Case Study — SaaS MVP · User Research · Product Strategy

SaaS MVP Research

Research isn't decoration. Hypothesis-driven validation that changed what a startup believed about their own users — before they built the wrong product.

My roleLead UX Researcher
Year2024
ClientTech Leap — Wingman AI
PlatformB2B SaaS (pre-build)
DeliverablesResearch report · Validated personas · Repositioned MVP brief
Wingman AI SaaS MVP — AI-assisted sales tool interface

A team ready to build, and a product no one had tested

Tech Leap had spent four months developing the concept for Wingman AI — an AI-powered coaching tool for B2B sales reps. The product vision was clear: real-time suggestions surfaced during live sales calls, helping reps handle objections, stay on script, and close faster.

The founding team was experienced. They had spoken to a handful of sales managers who were enthusiastic. Engineering was scoped and ready to sprint. And then someone asked the question that should have come first: have we actually talked to the people who would use this every day?

The answer was no. They had talked to buyers — the managers who would approve the purchase — but not to the sales reps who would be staring at a screen while on a call with a prospect. I was brought in to run four weeks of discovery research before a single line of product code was written.

Wingman AI — product dashboard interface and feature overview

Sole researcher. Direct line to the founding team

I led all research activities end to end: study design, participant recruitment, moderated interviews, synthesis, and the final strategic brief that redirected the team's product thinking.

I reported directly to the co-founders and presented findings in three working sessions over the four-week engagement. This was not research-for-the-sake-of-a-report — every finding connected explicitly to a decision the team was about to make. The framing I used throughout: what do we need to believe to be true for this product to work, and is it true?

Six assumptions. Two of them right

Before any interviews, I sat with the founding team to surface every assumption baked into their product concept. We listed them explicitly and scored each one on two axes: how central is it to the product concept, and how confident are we that it's true?

The riskiest assumptions — central to the product but unvalidated — became the research questions. We entered the field with six specific hypotheses to test.

Hypothesis Confidence going in Result
Sales reps want real-time suggestions during live calls High Invalidated
Pre-call preparation is a key pain point for reps Low Confirmed
Junior reps are the primary user segment High Invalidated
Reps want visibility into their own performance data High Partially true
Sales managers will champion adoption High Confirmed
CRM integration is a prerequisite for purchase Medium Partially true

Four of six central hypotheses were wrong or more complicated than assumed. The product that would have been built based on those assumptions would have served the wrong moment, for the wrong person, with the wrong feature prioritisation.

Structured to produce decisions, not insights

The research was designed from the output backwards: what decisions need to be made by the team at the end of week four? I mapped these decisions, then designed a mixed-methods study to answer them.

🎙️
Moderated interviews
18 sales reps across 6 companies. 45 minutes each. Mix of SDRs, AEs, and senior account managers. Recruited independently — not through the client's network.
👀
Contextual observation
Shadow sessions with 4 reps during real sales call prep and debrief. Observed what they actually did, not what they said they did.
📊
Competitive diary study
8 participants logged tool usage over 2 weeks. Revealed actual vs. reported tooling behaviour — significant gap.
🧩
Jobs-to-be-done mapping
Synthesised interview data into functional, emotional, and social jobs. Used to test fit of the proposed product concept directly against real needs.
Research synthesis — affinity mapping and pattern identification across user interviews

The product team didn't know their user

"I don't want suggestions popping up while I'm listening to a prospect. That would throw me completely. What I need is to walk into the call already knowing what to expect."
— Senior AE, B2B SaaS company, interview participant

Real-time in-call suggestions — the centrepiece of the product — was rejected by 14 of 18 reps. Not because the technology seemed far-fetched, but because the moment was wrong. A call is not a moment for AI assistance; it's a performance. The cognitive load of processing a suggestion while listening to a prospect and formulating a response was described, unprompted, by multiple participants as "exactly the kind of distraction that kills deals."

The real pain was upstream. Pre-call preparation — researching the prospect, reviewing previous interactions, predicting likely objections — was something every rep did, and almost everyone found frustrating. It was time-consuming, inconsistent, and scattered across four or five tools.

The secondary finding was about who the user actually was. The founding team had designed for junior reps who needed coaching. But junior reps were the least motivated to use a new tool — they followed process and hoped to get out of the role quickly. The power users were mid-level reps who had already developed an instinct for selling and wanted to sharpen it. They were more self-directed, more tool-aware, and more likely to champion adoption bottom-up.

78%
Rejected real-time suggestions
14 of 18 reps said in-call AI prompts would be distracting or harmful to their performance
94%
Struggled with pre-call prep
Nearly every participant named inconsistent or time-consuming prep as a genuine pain point
2.4×
More tools than reported
Diary study found reps used more than twice as many tools as they mentioned in interviews
#1
Trust barrier — privacy
"Will my manager see my scores?" — raised spontaneously by 11 of 18 reps, never anticipated by the team

The privacy finding was the one that most surprised the founding team. Performance data was assumed to be a feature. But for most reps, sharing call quality scores with management was a threat. If the product surfaced individual performance data to managers by default, adoption would collapse. The product needed to be clearly rep-controlled before it could earn trust.

Key research findings — user needs and product pivot recommendations

Three pivots the research made undeniable

Pivot 1 — Move the value from the call to before the call
The MVP would be redesigned around pre-call intelligence: automatic research briefs, prospect history summaries, and predicted objection maps — delivered to the rep 30 minutes before a scheduled call. The in-call assistant feature would move to a later roadmap phase, if at all.
Why this was hard: The in-call feature was the team's vision. It was what made the product feel like "AI" to them. Pivoting away from it required accepting that the most technically impressive capability was not the most useful one. The research made this argument impossible to dismiss — it wasn't one user's opinion, it was 14 out of 18.
Pivot 2 — Make individual data private by default
Performance data — call quality scores, objection handling rates, AI coaching suggestions — would be owned by the rep, not the manager. Managers would only see aggregated, anonymised team data unless a rep explicitly chose to share. The product would market this privacy posture explicitly.
Why this was hard: The business model assumed managers were the buyer and needed individual visibility to justify the spend. The research showed that making performance data visible to managers by default would kill rep adoption, and without rep adoption there was no product. The team needed to find a pricing model that worked with bottom-up adoption.
Pivot 3 — Redefine the primary user as the ambitious mid-level rep
All copy, onboarding, feature language, and in-product feedback loops would be designed for reps who already had instincts and wanted to sharpen them — not for beginners who needed to be told what to do. This changed tone, changed the default complexity level, and changed what a "good session" looked like.
Why this was hard: Junior reps are a larger segment and often the ones managers point to when justifying coaching tools. Choosing not to optimise for them meant walking away from the most obvious sales pitch. But research showed that optimising for juniors would produce a product that felt patronising to the people most likely to actually use it.

Three months of engineering time redirected before it was spent

~3
Months of engineering saved
6→3
Hypotheses validated before build
100%
Core assumptions revised
1
Repositioned MVP brief delivered

The founding team left the engagement with a repositioned MVP brief, two revised personas, a new understanding of their primary user's privacy concerns, and a roadmap in which the most technically complex feature had moved from phase one to phase three.

The research also produced an unexpected commercial output: by framing Wingman AI as "built for reps, not for managers," the team identified a positioning angle that differentiated them from every competitor in the space — all of whom sold top-down to management. This became their go-to-market angle for early trials.

Original product scope — before research-driven repositioning
Repositioned product scope — post-research MVP definition

The most valuable thing research can do is stop the wrong work

This project didn't produce a design. It produced a change in what the team was going to design. That is harder to package as a portfolio piece than a screen or a prototype — but it is often where UX creates the most leverage.

The challenge in research-only engagements is maintaining credibility when you are the person telling a founding team that their core assumption is wrong. The table of hypotheses — entered before the field, scored honestly afterwards — was what made the findings feel rigorous rather than opinionated. The team couldn't argue with a framework they had helped create.

If I were running this study again, I would add a concept test earlier — a lo-fi prototype of both the original and the repositioned concept, shown to the same 18 participants — to produce a direct comparison the team could see rather than read. The research was strong. The evidence would have been stronger.

Back to the beginning

PlutoPay Financial App