How to facilitate a backlog refinement session: an agenda, the DEEP criteria, estimation techniques, and methods to keep distributed teams engaged.

If your sprint planning always runs long, the cause may sit one meeting earlier: in refinement, and specifically in how refinement is facilitated.
Most Scrum teams run refinement as a read-through. The Product Owner shares their screen, walks through tickets, and now and then asks "does everyone understand this?" The team nods, stories get estimated, the session ends. Then sprint planning arrives and half the stories need explaining again.
The backlog is rarely the problem. Nobody was facilitating.
What backlog refinement is (and why facilitation matters)
The Scrum Guide describes refinement as the ongoing act of breaking down and further defining Product Backlog items, adding details such as order and size. It's where the Product Owner and team review, estimate and prioritise Product Backlog Items. A useful target for that work is a DEEP backlog: Detailed appropriately, Estimated, Emergent, and Prioritized.
The Scrum Guide doesn't say how to get people to think together in that session instead of passively taking in information.
Roman Pichler has written that refining the product backlog should be a collaborative effort involving the Product Owner and the development team. In many sessions it isn't. The PO knows the backlog best, so they do most of the talking and everyone else follows along. That is running a meeting.
Facilitating means setting up the conditions for the group to think together: designing the session, managing time, balancing participation and steering back when the conversation drifts. It's a different skill from owning the content, which is why the person most invested in the backlog usually makes the worst session leader.
Picture a Scrum team that makes a single change. It moves from a PO-led read-through to a facilitator-led format with silent reading, structured questions and dot voting to choose which topics get discussed. The aim is fewer stories coming back mid-sprint for clarification and shorter sprint planning. The backlog stays the same; only the facilitation changes.
The DEEP criteria: check them before you design the agenda
Before you open your backlog tool, ask which DEEP dimension is weakest this sprint. Mike Cohn and Roman Pichler use the DEEP acronym. Use it as a diagnostic as well as a checklist, and let the diagnosis shape your agenda.
- Detailed appropriately means near-term items are granular enough to sprint on, while distant items stay coarse. If your team keeps discovering unspoken acceptance criteria during sprint planning, this dimension needs work.
- Estimated means every item has a relative size. If planning poker keeps turning up stories that were never estimated, refinement is skipping the work.
- Emergent means the backlog changes continuously. This is the dimension teams ignore most. They spend whole sessions on known items and never bring in what recent customer feedback, technical debt or market shifts have revealed, and the backlog turns into a stale to-do list.
- Prioritized means the order reflects current business value, risk and dependencies, not the order in which stories were written.
If your team has just finished a round of user research, start with an Emergent exercise before touching existing items. If the next sprint has a hard deadline, put more time into Detailed and Estimated. Building the agenda around the weakest dimension gives the session a purpose beyond routine.
Roles in the room
Three roles have defined jobs in refinement. The Product Owner brings prioritised items and acceptance context. The Development Team brings technical feasibility, awareness of dependencies, and estimates. The Scrum Master or a designated facilitator holds the process.
The "hero PO" pattern, where the Product Owner both owns the content and runs the session, fails again and again. The person most invested in an outcome makes a poor neutral facilitator. Separating content ownership from process ownership improves participation straight away.
Rotating the facilitator role among squad members each sprint is worth trying. It builds shared ownership of the backlog and takes load off the PO.
Guest stakeholders and UX researchers can join for specific items, but they shouldn't sit through the whole session without a clear role. When they attend without framing, the session drifts towards a presentation, and developers become less likely to raise technical concerns openly.
A facilitation-first agenda
Here's a structure for a 60-minute session. Time-boxing each segment shows that the conversation has limits, which tends to make people more willing to speak up.
- Minutes 0-5: Check-in and purpose. Instead of "we're here to review the backlog," say what you want to leave with. A one-question check-in ("in one word, how ready does this backlog feel to you?") shows you the team's state before you get into content and lets you adjust the pace.
- Minutes 5-15: New items. Ask what has come up since the last session: customer feedback, production incidents, tech debt. Handling Emergent first keeps the session from settling into known items only.
- Minutes 15-40: Story readiness review. Give silent reading time (2-3 minutes per story, everyone reads alone) before any discussion. Before estimating, put each item through three questions: Do we know what done looks like? Do we know who the user is? Can we see any blockers now? Items that fail go back to the PO for follow-up instead of getting a speculative estimate.
- Minutes 40-55: Estimation. Use Planning Poker or T-Shirt Sizing depending on how many items there are and how soon they're needed (more on both below).
- Minutes 55-60: Prioritisation check. A quick dot vote or stack-rank confirms the order before you close.
This sequence treats refinement as a working conversation instead of a status update.
Estimation techniques that start the right conversations
The number matters less than the conversation the estimation forces.
Planning Poker works because everyone reveals at the same time, so nobody is anchored by the first number they hear. When one person plays a 2 and another a 13, that gap is the most useful result of the exercise. It exposes hidden assumptions and conflicting mental models that have to be resolved before the story is sprint-ready. Facilitators who rush past high-variance results to reach consensus throw that information away.
Enforce the simultaneous reveal strictly. Once a senior engineer calls out a number before the others flip their cards, the protection against anchoring is gone.
T-Shirt Sizing (XS through XL) suits early-stage or high-volume backlogs where Fibonacci precision would waste time. Use it to sort a large set of items quickly in the first half of a session, then use Planning Poker only for near-term stories that need a finer estimate.
For distributed teams, PlanningPoker.com keeps the simultaneous reveal in remote sessions. The facilitator controls when cards flip, so anchoring stays out across timezones.
Prioritisation: making the order visible and shared
Dot voting is a quick, low-conflict way to make the group's priorities explicit. Each participant gets a fixed number of dots (3-5) and places them silently on the items they think matter most. Everyone sees the distribution at once. That is faster than a verbal round-robin and much less dominated by whoever speaks first.
Two rules matter. No campaigning before the vote: nobody argues for items before the dots go down. And people vote silently and independently, then discuss the result. Two debrief questions tend to produce useful answers: "What surprises you about this result?" and "Is there anything in the bottom half we're undervaluing?"
They catch familiarity bias, where teams vote up the items they know best instead of the ones with the most value.
Techniques that keep energy up
Silent reading time is borrowed from Amazon's meeting culture, and it works. Instead of having the PO narrate a story, give the team 2-3 minutes to read the ticket on their own. Everyone gets through it at their own speed, and the PO's framing doesn't shape the team's reading before discussion starts.
Use open question stems in place of yes/no questions. "Does everyone understand this story?" gets nods. "What would need to be true for this to be done in one sprint?" gets real answers. "What is the riskiest assumption in this story?" brings out concerns nobody would volunteer in an open round.
Managing energy is part of the facilitator's job. Refinement scheduled on Friday afternoons or straight after sprint demos tends to go worse. If energy drops mid-session, a change of perspective helps: ask each person to describe the story from the end user's point of view before you estimate. The Liberating Structures menu has several structured interaction patterns that work for this kind of reset.
Remote and async refinement
Distributed teams have a harder version of the problem. Synchronous sessions across timezones are expensive, and fully async refinement loses the live debate that brings hidden assumptions out.
A hybrid model works reliably. Team members annotate stories asynchronously in Jira or Notion before the session and flag questions and concerns. The synchronous session covers only the flagged items. Total session time goes down and the important conversations still happen.
Remote facilitation needs more explicit structure than in-person work because you can't read body language. Named turn-taking ("Let's hear from the backend team first, then frontend"), a visible timer on the shared screen, and a digital tool for dot voting are required, not nice to have. GitLab's public handbook on all-remote work is worth reading as an example of a company that works remotely by default.
For async estimation, review divergent estimates before the synchronous session and set aside time for the high-variance items. Averaging divergent estimates without discussing them is the async version of estimation theatre.
Common failure modes and how to prevent them
Scope creep inside the session is the most common failure: the team starts designing a feature instead of refining a story. When it happens, say so: "I notice we've moved from refinement into design. Let's capture this and schedule a separate spike." Keep the insight and return to the agenda. Without that intervention, a 60-minute session turns into a 3-hour design session with zero ready stories at the end.
Estimation theatre is harder to spot. It happens when a senior engineer shows their card first, when the PO hints at a preferred estimate, or when the team has learned to converge quickly to avoid conflict. The fix is in behaviour: enforce the simultaneous reveal, thank people for divergent estimates, and treat high variance as a sign the story needs more work rather than a sign that someone is being difficult.
Backlog bloat, with stale stories piling up because nobody wants to delete work, can also be fixed through facilitation. A quarterly pruning exercise with time-boxing and dot voting to pick items for archiving keeps the backlog useful for strategy.
When refinement sessions keep overrunning or producing poor estimates, run a facilitated retrospective on the refinement process itself before the next session. A new structure designed with the team lasts longer than a fix imposed from outside.
Refinement is your highest-leverage session
Every Scrum event downstream (sprint planning, daily standups, review) runs better when refinement has done its job. That makes it the facilitation session with the most leverage a Scrum team has each sprint. Run it as a facilitated conversation, because it decides whether the team spends the sprint building the right things or working out what the right things are.
Before your next session, check it against DEEP and find the weakest dimension. Then compare your agenda with the five-part structure above. If it lacks a check-in, a step for new items or a prioritisation close, you know what to add.
For the estimation and prioritisation techniques mentioned here, the practical detail is in the Planning Poker, T-Shirt Sizing, and Dot Voting method guides. Use the agenda structure from this guide and the step-by-step instructions from those.