Latest / 5 Minute UX / Revisiting Assumptions After Research
Transcript
- 0:07This lesson is Revisiting Assumptions After Research.
- 0:10Let's get started.
- 0:18The first move is suspension. Take the original assumptions
- 0:20and any provisional user models created before research began
- 0:23and set them aside. Do not attempt to force new data into
- 0:25the old framework. When you try to fit fresh evidence into a
- 0:27pre-existing structure, you distort the findings. The goal
- 0:30is to clear the mental space so the data can speak for itself.
- 0:32Next, reconstruct the user groups from the evidence. Ask
- 0:35yourself, "What user groups would I create now that I have
- 0:37more information?" This question forces a fresh look at the
- 0:39patterns in the data. Use the new knowledge to define the
- 0:42groups that should actually be the focus of the project. This
- 0:44revised definition replaces the initial guesswork with grounded
- 0:47reality. This shift changes the target of your design work. You
- 0:50are no longer designing for who you thought the users were, but
- 0:52for who they actually are. The next step involves comparing
- 0:55these newly defined groups against the participants you
- 0:57actually researched to identify any coverage gaps.
- 1:05You have already set aside the old assumptions and defined the
- 1:07new user groups based on your latest evidence. Now you need to
- 1:10check if your research actually covers them. The critical step
- 1:12is to compare the newly defined groups against the participants
- 1:15you have actually spoken with. This comparison reveals whether
- 1:18any earlier assumptions were invalid. Often, the data shows
- 1:20that a key user group was completely excluded from the
- 1:23research. This happens because that group was not identified in
- 1:25the initial assumptions. If you only spoke to power users, you
- 1:28might have missed the casual users who struggle with basic
- 1:30navigation. That gap represents a blind spot in your current
- 1:33understanding. If you identify this gap early enough in the
- 1:35research activity, you can still adjust the plan. The move is to
- 1:38add another set of participants to the research in progress.
- 1:40This expansion ensures you get a full picture of the user base
- 1:43before you move forward. You are not starting over; you are
- 1:46simply widening the net to capture the missing perspective.
- 1:48This adjustment is only possible if time permits. If the research
- 1:51phase is winding down, you may not have the budget to recruit
- 1:53new participants. In that case, you must acknowledge the
- 1:56limitation and note that specific group for future
- 1:58validation. The goal is to match your research scope to your
- 2:00actual user groups, not just the ones you thought existed at the
- 2:03start. By applying this gap analysis, you determine if your
- 2:05research scope needs adjustment. This ensures your final insights
- 2:08are grounded in the reality of your users, not just your
- 2:11initial guesses. Once you have this validated coverage, you can
- 2:13move to revising the user model itself.
- 2:20You’ve just compared your newly defined groups against the
- 2:22participants you actually spoke with, and now you need to close
- 2:25the loop. Step five is where you update the user model to
- 2:27accurately reflect the groups identified in Step two. This
- 2:30isn’t a minor tweak; it’s a structural revision that aligns
- 2:32your working assumptions with the evidence you’ve gathered.
- 2:35Think about the specific attributes you listed for your
- 2:37initial personas. Did the research confirm those
- 2:39motivations, or did it reveal a completely different primary
- 2:42driver? If the data shows a shift, you have to rewrite that
- 2:44section of the model. This revised model becomes the basis
- 2:47for creating more detailed tools, such as personas. When
- 2:49you build those detailed artifacts, you are no longer
- 2:51guessing; you are translating validated insights into concrete
- 2:54design constraints. Take a moment to look at your current
- 2:57model. Which specific user group changed the most based on what
- 2:59you learned? If you can’t point to a specific shift, you haven’t
- 3:02fully integrated the new data. Once that model is accurate, you
- 3:05use it to generate new ideas for exploration. The model should
- 3:07feel different now, right? If it doesn’t, you’re still working
- 3:10from the old assumptions. The revised model is the foundation
- 3:13for your next moves, but it also sets the stage for how you
- 3:15handle projects where you can’t speak to users directly.
- 3:24There is a specific moment in real work where direct user
- 3:27contact simply isn't possible. In that situation, using an
- 3:29unvalidated model introduces a higher risk that your
- 3:32assumptions will be incorrect. This isn't a failure of the
- 3:34process; it's an inherent constraint of the project's
- 3:36scope. You still need a model to move forward, because a well-
- 3:39thought-out model is always preferable to having no model at
- 3:42all. Your mitigation strategy shifts from direct observation
- 3:44to external verification. You need to check your model against
- 3:47sources outside the project team, specifically customer
- 3:49service groups or training groups. These teams interact
- 3:51with your actual users daily, so their insights provide a
- 3:54critical check on the assumptions embedded in your
- 3:56provisional model. This step grounds your work in reality,
- 3:59even when you can't sit across the table from the user.
- 4:01Finally, you must handle the team communication with
- 4:03precision. Be explicit with your team that these models are based
- 4:06on assumptions and require refinement if research becomes
- 4:08possible later. This clarity prevents the model from being
- 4:11mistaken for validated truth. It sets the expectation that the
- 4:13current state is a working hypothesis, not a final
- 4:16conclusion. When you finally get access to direct user feedback,
- 4:18you know exactly what needs to be revisited and corrected.
- 4:24Welcome to The Take. I'm joined by a fellow practitioner and
- 4:27together we're going to dig into Revisiting Assumptions After
- 4:29Research — what it means in practice and what's worth
- 4:32remembering.
- 4:33[HOST] There's a specific moment in real work where direct user
- 4:37contact simply isn't possible. [LEARNER] That constraint feels
- 4:40like it forces a false sense of security. [HOST] It does. We
- 4:43often treat the model as a static artifact rather than a
- 4:46hypothesis. [LEARNER] But if we can't validate it directly, how
- 4:49do we know the assumptions are even close? [HOST] You check it
- 4:53against customer service logs or training feedback instead.
- 4:56[LEARNER] That’s a good proxy, but it’s still second-hand data.
- 4:59[HOST] Exactly. It’s a mitigation, not a validation.
- 5:02[LEARNER] Which brings me back to the suspension step. I find
- 5:05it hard to truly let go of the original persona. [HOST] Why is
- 5:09that so difficult for you specifically? [LEARNER] Because
- 5:12the initial model often drives the project scope and budget.
- 5:15[HOST] So the model becomes a political asset before it’s even
- 5:18tested. [LEARNER] Precisely. Changing it feels like admitting
- 5:21the initial planning was flawed. [HOST] But if you don’t suspend
- 5:25it, you distort the new evidence to fit the old box. [LEARNER] I
- 5:28see that now. The distortion is the real danger, not the change
- 5:32itself. [HOST] And that’s where AI tools can accidentally make
- 5:35this worse. [LEARNER] How so? They usually help us organize
- 5:38data faster. [HOST] They can generate plausible personas from
- 5:41your old assumptions with high confidence. [LEARNER] So they
- 5:45might hallucinate validation for the very biases we’re trying to
- 5:48suspend. [HOST] Yes. The fluency of the output masks the lack of
- 5:51grounding. [LEARNER] That’s a subtle trap. The model looks
- 5:54polished, so we trust it more. [HOST] You have to actively
- 5:58interrogate the source of the inputs. [LEARNER] So the tool
- 6:01doesn’t remove the need for critical thinking, it just
- 6:04automates the risk. [HOST] It amplifies the speed of error if
- 6:07your foundations are weak. [LEARNER] I’ll need to be more
- 6:10careful about labeling those models as provisional. [HOST]
- 6:13That transparency is the only safeguard you have. [LEARNER]
- 6:16Agreed. I’ll start tagging every assumption explicitly in the
- 6:19next sprint. [HOST] That’s the right move. It keeps the model
- 6:23honest. [LEARNER] And it helps the team understand what still
- 6:26needs validation. [HOST] Exactly. Clarity prevents the
- 6:29model from calcifying into truth. [LEARNER] I appreciate
- 6:32that distinction. It changes how I’ll brief the stakeholders.
- 6:35[HOST] It should. The goal is to keep the model a working
- 6:38hypothesis. [LEARNER] Not a final conclusion. [HOST] Right.
- 6:41And that flexibility is what allows the research to actually
- 6:44guide the design. [LEARNER] I think I’m ready to try that
- 6:47suspension step again. [HOST] Good. Let the data speak for
- 6:51itself this time. [LEARNER] I’ll make sure to leave the old
- 6:54persona in a drawer. [HOST] Metaphorically speaking, yes.
- 6:57But keep it accessible for comparison. [LEARNER] Smart. I
- 7:00can see exactly what shifted. [HOST] That visibility is what
- 7:03turns a guess into an insight. [LEARNER] Thanks for pushing on
- 7:06that point. [HOST] It’s a common blind spot. We’re glad you
- 7:09caught it. [LEARNER] I feel more confident about the next phase
- 7:13now. [HOST] That confidence should be tempered, not
- 7:16eliminated. [LEARNER] Understood. Humble confidence.
- 7:18[HOST] Exactly that. Let’s see what the new data reveals.
- 7:21[LEARNER] I’ll bring those findings to our next sync.
- 7:24[HOST] Looking forward to seeing the revised model. [LEARNER]
- 7:27It'll definitely look different. [HOST] It should. That’s the
- 7:31whole point of the exercise. [LEARNER] I’ll let you know how
- 7:34it goes. [HOST] Keep me posted. And trust the process. [LEARNER]
- 7:37I will. Thanks again. [HOST] Any time. Go revise that model.