How to run agile t-shirt sizing sessions: anchoring, simultaneous reveal, mapping to story points, and when to use it instead of planning poker.

Your team has 30 backlog items to size, a 60-minute window and a roadmap decision riding on the outcome. Debating whether a story is a 5 or an 8 doesn't fit into that. Agile t-shirt sizing is a faster way to reach a shared view of effort, and it is more accurate than its simplicity suggests.
What agile t-shirt sizing is
T-shirt sizing gives each work item a size label (XS, S, M, L, XL, sometimes XXL) instead of an estimate in hours or days. The group captures effort, complexity and uncertainty together. Nobody should read the result as a literal forecast.
The technique rests on a simple observation: people are much better at comparing things with each other than at judging absolute values in isolation. Atlassian's agile estimation guide makes a related point: using story points instead of hours encourages team collaboration, reduces emotional bias and improves forecasting. "Is this a Small or a Medium?" is faster to answer, and less error-prone, than "how many hours will this take?"
T-shirt sizing suits early backlog grooming, roadmap planning and PI Planning in SAFe contexts, where epics and features need a quick size before anyone breaks them down. A product team that sizes this way during quarterly roadmap reviews can get through a whole quarter of work in one session, because each item takes a couple of minutes instead of the longer discussion that planning poker invites. With 30 or more items on the table, that difference adds up quickly.
How to run a t-shirt sizing session
A session has five phases.
- Preparation: the facilitator checks that every backlog item has enough context for a useful discussion. If a story has a one-line title and no acceptance criteria, the team will spend 10 minutes guessing what it is before sizing can start. Make the items readable before the session.
- Anchoring: before voting, the team agrees on a reference story for each size. Teams skip this step more than any other. Without a shared reference, "Medium" means something different to every participant. Pick one completed story per size that everyone accepts, and keep those examples visible for the whole session, on a Miro board, a Confluence page or a physical card. They work like a ruler, and they decide whether the team stays calibrated or starts from scratch every session.
- Silent simultaneous reveal: everyone writes down a size privately and all reveal at once. Nobody speaks before the reveal, so the first voice in the room can't anchor the others.
- Discussion: discuss only the outliers. If everyone says M and one person says XL, that conversation is worth having. If the spread is M and L, pick one and move on. Sessions fall apart when teams debate every small gap in full.
- Consensus record: enter the agreed size in the backlog tool before moving to the next item.
For remote teams, Miro's agile estimation templates and similar digital boards make simultaneous reveal possible with a single "flip all" action. The tool has to enforce the rule. If participants type answers into a visible shared doc, they read each other's responses before committing, and the bias you wanted to avoid is back.
If you pin the reference stories to a shared Confluence page, new team members can learn what "Medium" means in your context without a full calibration exercise every time the team changes. It is a cheap way to keep sizes consistent over time.
Mapping t-shirt sizes to story points
Mapping is optional. It helps when your delivery tooling (Jira, Azure DevOps, Linear) tracks velocity in story points, but the technique doesn't need it.
A common mapping is XS=1, S=2, M=3, L=5, XL=8, which follows the Fibonacci sequence used in planning poker. Some teams use XS=1, S=2, M=5, L=8, XL=13 to put more distance between large items and force clearer distinctions when the backlog leans toward complex work.
Whichever mapping you pick, treat it as a team convention and keep it for at least a quarter. Changing it mid-project is like changing your unit of measurement halfway through. Anyone on the team who equates story points with time should read Scrum.org's myth-busting piece on story points.
Picture a product team that maps XS=1, S=2, M=3, L=5, XL=8 in Jira, tracks velocity for two quarters and notices that XL stories land far less predictably than M stories. A sensible response is a team rule that every XL item gets split before it enters a sprint. Sizing then doubles as a trigger for story splitting.
Mature teams that track flow metrics such as cycle time and throughput instead of velocity can use t-shirt sizes as a rough prioritisation filter without converting them to points.
T-shirt sizing vs planning poker vs dot voting
People mix these three techniques up, but they do different jobs.
T-shirt sizing is fast and scales well. Use it when you have 20 or more items, little time or little detail at story level. Planning poker is slower and more precise: the Fibonacci scale makes participants choose between a 5 and an 8, which brings out more nuance than "Medium vs Large". Dot voting is a prioritisation method. It helps a group agree on which items to tackle, and it says nothing about how large they are.
A rule many agile coaches use: t-shirt sizing for anything above sprint level (epics, roadmap features, program-level objectives) and planning poker for user stories being refined for the next sprint. T-shirt sizing works as the coarse first pass, and planning poker's precision goes where it pays off.
In client discovery phases this often means t-shirt sizing covers whole feature sets quickly, and planning poker takes over for story-level refinement once development starts. The two techniques complement each other.
As for where dot voting fits: dot voting tells you what to work on, and t-shirt sizing tells you how large the work is. They answer different questions.
Common failure modes
Anchoring bias is the most common failure. Whoever speaks first sways the group. Simultaneous reveal is what stops the session from simply confirming the tech lead's opinion, so don't treat it as optional.
The second is size creep. Over time, teams quietly stretch their definition of Small until something that used to be a Medium gets called Small. Recalibrating monthly or quarterly against the reference stories fixes this. Without it, your velocity data drifts in ways that are hard to diagnose.
When stakeholders treat sizes as commitments, that is a problem of facilitation and stakeholder management. Sizes are rough order-of-magnitude estimates with wide uncertainty. Wording helps: "our best shared guess today" sets a different expectation than "our estimate".
The last failure is endless debate. Teams try to agree on an exact size for every story and turn a 2-minute exercise into a 15-minute argument. If the team can't agree within 3-4 minutes, take the larger size and flag the story for breakdown. A strict 3-minute timebox per item and a "park it" column on your Miro board for contested stories can shorten a session considerably. Scrum.org's rundown of common estimation mistakes covers several of these patterns in more depth.
T-shirt sizing in remote and distributed teams
Remote sessions need explicit structure that co-located teams can sometimes do without: a shared digital board everyone can see, a strict reveal protocol and a named timekeeper. Without them, remote sessions slide into sequential sharing, where participants hear each other's answers before committing.
Distributed teams across time zones can size asynchronously. Tools like Parabol let team members record their estimate independently before a synchronous debrief that covers only the outliers. A common pattern for distributed engineering teams is async sizing followed by a short sync of around 20 minutes to settle contested items, which takes far less meeting time than fully synchronous planning poker.
Psychological safety matters more in remote settings. Junior engineers and new team members can feel exposed when their estimate differs from a senior colleague's. A blind simultaneous reveal protects minority opinions that might otherwise be dropped before the conversation starts.
Integrating t-shirt sizing into backlog refinement
T-shirt sizing fits best into the triage and coarse grooming stage of backlog refinement, where items are checked for readiness rather than planned into a specific sprint. Many teams use it as a readiness check: XS, S and M items can go into sprint planning, while L and XL items get flagged for splitting first.
In SAFe environments, t-shirt sizing often appears in program-level refinement to size Features before they are broken into Stories. Product Managers can then forecast capacity across program increments without waiting for a full story-level breakdown.
Recording the sizes in your backlog tool gives you a feedback loop. If your M items are usually done in 1-2 sprints, that is calibration data. If they regularly spill over, the team's definition of M needs a look. Checking the reference stories against delivery history is how teams improve, instead of repeating the same miscalibration every quarter.
Workshop Weaver includes facilitation templates for t-shirt sizing sessions, with the reference story setup and the timebox structure described here, so teams can run the process properly from the first session.
Closing thoughts
The size label matters less than the conversation it forces. When a team splits between XS and L on the same story, something has come to light: the story is unclear, people picture the work differently, or nobody knows how complex it is. That conversation has to happen before development starts.
If you haven't run a t-shirt sizing session yet, try it at your next backlog refinement. Use the five phases above, set up the reference stories beforehand, enforce simultaneous reveal and timebox discussions to three minutes.
After that, read about planning poker for when you need more detail at story level, and dot voting for a fast prioritisation method alongside estimation. Between them, the three techniques cover most of what a team needs for estimation and prioritisation.