Skip to main content

The Incremental Revenue Channel Nobody A/B Tests: How Phone CTAs Generated Hundreds of Sales Without Cannibalizing Digital

We tested phone CTAs across 3 brands. Every test passed non-inferiority on digital AND generated incremental sales. Here's the playbook for testing additive channels.

A
Atticus LiApplied Experimentation Lead at NRG Energy (Fortune 150) · Creator of the PRISM Method
14 min read

Editorial disclosure

This article lives on the canonical GrowthLayer blog path for indexing consistency. Review rules, sourcing rules, and update rules are documented in our editorial policy and methodology.

Fortune 150 experimentation lead100+ experiments / yearCreator of the PRISM Method
A/B TestingExperimentation StrategyStatistical MethodsCRO MethodologyExperimentation at Scale

The conventional wisdom in digital optimization is that phone CTAs cannibalize digital conversion. The reasoning sounds airtight: every user who calls instead of clicking "submit" is a user who would have converted online. Adding a phone number is a leak in the funnel, not a feature.

I believed this too, until the data proved otherwise — three times in a row, across three different brands, in the same regulated consumer service category.

Every phone CTA test we ran passed non-inferiority on digital enrollment. Every single one. And every single one generated incremental sales that would not have existed without it. In one test, phone sales nearly doubled from roughly 234 monthly conversions to roughly 469 while digital enrollment held perfectly flat. In another test, we went from zero phone-originated sales in the control — because the control had no phone CTA at all — to 577 calls and 23 confirmed sales in the variant.

These were not marginal effects. They were the kind of results that change how you think about channel design.

The Cannibalization Myth: Why the Conventional Wisdom Is Wrong

The cannibalization argument assumes that phone-preferring users are digital users who have been offered a detour. Under this model, adding a phone number pulls users out of a funnel they would have completed online, trading a low-cost digital conversion for a high-cost phone conversion.

This is wrong at the level of user segmentation.

Phone-preferring users are not digital users who prefer to call. They are a distinct segment with a different psychological profile, different trust requirements, and in many cases, different product complexity. In a high-consideration service category — think energy plans, insurance, financial products — the decision involves enough friction, enough fine print, and enough perceived risk that a meaningful percentage of users will not complete the online flow without human guidance. Not because the online flow is broken. Because their comfort threshold for committing without talking to a person has not been reached.

What happens to these users when no phone CTA is available? One of two things: they struggle through the digital flow at a lower conversion rate than a phone-confident user would achieve, or they abandon entirely and you never see them again. Either way, you are losing them. You are not keeping them.

Adding a phone CTA does not divert users you were keeping. It captures users you were losing. The segmentation difference is the entire argument.

Our data confirmed this directly. Across every test where we could track call outcomes, phone callers converted at roughly three times the rate of digital visitors. Not because phone agents were more persuasive — in some cases they were, in some cases they were not. But because the users who self-selected to call were already at a higher intent level than the average digital visitor. They had read enough to want to proceed. They just needed one more thing: a human who could answer their specific question.

Key Takeaway: Phone-preferring users are a distinct segment, not diverted digital users. Without a phone CTA, they either convert at lower rates or abandon entirely. Adding a phone CTA captures users you were losing — it does not divert users you were keeping.

The Non-Inferiority Framework: Testing Additive Channels Without Risking Core Metrics

Standard A/B testing is designed to answer: "Does variant B outperform control A on our primary metric?" Non-inferiority testing answers a different question: "Does variant B perform no worse than control A, within a defined acceptable margin?"

The distinction matters enormously when you are testing an additive channel.

If I add a phone number to a digital enrollment page, my hypothesis is not that digital conversion rates will increase. My hypothesis is that (a) digital conversion rates will not meaningfully decrease, and (b) the additional phone conversions will more than compensate for any minor changes in digital behavior. The primary success condition is non-inferiority on the digital metric, not superiority.

Running this as a standard superiority test is not just methodologically wrong — it is organizationally dangerous. A standard test that shows flat digital conversion will be read as "no significant improvement," which is technically accurate and practically misleading. A non-inferiority test that shows flat digital conversion while also generating incremental phone sales is a clear win, framed correctly from the start.

The way to structure this test:

Primary metric: Digital enrollment (must pass non-inferiority, meaning the lower bound of the confidence interval must not fall below a pre-specified margin, typically 2-5% below control).

Secondary metric: Phone-originated enrollments (incremental volume above the control baseline).

Success condition: Non-inferiority on primary AND any positive incremental volume on secondary.

With this framing, the business case is constructed before the test launches. The product team, the call center team, and the CRO team are aligned on what success looks like. There are no surprises in the readout — just a check against a pre-committed decision framework.

In our three tests, the actual margin on digital enrollment was narrower than the pre-specified threshold in every case. The phone channel added net-new volume. Non-inferiority was achieved comfortably. The result in each case was an unambiguous yes.

Key Takeaway: Frame phone CTA tests as non-inferiority tests, not superiority tests. Pre-specify the acceptable margin before launch. This prevents a flat digital result from being misread as "no effect" when the incremental phone volume is a genuine win.

Three Brands, Three Wins: The Pattern Behind the Results

What makes the three-brand consistency noteworthy is that these tests were not designed together. They emerged independently from different test roadmaps, different pages, and different hypotheses. The convergence of the results was not planned — it was discovered in retrospective analysis.

Brand A ran a phone CTA variant on the primary digital enrollment flow as part of a broader test on conversion channel support. The control had no phone option. The variant added a persistent phone number with a supporting headline designed to address trust concerns. Digital enrollment in the variant was statistically equivalent to the control — within 0.4% of control performance, well inside the non-inferiority margin. Phone-originated enrollments in the variant: 577 calls, 23 confirmed sales, all incremental.

Brand B tested a phone CTA embedded in a mid-funnel page where users who had started the enrollment process but not completed it were most likely to drop. This was a high-anxiety moment in the flow — users had already shared personal information and were about to commit. The phone CTA gave them an exit ramp to human support rather than exit from the funnel. Digital enrollment held flat. Phone sales nearly doubled relative to the brand's existing phone baseline, growing from roughly 234 to roughly 469 monthly conversions in the variant period.

Brand C ran a sitewide phone CTA test — the most ambitious of the three — placing the phone number in the persistent navigation header across all enrollment-relevant pages. The concern going in was that sitewide placement might introduce more ambient distraction and pull users out of the digital flow earlier. The data did not support this concern. Digital enrollment was statistically non-inferior, and phone volume grew proportionally with the increased visibility.

Three different brands. Three different test designs. Three identical outcomes. The consistency is the signal.

Key Takeaway: Across three independent tests on different brands and different page types, every phone CTA test passed non-inferiority on digital AND generated incremental phone sales. The pattern is robust enough to treat as a near-universal truth in high-consideration service categories.

The Mobile Dominance Pattern: 78-88% of Phone CTA Engagement

If there is one execution insight that shapes how I now design these tests, it is the mobile engagement concentration.

Across every phone CTA test where we could segment engagement by device, mobile accounted for 78-88% of total phone CTA interactions. This is not a trivial distribution skew. It is a near-total concentration of usage.

The implication is simple but frequently missed in execution: the phone CTA must be designed for mobile first. This means:

Click-to-call functionality is mandatory. A phone number that requires the user to manually dial on mobile is a friction-added feature, not a convenience. The CTA should trigger the native dialer directly. Any implementation that displays the number without enabling click-to-call will substantially underperform.

Mobile placement must be primary. In desktop layout, a phone number in the header or alongside the enrollment form is visible without effort. On a 375px screen, placement choices matter. A phone CTA buried below the fold or tucked into secondary navigation will generate a fraction of the engagement of one that appears in a sticky header or immediately following the primary form.

Mobile design should not be an afterthought. In several of the early iterations, mobile styling was treated as a responsive breakpoint of the desktop design. The results improved when the mobile phone CTA was designed as a standalone element — distinct size, distinct visual treatment, distinct context copy — rather than a scaled-down version of desktop.

The 78-88% mobile share also has implications for call center staffing. Mobile users calling from a service enrollment flow are typically calling in the early evening, on weekends, and during commutes — not during standard business hours. If your call center is staffed 9-5 Monday-Friday, you are capturing the minority of your phone CTA-driven call volume. This is a real operational tradeoff, but it is a manageable one — and it is not a reason to avoid the phone CTA.

Key Takeaway: Mobile accounts for 78-88% of phone CTA engagement. Design click-to-call as the default, prioritize mobile placement, and plan call center staffing around mobile usage patterns (evenings and weekends), not desktop office-hour assumptions.

The Call Center Efficiency Tradeoff: More Volume at Lower Average Quality

One nuance I want to address directly because it came up in every post-test business case discussion: phone CTAs on digital enrollment pages typically generate more call volume at a lower average conversion rate per call than traditional call-center-originated traffic.

This is not a paradox. It is a selection effect.

Traditional call-center traffic — users who found a phone number through a directory, a TV ad, or a direct mail piece — has been pre-filtered by the friction of seeking out the number. These are high-intent users who have made a deliberate decision to call as their first interaction. Their conversion rate per call is high.

Digital enrollment users who call via phone CTA are a more mixed segment. Some are at the "just need one more question answered" stage and convert immediately. Others are earlier in their decision process and are using the call to gather information they will act on later. The average conversion rate per call is lower than traditional call-center traffic — in our tests, meaningfully lower, roughly 40-60% of the traditional rate depending on brand and placement.

But the question is not what the per-call rate is. The question is whether the incremental sales volume is positive. And across every test we ran, it was. Even at a lower per-call conversion rate, the volume of calls generated by the phone CTA produced more net-new sales than zero — because zero is the counterfactual.

The call center efficiency calculation is: (incremental call volume) × (per-call conversion rate) × (average sale value) versus (cost of handling the call volume). In every case we modeled, this was clearly positive. The margin on incremental sales more than covered the cost of the additional calls, with room to spare.

There is also a secondary efficiency gain that is harder to quantify: users who call and do not convert in the first interaction do not convert to zero. Some of them convert on a follow-up call, some convert digitally later, and some generate referrals. The immediate-conversion calculation understates the total value.

The Phone Number Tracking Challenge: What Happens When You Get It Wrong

There is one operational detail that caused us real measurement pain in one of the three tests, and it is worth documenting specifically because it is easily avoided.

Accurate phone CTA testing requires a unique toll-free number (TFN) for each variant — a TFN that is exclusively associated with the test variant and not used anywhere else. This is the only way to attribute phone-originated sales cleanly to the variant versus the control.

In one test, we learned partway through the runtime that the control variant's phone tracking was contaminated — the number used in the control had previously appeared in other marketing contexts, and some of the "control" call volume was actually coming from users who had seen the number elsewhere and were calling that number directly. The control phone sales figure was inflated, which made the incremental lift calculation unreliable.

The fix was not complicated after the fact, but the test had to run longer to accumulate clean data. The lesson is simple: assign a fresh, never-before-used TFN to each variant before the test launches. Do not reuse numbers from previous tests, campaigns, or channels. The attribution integrity of the entire test depends on this single operational step.

Dynamic number insertion (DNI) systems make this tractable at scale — a technical layer that serves variant-specific TFNs based on the session cookie that identifies which test variant the user is in. If your testing platform supports DNI integration, use it. If it does not, assign static unique TFNs manually and verify they are clean before launch.

Key Takeaway: Assign a fresh, never-reused toll-free number to each test variant. Contaminated control phone tracking invalidates the incremental lift calculation. Dynamic number insertion systems automate this — use them if available.

Sitewide vs. Page-Specific: Where Phone CTAs Work Best

The three tests used different placement strategies — mid-funnel page, end-of-enrollment flow, and sitewide header — and each produced positive results. But the economics were meaningfully different.

Page-specific placement, particularly at high-anxiety moments in the funnel (just before the commitment step, on the plan selection page, or in the confirmation flow), generates the highest conversion rate per call. These are users who are close to converting and have a specific blocking concern. A phone call resolves that concern and closes the sale.

Sitewide placement generates higher raw call volume but lower per-call conversion rates. More users who are earlier in their research process will call. The economics work out, but the staffing requirement is higher and the average handle time per sale is longer.

My current recommendation for teams building the business case for a phone CTA program: start with page-specific placement at the highest-friction moment in your funnel. The per-call economics are easier to defend, the staffing ramp is more manageable, and the test is simpler to instrument. Once non-inferiority is established and the incremental value is demonstrated, expand to sitewide placement with the first test as the proof of concept.

The one context where I would recommend against phone CTAs: purely transactional, low-consideration digital products where the decision cycle is measured in seconds, not minutes, and where the audience has self-selected into a digital-first channel. In those contexts, the phone-preferring segment is small, the per-call economics are poor, and the risk of funnel distraction outweighs the incremental gain. But in any category where the purchase decision involves uncertainty, risk perception, or product complexity — which covers most high-consideration consumer services — the phone CTA tests consistently.

Building the Business Case Before the Test

One of the structural advantages of non-inferiority framing is that it forces the business case to be built before the test runs. This is unusual in CRO, where most teams build the business case after the results are in — and adjust the framing based on what came out.

For a phone CTA test, the pre-test business case requires two estimates: the acceptable non-inferiority margin for digital enrollment, and the minimum incremental phone volume required to justify the call center investment. Both are answerable before the test launches.

The non-inferiority margin is typically set by the business stakeholders who own digital enrollment metrics. In most organizations, a margin of 2-3% below control is acceptable — meaning if the variant produces digital enrollment within 2-3% of control, the digital metric is considered held. This number should be explicit and agreed-upon before the test launches.

The minimum incremental phone volume calculation is: (call center cost per acquired customer via phone) versus (value of an incremental customer). If your call center costs $40 per phone-acquired customer and your average customer value is $200, you need to generate one phone sale per roughly 6-7 handled calls to break even. Given our observed per-call conversion rates, this threshold is typically easy to clear at even modest call volumes.

GrowthLayer's test design workflow includes a pre-test calculator for exactly this kind of business case — it forces the team to specify non-inferiority margins and minimum effect sizes before any variant is built. The discipline of doing this work upfront is what separates tests that produce clear decisions from tests that produce ambiguous data and inconclusive readouts.

Key Takeaway: Build the business case before the test — not after. Specify the non-inferiority margin and minimum incremental phone volume required to justify call center investment. This framing turns a flat digital result into a clear win when phone sales are additive.

The Next Frontier: What Phone CTA Programs Should Track

The three tests I have described were, in retrospect, first-generation phone CTA work. The incremental value was real and the non-inferiority framework worked. But there is a second layer of analysis that most teams are not yet doing, and that second layer is where the program gets significantly more valuable.

Call type segmentation: Not all calls are equal. A call from a user on the plan selection page has a different conversion profile than a call from a user who hit an error in the payment flow. Systems that can tag calls by entry page, by session stage, and by call outcome create a data asset that enables much more targeted phone CTA placement. If calls from page X convert at 3x the rate of calls from page Y, the placement strategy changes.

Callback rate and timing analysis: In our tests, we observed that users who engaged with the phone CTA during off-hours — when the call center was closed — had a measurably lower conversion rate even when they called back during open hours. The intent had cooled. A callback reservation system, where users who click the phone CTA outside business hours can schedule a call for the next available slot, captures some of this lost volume. We have not tested this systematically yet, but the signal is clear enough that it is on the roadmap.

CSAT monitoring by acquisition channel: Phone-acquired customers may have a different support burden and a different retention profile than digitally-acquired customers. If phone-acquired customers contact support twice as often in the first 90 days, the per-call economics need to be adjusted to reflect the higher lifetime cost. Monitoring this allows the business case to be continuously refined rather than locked in at initial test conditions.

The phone CTA is not a campaign tactic. It is a channel decision. When it is treated as one — with proper instrumentation, ongoing monitoring, and continuous optimization — it becomes one of the most defensible incremental revenue streams in a digital enrollment program.

Conclusion

The data from three brands is unambiguous: adding a phone CTA to digital enrollment experiences does not cannibalize digital conversion. It captures the users you were already losing to abandonment or friction. Phone-preferring users are a distinct segment who self-select into a channel that matches their trust threshold — and who convert at three times the rate of digital visitors because they are already at high intent.

The non-inferiority framework is what makes this testable in a way that the business can commit to before the test runs. Specify the acceptable margin. Define the incremental volume threshold. Run the test. Read the result against the pre-specified criteria.

In every case we tested, the result was the same: digital held, phone added, and the incremental value was clearly positive. The test that would have taken weeks to approve under a superiority framing took days under a non-inferiority framing, because the decision criteria were clear and the risk was bounded from the start.

The phone channel is not a relic of pre-digital commerce. In high-consideration categories, it is an active conversion lever that most digital programs have left sitting on the table.

If you want to design non-inferiority tests with pre-specified margins, track phone-channel attribution alongside digital metrics, and build the business case before your test launches, [GrowthLayer](https://growthlayer.app) gives you the framework to run additive channel tests with the rigor they require.

About the author

A
Atticus Li

Applied Experimentation Lead at NRG Energy (Fortune 150) · Creator of the PRISM Method

Atticus Li leads applied experimentation at NRG Energy (Fortune 150), where he and his team run more than 100 controlled experiments per year on customer-facing surfaces. He is the creator of the PRISM Method, a framework for high-velocity experimentation programs at large enterprises. He writes regularly about the statistical and operational details of A/B testing — the parts most CRO content skips.

Keep exploring