

How Do I Explain a Survey Trend Break to Executives?

Explain a survey trend break to executives before it happens. Tell them which metrics are being bridged, the tolerance for each, and how the chart will be labeled if the gap falls outside it. After cutover, report any move as the platform effect plus the remaining change, with the bridge result beside the number.

Why Does an NPS Bridge Need More Responses Than a CSAT Bridge?

An NPS bridge needs more responses than a CSAT bridge because NPS splits respondents into three groups, which makes each response more variable than a yes-or-no satisfaction share. In one illustrative case, measuring the gap within ±3 points took about 5,000 responses per arm for NPS, against about 1,400 for an 80% top-2-box CSAT share.

How Do I Keep Alerts and Integrations Running During Cutover?

Keep alerts and integrations running during cutover by listing every automated action the old platform triggers, like sends from a customer relationship management (CRM) tool or low-score alerts to a support queue, and rebuilding each one before the old platform stops fielding. Then send a test response through each path. A low score that triggers nothing during the switch is a missed customer follow-up.

Does Changing Survey Vendor Change Results if the Mode Stays the Same?

Changing survey vendor can change results even when the mode stays the same, which survey researchers call a house effect. Gallup tested a switch of phone interviewing vendors with three paired months in 2023 and found no statistically significant differences for most results, but it still adjusted the new vendor’s procedures to close the gaps that remained.

Do I Need a Statistician to Run a Bridge Study?

You don’t need a statistician to run a bridge study, but someone has to own two calculations: the sample size each arm needs to measure the gap within your tolerance, and the confidence interval on the gap once the data is in. Both are standard formulas for a difference in proportions or means, and both should be written down before fielding starts.

What Should I Export Before My Survey Platform Contract Ends?

Before a survey platform contract ends, export every instrument’s full response data in both labeled and numeric form, the rendered questionnaires, codebooks or datamaps, user attributes, and any computed fields you report. Open the files and check row counts before the account closes, since a finished download isn’t a verified export.

What if My Old Contract Ends Before a Bridge Study Can Finish?

If your old contract ends before a bridge study can finish, you can extend the contract for the overlap, run a shorter bridge and accept a less precise result, or break the series on purpose and label it. Export and verify the full archive first in every case, since the data deadline is the one you can’t recover from.