Case study
The Cheapest Plan Was the Wrong Plan
A pricing table turned into a decision system, and realized about $1.8M in potential ARR at the point of sale, without a single dark pattern.
Haim Repael · 4 min read

STAR AT A GLANCE
- Situation
- About 22% of self-serve buyers picked a plan their own traffic had already outgrown on the day they signed up.
- Task
- Move buyers to the plan that fits, at the moment of purchase, without losing conversion and without a dark pattern.
- Action
- Together with Product, turned the catalog into a decision system that measures the buyer's traffic live, shows two plans instead of four, and keeps every plan one click away.
- Result
- Conversion held, plan-fit accuracy roughly doubled, about $1.8M in potential ARR was realized at the point of sale, and the page became the platform standard.
Picture a buyer with 18,000 monthly visits. They open the pricing page, look at four columns and dozens of features, and pick the cheapest one. Its cap is a fraction of their traffic. Nothing on the screen says so. The system let them choose it, so it must be enough.
Now multiply that buyer. About 22% of self-serve purchases went to the entry tier on sites that had already outgrown it. Not months later. On day one.
SituationThe data saw it. The page didn't.
We priced the gap at roughly $1.8M in annual recurring revenue the product was entitled to and never asked for. The money was the easy part to see.
The harder part came after the sale. Overage warnings. Feature limits. Upgrade prompts arriving weeks after the decision. Support tickets. A customer relationship that started with a quiet mistake nobody caught.
The context: a B2B SaaS product, an embedded web component priced in four tiers by monthly site traffic, and a self-serve checkout.
TaskThis was never a pricing table
The task came up in our meetings with Product. We were trying to understand how to fix one specific thing: customers choosing the wrong plan. Redesigning the pricing table was the direction that came out of those meetings. I led the design, together with Product and Ofri, a product designer on my team.
We treated it as a decision problem at the point of sale. The buyer had every reason to choose correctly. The page simply never showed them the one number that mattered.
The usual SaaS answer is enforcement: detect the overage, lock the widget, force the upgrade. It works on a spreadsheet and fails as a relationship. Even when the terms are clear, it feels like bait-and-switch.
So the brief carried three constraints that pulled against each other:
- Fit: move buyers to the plan their traffic actually requires, at the moment of purchase.
- Guardrail: do not reduce overall conversion. A higher average price with fewer buyers is not a win.
- Ethics: no dark patterns. No hidden plans, no pre-checked upsells, no shame copy. The buyer keeps the last word.
ActionDiagnose first, then design
Before sketching anything, we read the old screen against behavioral research. Not to decorate the case study, but to know which failure we were fixing. Each one called for a different move.
- Working memory ran out before the decision did. To choose correctly, a buyer had to leave checkout, open their analytics, average a few months of traffic and come back. Most chose fast to get it over with.
- The lowest price became the anchor and the default. With no data on the page, the cheapest column did the guiding.
- Equal options produced paralysis. The minimal exit from a paralyzing screen is the smallest commitment: the entry tier.
That diagnosis ruled out the obvious fix. Better feature copy would not help a buyer who never had their traffic number in front of them. The missing input was data, not persuasion.
So we stopped treating the page as a catalog and built a decision system that already knew the fact the buyer was missing. Four mechanisms, one for each failure:
- A verification banner that shows its work. The page read the buyer's domain and aggregated its traffic live. A meter filled with a deliberate ease, then revealed the number against the recommended plan: "Your site generates 18,500 visits a month. Fits the Growth plan, which covers up to 30,000." People trust a result more when they watch the effort behind it.
- Two plans instead of four. The plan that fits, marked Recommended, and the one above it. The decision space halved.
- A safety valve. Under the two cards, a plain control: Show all plans. This was the line between a nudge and a dark pattern, and we held it.
- Advice instead of an error. Choose a plan your traffic exceeds and nothing blocks you. A comparison modal lays both plans side by side, and you can still proceed.
The craft carried the mechanism. Two wide cards instead of four dense columns. A 2px accent border on the recommended plan. A cyan-to-blue meter with AAA contrast, because a meter that looked like an alert would have brought back the enforcement posture we were designing away from.
ResultConversion held. Fit doubled.
We tested the new page against the old one in a 50/50 A/B across thousands of self-serve purchases.
| Measure | Result |
|---|---|
| Conversion | Held, non-inferior to control |
| Plan-fit accuracy | Roughly doubled |
| Expansion ARR | About $1.8M potential realized at the point of sale |
| Support contacts on overage, billing and cancellation | Sharp decline |
The feature flag came off. The dynamic page became the standard configuration of the payments platform.
Two honest limits. These are directional results; the per-metric deltas live in the test readout. And the ARR figure is recovered potential at purchase, not a claim about later retention.
What travels
The work here was not a screen. It was changing the unit of decision: from "which plan do you want?" to "here is what your site needs, choose from there." It was deciding, together with Product, not to take the cheaper option of hiding the lower tiers. And it was having product, analytics, data and engineering on one hypothesis before the first build, so the result belongs to everyone.
Nobody was tricked into a bigger plan. The buyer with 18,500 visits can still see all four plans. They just don't have to ask anymore.
