Studies · Usability and decision test · Synthetic research
They trusted the arithmetic, then questioned the verdict. Here is what twelve founder and growth-leader archetypes did with the Retained New Growth Model, and what the results raise for the calculator.
Why we ran it
The Retained New Growth Model multiplies four customer decisions (Act, Buy, Adopt, Renew) to show how much new growth survives, and which decision carries the most leverage. Entering numbers is the easy part. We wanted to know whether a founder or growth leader can recognize their own problem, configure the model, believe the output, find the leverage point, and choose a sensible next step.
We used ten tasks across eight stages, each tied to one question about the page.
Does the archetype see their own problem in Act, Buy, Adopt or Renew?
Do they understand that the metric locates the loss and the decision explains it?
Can they configure the model and enter credible numbers unaided?
Do they understand the outputs, and does the model behave as they predict?
Can they find the highest-leverage decision and tell it apart from the largest loss?
Does the result confirm, challenge or refine their first judgment?
Do they connect the result to the right research pathway?
Is the problem important and credible enough to move toward a conversation?
What we put in front of them
Every archetype worked the Retained New Growth Model on the live page. The numbered outlines mark the parts of it that the findings below keep returning to.
1
2
3
4
5
6
Enter your own numbers, find your lowest rate, and see how the five-point table and the installed-base note respond. Then compare what you notice with the six places we watched.
Who we tested
Every archetype was assigned its own OCEAN personality profile (Openness, Conscientiousness, Extraversion, Agreeableness, Neuroticism). The platform uses that profile to shape how the archetype's reactions shift over a task, so a session does not read as one uniform voice.2 Each card leads with the archetype's primary growth motion and metric decision style; expand it for the rest of the attributes. Cards tagged Highlighted are the six sessions we highlighted, chosen for contrast.
Founder and CEO of a B2B SaaS workflow analytics company
32 · Durham, North Carolina, USA
Solo founder and growth advisor
77 · San Juan, Puerto Rico
Fractional Head of Growth for early-stage B2B SaaS startups
31 · Boise, Idaho, USA
Growth Programs Manager
29 · Dearborn, Michigan, USA
Founder and Head of Revenue Growth
32 · Brno, Czech Republic
VP of Lifecycle Growth
61 · Aarhus, Denmark
Founder and President
55 · Lyon, France
Founder and CEO
79 · Burlington, Vermont, USA
Head of Demand and Retention Growth
38 · Lagos, Nigeria
Growth Analytics Lead
28 · Monterrey, Mexico
Founder and Chief Growth Officer
33 · Austin, Texas, USA
VP of Growth and Customer Revenue
65 · Cebu City, Philippines
How the panel compares with the brief. We recruited for U.S.-based B2B SaaS leaders across every go-to-market motion. Six archetypes sit outside the U.S. Three run a hybrid product-led and sales-led motion that the calculator does not offer as an option. No archetype runs a community-led or event-led motion, so the guidance for those two motions went untested.
How the evidence was produced
Research platform
Each archetype operated the live page in a browser and worked through the ten tasks. The platform produced three kinds of evidence, and each carries its own limit.
Session transcripts. Step-by-step actions with the archetype's reasoning at each step.
Step-level state labels. Each step is tagged with a state such as oriented, positive or hesitant. Synthetic Users describes a "chain-of-feeling" method that ties emotional states to OCEAN traits and models perceived feelings, not physiology.2 We read the labels as markers of where a session slowed, not as measured emotion.
Synthetic eye tracking. For each screen, the platform predicts where the archetype's attention goes. It blends a saliency model trained on recorded human gaze with a persona layer, and for user tests it also draws on what the archetype hovered and clicked. The maps are model predictions, not recorded gaze.1 The platform reports validating them against human eye-tracking data, including a figure of 78% of a real viewer's agreement with a crowd on interfaces, and describes the method in Where do synthetic eyes look?1 That validation is the vendor's own, and we did not reproduce it.
What happened
Nkechi, Helga, Ana Lucía, Petra, Claire and Evelyn were picked for contrast across motion, seniority and geography, so they do not stand in for the whole panel.
Head of Demand and Retention Growth
Used the calculator, then questioned whether its 30% win rate matched her real number.
VP of Lifecycle Growth
Chose sales-led because no hybrid option exists, and pointed to the steepest drop in her funnel.
Growth Analytics Lead
Read the flag as compounding, and followed the note to the Churn Decision Study.
Founder and Head of Revenue Growth
Took the second hero button and ran the Adoption Reality Check's own calculator.
Founder and President
Called the flag a funnel-position artifact.
Founder and CEO
Concluded her win rate was where the money leaked, and said the installed-base view reframed the problem.
Seven findings
Nkechi and Ana Lucía pointed to adoption. Helga and Evelyn pointed to renewal. Claire described adoption bleeding into renewal, and Petra named adoption first with renewal close behind.
named first named second Petra used the Adoption Reality Check's own calculator.
Five of the six used the main calculator. In all five, win rate was the lowest of the four rates they entered, and the five-point table flagged Buy. Helga pointed to the steepest drop in her funnel. Ana Lucía pointed to compounding. Claire called it a funnel-position artifact. Evelyn concluded her win rate was where the money leaked.
The rule behind the flag is that in a multiplied chain, five points added at the lowest rate always moves the most. Only Claire came close to saying so in the sessions. In the follow-up questions, Claire and Evelyn stated it correctly, Helga described it loosely, and Ana Lucía explained Buy's lead by downstream surface area while denying it was because 30% is the lowest rate. Those are the same thing. All four called the flag a place to look, not a diagnosis. Because Buy was the lowest rate every time, this test cannot yet show whether the flag works as well when Adopt or Renew is the weakest decision.
“Buy is still flagged as highest leverage at 30% win rate, adding 2.2 retained customers and $56,100.”
Claire predicted what a change to win rate would do and the model matched. Ana Lucía predicted the ranking would change, was wrong, and treated the surprise as a finding, not a fault. Her later churn-page prediction held. Helga called the funnel logic clean. Helga and Evelyn changed inputs without stating a prediction first, so the test step was only partly observable.
The note beneath the table says a different model applies when churn hits revenue already booked. Helga entered an $18M book and saw five points of retention worth $900,000, 22.2 times the new-cohort effect. Claire saw 13.2 times and Evelyn 5.4 times. Helga and Ana Lucía followed the note's link to the Churn Decision Study. Evelyn said it reframed the conversation, moving retention from secondary to the larger lever.
Five of the six archetypes clicked the primary hero button, which lands on the calculator below the four decision cards. Petra took the second button, matched her problem to the Adopt card, and went straight to the Adoption Reality Check, where she ran that page's own calculator. Of the five calculator-first archetypes, only Ana Lucía reviewed all four research pathways. Four of the five said the Renew card or example was missing, because the page's example scenarios cover Act, Buy and Adopt only.
Four archetypes changed the go-to-market motion, including Helga, who chose sales-led because no hybrid option exists. In the sessions we read, none opened a "How to estimate this" panel, and none changed the average new-customer ARR, so every Part 1 dollar figure rests on the $25,000 default. Part 2 was used by three of the five calculator sessions. Claire and Evelyn cited its break-even figure (2.0 and 3.6 points against a $10,000 study) to support booking a conversation, and Nkechi said the break-even math checked out as she chose a time slot.
Five archetypes chose a slot, and the four whose sessions we read to the end stopped at the name and email form. Ana Lucía reviewed the page without choosing one. That shows the task was completed, not that anyone would buy. In the five sessions where state labels were available2, the four that recorded a hesitant state (Helga, Ana Lucía, Petra and Claire) recorded it at booking. Helga, Petra and Claire, all in Europe, saw midnight, 12:30am, 1am and 8pm slots. Evelyn saw a mid-morning slot and recorded no hesitation. Ana Lucía, Petra and Claire noticed the page says both 30 and 45 minutes, and Ana Lucía paused over the "sprint" branding on a call she reached from an adoption study button.
“There's a cookie banner blocking part of the screen. I need to dismiss it first.”
Follow-up questions
After their sessions we asked Claire, Helga, Ana Lucía and Evelyn the same questions again, as follow-ups to the same synthetic archetypes. They are answers to hypotheticals from archetypes that remember their session, so we read them as hypotheses and not as new evidence.
| What we asked | Claire | Helga | Ana Lucía | Evelyn |
|---|---|---|---|---|
| Called the flag a place to look, not a diagnosis | Yes | Yes | Yes | Yes |
| Instinct stayed on their own decision | Adopt | Renew | Adopt | Renew |
| Said five points are not equally achievable | Yes | Yes | No | No |
| Stated the leverage rule correctly | Yes | Partly | No | Yes |
| Would have reached the installed-base number sooner | Yes | Yes | Yes | Yes |
| Reran both scenarios with values shown | No | No | No | Not shown |
“So the math earned more trust, the conclusion earned the same amount of scrutiny it had at the start.”
“The table treats all five-point improvements as equally achievable, and they are not.”
“When the model surfaces something I might have missed, I trust it more than when it confirms what I already see clearly.”
Two answers we did not use. Three archetypes reasoned about the Adopt-lowest and Renew-lowest scenarios without rerunning them, and Evelyn reported a rerun without values, so we report no scenario results. Some answers about what they expected before changing an input were reconstructed after the fact, and the sessions show no stated prediction.
Synthetic eye tracking
Nkechi's session gives the clearest view, with twelve of her 24 screens available to us. Seven of them appear below, each with what she said aloud at that step. Predicted attention was heaviest on the hero headline and primary button, on the motion and period dropdowns, on the $420,750 output, on the Buy row and the summary sentence beneath the table, on the first and third qualification cards, and on the closing call to action. It stayed light on the secondary hero button and on the FAQ card comparing this research with win/loss analysis. Each pattern fits what she said at that step.
These maps are model predictions from Synthetic Users, and in user tests they draw partly on hover and click behavior, so hot spots near the cursor are expected.1 We read them as hypotheses about where attention settled once an archetype was working, not as a record of what they noticed first. We have not analyzed maps for the other five sessions.
“The calculator is the task, so starting there.”
“The Adopt one jumps out first because ‘customers buy but do not change behavior’ is exactly what I deal with.”
“GTM motion is now ‘Content-led marketing’ and measurement period is ‘Trailing 12 months.’ Both look right for our setup.”
“Buy is flagged as highest leverage at 30% win rate, adding 2.8 customers and $70,125 from a five-point improvement.”
“The four checkmarks are all lit up, which means the study qualifies.”
“Going to click ‘DISCUSS THE DECISION.’”
“The 3:00pm slot looks fine.”
What the results raise
We have not yet evaluated these findings to decide what to change. Each card pairs an observation with the question it puts to the page, a preliminary priority, what to watch for in live tests, and the method that would show it: a moderated session, a session recording, a click or scroll map, or a counted event. The order reflects how many sessions showed the pattern and how much it could change what visitors do, and it may move as live data comes in.
It appears in the text below it. In the follow-ups, all four archetypes called the flag a place to look and not a diagnosis, but only two stated the rule correctly, and Ana Lucía asked for the chain to be shown beside the table.
All four archetypes said they would have reached it sooner, and the result only appears after an ARR book is typed. Ana Lucía's own figures show it will not always favor retention ($100,000 on the book against $114,750 from Buy), so it should be shown neutrally.
Buy was the lowest rate in every session, so a follow-up with adversarial inputs would show whether the other three flags land as well. The follow-up reruns were not verified, so this is still untested.
Five of six took the calculator first, and three never reached a pathway page.
Claire and Helga said so unprompted: the table treats all five-point improvements as equally achievable, and they are not. Moving a partner-led win rate is harder than fixing onboarding.
Four of five calculator-first archetypes said so.
Nobody opened it. Evelyn asked for a prompt to triangulate an uncertain win rate and wondered whether her 45% was "a real number or a hopeful one," which is what the toggles are for.
Nobody changed ARR per customer, so every Part 1 dollar figure used $25,000. Ana Lucía read the greyed 2000000 in the ARR book field as already filled in, so she never saw the book result and worked it out in her head.
Three of the twelve archetypes run one, and Helga picked a motion to work around it.
Half the panel sits outside the U.S., and the page and the booking page give different meeting lengths.
Limits
Use these results to decide what to fix and what to test next with real buyers. Do not use them to predict conversion.
Sources
Customer decision review
Try the calculator with your own numbers, then tell us which decision it flags and whether you agree.