Most "ideation workshops" fail in the first ten minutes, and they fail the same way: the most senior or loudest person says what they think the answer is, and the room quietly converges on it before a single alternative has been drawn. The techniques below exist to stop exactly that. They are not interchangeable team-building games. Each one is tuned for a specific job, diverging to more options, or converging a mound of raw input into something you can act on, and using the wrong one is why so many sessions produce sticky notes and no decisions.

The quick version

  • Diverge before you converge. Crazy 8s and design studios force a wide spread of options on the table before anyone judges, so the first idea doesn't win by default.
  • Affinity mapping converges. It turns a wall of research notes or ideas into a handful of themes the group can actually reason about.
  • Card sorting is for structure, not ideas. It reveals how real users would group and label your content, the menu, not the message.
  • The method is the cheap part. A clear question, the right people, and silent independent work first are what make any of them work.

The idea in depth: diverge, then converge

The spine that ties these four techniques together is older than any of them: separate generating options from choosing between them. Crowd everything into one open discussion and you get anchoring, the room latches onto the first plausible idea and spends its energy defending it. So nearly every good ideation method builds in a wall between the two phases.

Crazy 8s is the sharpest expression of that wall. The exercise comes from Jake Knapp's design-sprint work at Google and Google Ventures, written up in Sprint (Knapp, Zeratsky & Kowitz, Simon & Schuster, 2016): everyone folds a sheet of paper into eight panels and sketches eight distinct ideas, one per panel, against a tight clock, Google's own Design Sprint Kit runs it at eight minutes, roughly a minute each. The point of the time pressure isn't speed for its own sake. It's that your first idea is usually your most obvious one, and the constraint drags you past it into the variations you'd never have volunteered out loud. So when a team keeps circling one solution, don't debate it: hand out paper and a timer and demand eight alternatives, in silence, before anyone speaks.

The design studio wraps that same sketch-in-silence step in a full diverge-and-converge loop. Nielsen Norman Group describes it as "a type of UX workshop that combines divergent and convergent thinking," and lays out the cycle plainly: sketch individually, then present and critique, then converge by combining the strongest elements, then prioritise (Kate Kaplan, "Facilitating an Effective Design Studio Workshop," NN/g, 2017). The method grew out of practice in architecture and industrial design and was carried into UX largely by practitioners such as Will Evans and Todd Zaki Warfel. Use it when a cross-functional group needs a shared direction and not just one person's vision: have the engineers, marketers and designers each sketch, critique each other's work, and merge. The decision then carries the fingerprints of everyone who has to build it.

flowchart LR
    A(["Frame one clear question"]) --> B(["Diverge: sketch in silence
Crazy 8s · individual ideas"])
    B --> C(["Present & critique
no defending, just listening"])
    C --> D(["Converge: combine the
strongest elements"])
    D --> E(["Prioritise & decide
what to prototype next"])
The diverge-converge loop a design studio runs, generate wide, then narrow on purpose. Leaders Loop

Where the sketching methods diverge, affinity mapping converges. You take a wall of raw input, research observations, interview quotes, half-formed ideas, each on its own note, and cluster them by natural relationship until themes emerge from the bottom up. The technique has a real lineage: it's the KJ method, devised by the Japanese anthropologist Jiro Kawakita in the 1960s to make sense of unwieldy ethnographic field data, and later adopted into the "seven management and planning tools" used in Japanese quality practice (see the affinity diagram and its KJ-method roots). The discipline that makes it work is grouping before you name the groups, let the clusters form, then label them, so the categories come from the data rather than from your prior assumptions. After any round of customer interviews, or staring at a flip-chart of fifty ideas, resist the urge to summarise from memory. Put each point on its own card, cluster silently, and let three or four themes surface that you can actually prioritise.

Card sorting is the odd one out, and the one most often misused. It is not for generating ideas at all, it's for structure. You give participants a set of labelled cards (your content, features or topics) and ask them to group them in the way that makes sense to them. NN/g defines it as "a research method in which study participants place individually labeled cards into groups according to criteria that make the most sense to them" (Tankala & Sherwin, "Card Sorting," NN/g, 2024). An open sort lets people create and name their own categories, good when you're building an information architecture from scratch; a closed sort gives them your existing categories to file cards into, good for testing a structure you already have (NN/g, "Open vs. Closed Card Sorting"). When users can't find things in your product, the fix isn't to redraw the menu in a meeting. Run a card sort with real users and let their mental model dictate the navigation.

Where this breaks down

An honest limitation runs through all of it: these methods organise thinking, they don't supply it. A design studio of the wrong six people produces six confident sketches of the wrong product. Affinity mapping can manufacture false consensus, clusters look authoritative on a wall even when they rest on three weak data points, so a tidy map is not a true one. Card sorting tells you how people would group content but not whether they'd complete a task, which is why it pairs with, rather than replaces, usability testing. The technique is the cheap, reliable part. The expensive part, a sharp question and the right people in the room, is still on you.

flowchart TD
    Q(["What's the job
in front of you?"])
    Q --> G(["Need MORE options?"])
    Q --> S(["Need to make SENSE
of messy input?"])
    Q --> N(["Need the right
STRUCTURE / labels?"])
    G --> G1(["Crazy 8s · Design studio
diverge by sketching"])
    S --> S1(["Affinity mapping (KJ)
cluster, then name"])
    N --> N1(["Card sorting
open or closed"])
Pick by job-to-be-done, not by which method is fashionable this quarter. Leaders Loop

A worked example

Picture a small payments team whose support queue is clogged with the same complaint: people can't find how to dispute a transaction. The product lead's instinct is to add a big "Dispute" button to the home screen. Instead, the team runs the loop. (Figures below are illustrative.)

Monday, they card-sort the twenty help topics with ten real customers, an open sort. The results are unambiguous: nobody files "dispute a charge" under "Payments," where the team had put it; they look under "Problems" or "Security," a category that didn't exist in the product. That's a structure finding, and it would have been invisible in a meeting.

Wednesday, with the where settled, they tackle the how. Six people, two engineers, a designer, a support lead, the PM and a data analyst, run a design studio. Eight minutes of Crazy 8s each produces roughly forty rough sketches; the support lead's panel, the one person who reads the complaints daily, contains the idea no designer drew: let users start a dispute straight from the transaction line, not a separate flow. In the critique-and-converge step, that becomes the spine of the design.

Thursday, they affinity-map the forty sketches plus the support tickets. Three themes cluster out, find it fast, explain what happens next, reassure me it's handled, and those become the three things the new flow must do. No single technique solved the problem. The card sort fixed the location, the studio widened the solutions, and the affinity map turned the pile into three testable bets. That sequencing, structure, then options, then sense-making, is the reusable part.

Frequently asked questions

Isn't this just brainstorming with extra steps?

Classic open brainstorming has a well-known weakness: people anchor on the first ideas and the confident voices dominate, so the group generates fewer distinct ideas than the same people would working alone first. These methods all share one fix, independent, silent generation before any discussion. That's the "extra step," and it's the entire point. Skip it and you've just run a meeting.

How many people, and do they need to be designers?

For a design studio or Crazy 8s, four to eight is the practical band, enough variety, few enough that everyone presents. None of them need to draw well; boxes, arrows and stick figures are fine, and saying so out loud is part of the facilitator's job. For card sorting, NN/g's guidance is to recruit actual users of the product (around 15 participants gives reasonably stable patterns), not the internal team, the whole value is in an outside mental model.

When do I use card sorting versus affinity mapping?

They look similar, both involve grouping cards, but answer different questions. Card sorting asks users how your content should be structured (an input to information architecture). Affinity mapping is your team making sense of raw material you've already gathered (an output of research). One designs the menu; the other digests the findings.

What if the most senior person hijacks the room anyway?

This is the failure these methods are built to resist, so lean on their mechanics rather than your authority. Make the first round silent and individual, collect every sketch before anyone speaks, and present them anonymously or in a random order. When the boss's idea competes on the wall as one option among eight, its gravitational pull drops sharply.

Do these work remotely?

Yes, and arguably better in one respect: a shared online whiteboard makes truly parallel, simultaneous work easy, and timed silent rounds enforce themselves. The thing to protect remotely is the silence, it's tempting to fill a quiet video call with chatter, which collapses the independent-generation step that does the heavy lifting.

Related in the Toolkit

Where to go next