A pattern that already won once is not evidence it will win again somewhere else in the funnel — it's a better-odds hypothesis, and the discipline is in testing that transfer instead of assuming it.
TL;DR
- A four-step progress bar with the first step pre-checked, added to a plan-selection page at a large energy retailer, lifted conversion 5% to 10% (Exp-049).
- The same pattern had already won earlier in the funnel, at mobile checkout. We didn't assume a repeat — we ran a separate experiment to test whether the goal-gradient effect would transfer to a structurally different stage.
- Checkout and plan selection put the visitor in different psychological postures: one is committed and transactional, the other is still comparing options. That difference is exactly why the transfer wasn't automatic.
- The win produced more than one page's worth of lift — it gave us portfolio-level evidence about which behavioral patterns are stage-agnostic and which are stage-specific, a distinction visible only across multiple experiments, never from one.
- Outcome: the variant was scaled into the permanent experience.
Where the pattern had to prove itself twice
| Funnel stage | Visitor's posture | Progress bar tested here before Exp-049? | Result |
|---|---|---|---|
| Mobile checkout | Committed — has chosen, now finishing the transaction | Yes — already a validated winner | Established pattern |
| Plan selection (Exp-049) | Comparing — still deciding, weighing alternatives | No — untested at this stage | +5% to 10% lift, scaled |
The decision at stake
This ran on a multi-step plan-selection page in the energy-retail category: a visitor enters a zip code, lands on a page to select a plan, moves on to enter their details, and confirms the order. Plan selection is the pivotal screen in that sequence — it's where someone who has already spent a small amount of effort deciding whether to keep going or leave for a competitor's site or a comparison page. The same fork shows up anywhere a business asks a visitor to clear several screens before they get value: multi-step signup flows, insurance quote engines, SaaS onboarding wizards. The underlying question repeats regardless of category — does showing people how far they've come make them more likely to finish?
Two pieces of soft evidence existed before anyone ran anything. A nearly identical progress bar had already been validated at checkout, later in this same funnel. And competitors in the category had progress indicators live on their own plan-selection pages. Neither piece of evidence resolved the decision on its own — the checkout win came from a different stage of the journey, and a competitor shipping a pattern is evidence of a bet, not evidence of a result. Both signals pointed toward the same design change, but neither one was proof it would hold at plan selection specifically, and that gap is exactly why this ran as its own experiment instead of getting shipped on the strength of the earlier win.
The mechanism: the goal-gradient effect and endowed progress
The behavioral bet here is old and well studied. In 1932, Clark Hull documented what became known as the goal-gradient hypothesis, observing that effort and speed increase measurably as a subject nears a reward, not before it. The same acceleration shows up in people completing any bounded task: the closer the finish line feels, the harder they push to reach it.
The refinement that makes this useful in growth work came from Ran Kivetz, Oleg Urminsky, and Yuhuang Zheng's 2006 study in the _Journal of Marketing Research_, "The Goal-Gradient Hypothesis Resurrected." They issued loyalty cards requiring ten purchases for a reward to two groups: one started at zero of ten stamps, the other started at two of twelve stamps — the same number of purchases still required, but with an artificial two-stamp head start already applied. The group with the head start completed the program faster, even though the real distance to the reward hadn't changed. Kivetz and his coauthors named this endowed progress: progress that's granted rather than earned still produces the acceleration effect.
That's the exact mechanic at work here. A visitor arriving at the plan-selection page has done exactly one thing — entered a zip code. The variant showed that action as the first of four steps already checked off, reframing it as a fifth of a short, finite process instead of the start of an open-ended one. The goal-gradient effect predicts they push harder toward the next step once that mark is visible. Endowed progress predicts it doesn't matter that the head start was small, or that the business handed it to them rather than them earning it outright — the completion behavior still shows up.
The judgment call: testing transfer, not assuming it
This is the part that's easy to get wrong, and it's where the actual experimentation judgment lives — not in picking the pattern, but in refusing to treat "it worked at checkout" as license to skip a test at plan selection.
Checkout and plan selection are not equivalent moments, even though both sit inside the same conversion sequence. Checkout is the closing phase: the visitor has already selected a plan and entered their details, and their posture is committed — they're not evaluating alternatives anymore, they're finishing a transaction they've already decided to complete. A progress bar at checkout is talking to someone who wants confirmation they're almost done.
Plan selection is an earlier, comparison-heavy stage. The visitor hasn't chosen anything yet. They're often working through this exact page next to a competitor's plan-selection page in another tab, actively weighing price and terms. A pattern that reduces friction for someone who has already decided to buy is not automatically safe for someone who is still deciding whether to buy at all — a progress bar could just as plausibly read as pressure to an undecided visitor, making the page feel more like a funnel than a comparison tool, and push them toward the exit instead of the next step.
That's the actual hypothesis under test: not "does a progress bar increase conversion," which had already been answered at a different stage, but whether the underlying goal-gradient effect generalizes across a change in visitor posture, from committed-transactional to comparing-undecided. Running a second experiment instead of shipping the change sitewide is the slower, more expensive option in the short term. It's also the only way to know, rather than assume, whether a pattern that worked once will keep working somewhere structurally different.
The result
Exp-049 came back a winner: the variant with the pre-completed, four-step progress bar lifted plan-selection conversion 5% to 10% over the control. The team scaled it into the permanent experience.
That range matters more than it looks. It sits in the same order of magnitude as the checkout win, at a stage of the funnel with a meaningfully different visitor posture. The effect didn't happen to show up twice by coincidence — it showed up twice because the psychological trigger underneath it, a visible marker of progress within a bounded task, doesn't require a committed buyer to function. It requires a short, finite process and a visible mark of progress inside it. Both stages had that structure, even though almost nothing else about the visitor's mindset matched.
Why this matters beyond one experiment
One winning experiment tells you a pattern worked once, in one context. It doesn't tell you whether that pattern is a general-purpose lever or a one-off that happened to fit its stage. That distinction only becomes visible across a portfolio of experiments run at different funnel stages, in different categories, over enough time to see where a pattern repeats and where it breaks.
This experiment is one data point in a larger, ongoing question I track across every program I run: which behavioral triggers are stage-agnostic — they work wherever there's a bounded task, regardless of the visitor's commitment level — and which are stage-specific, dependent on the visitor already being either committed or uncertain to land at all. Progress indicators, urgency framing, social proof, default options: each has to earn its place at every stage separately, because assuming a category-level pattern like "progress bars work" transfers automatically to a specific stage is exactly the kind of unearned assumption a rigorous experimentation program exists to catch. Sometimes the pattern transfers and sometimes it doesn't, and you only build a reliable model of which is which by testing the boundary every time, not by extrapolating from the first win.
That's the case for treating experimentation as an ongoing capability rather than a series of disconnected wins: the value compounds. This win alone is a lift range on one page. Read alongside the checkout win, it becomes evidence about how the underlying effect behaves across an entire category of multi-step consumer flows — evidence a founder or CMO can actually generalize from when deciding where else in their own funnel the same lever is worth testing.
FAQ
Why not just assume the checkout progress bar would work at plan selection too?
Because the visitor's psychological posture is different at each stage. Checkout visitors have already committed to a plan; plan-selection visitors are still comparing options, sometimes against a competitor's page in another tab. A pattern that reduces friction for a committed buyer isn't automatically safe for someone who hasn't decided yet — it could just as easily read as pressure and push them to leave instead. The only way to know is to test it at the stage where it's meant to run.
Isn't a 5% to 10% lift too vague to act on?
The range reflects a real, statistically supported result. Exact figures aren't published out of respect for confidentiality obligations to past employers and clients, but the range is precise enough to justify scaling the change and precise enough to compare against other wins in the same portfolio. What a founder or CMO needs from a case study isn't a decimal point — it's confidence that the mechanism is real and that the decision to scale was earned, not assumed.
Does this mean progress bars work on every page?
No, and that's the point. This experiment is evidence that the underlying effect transferred across two specific, different stages of one funnel. It isn't evidence that a progress bar is a universal fix. The pattern needs a bounded, sequential task for the mechanism to apply at all — it wouldn't make sense on a single-step form or an open-ended browsing page. Every new stage or context is its own question, answered on its own evidence.
How do you decide which validated patterns are worth re-testing at a new stage?
By checking whether the new stage shares the structural conditions the original mechanism depends on — here, a short, finite, multi-step process — and whether the visitor's psychological posture is similar enough that the same lever should plausibly apply. If both hold, it's worth a fast, low-cost experiment. If the posture is meaningfully different, the way it was here, the experiment is the honest way to answer the question, not a formality before shipping something already assumed to be true.
What does a program like this look like for a company that hasn't run experiments before?
It starts smaller than most people expect — a handful of well-chosen experiments on the highest-traffic, highest-friction pages in a funnel, with enough rigor to trust the results before scaling anything. The bigger unlock isn't any single win. It's building the kind of portfolio where results like this one start telling you something about your customers' psychology in general, not just about one page.
Bottom line
The four-step progress bar won twice — once at checkout, once at plan selection — but the second win wasn't inherited from the first. It came from treating "this pattern worked somewhere else in the funnel" as a hypothesis worth testing rather than a fact worth assuming, and that discipline is what turned one good idea into portfolio-level evidence about how the goal-gradient effect actually behaves across a real, multi-stage customer journey.
If you're deciding whether your team's experimentation program is generating isolated case studies or a compounding model of your customers' behavior, that's the conversation worth having. I design and run experimentation programs for founders and growth leaders who want the second thing — get in touch about working together.
Evidence sources and free next step
Nielsen Norman Group's progress-indicator guidance supplies a usability mechanism, while the second first-party test supplies a comparable production result. Review the checkout page design evidence and experiment diagnostic checklist. Then try GrowthLayer free to link replications without overstating them.