Designing the application form
The form is a screening tool that runs before any human reviewer reads anything. Every field is a trade-off: it either improves your ability to compare applications or it adds friction that loses you good submissions. The right number of fields is fewer than most organizers think.
The fields that consistently earn their place:
• Talk title and 100–200 word abstract. The single best predictor of talk quality. If you only had one field, this would be it.
• Three to five learning outcomes or "what the audience will leave with". Forces the applicant to articulate value rather than topic.
• Format preference (talk, workshop, panel, lightning) and length.
• Track or theme selection from a fixed list.
• Speaker bio (one paragraph, plus a long-form version optional).
• One or two links to prior talks (video preferred, deck or transcript acceptable).
• Logistical fields: location, willingness to travel, accommodation needs, accessibility requirements.
Fields to drop unless you have a specific reason to keep them: full draft of the deck, full talk script, three references, repeat fields that ask the same question from different angles, free-text "anything else?" boxes that nobody reads. Each one shaves a measurable number of senior applicants off the top.
Mark optional fields explicitly. Senior speakers fill out forms quickly when they trust the form is not going to surprise them in the middle.
A structured speaker request form on SpeakUp captures the fields above by default and routes applications to a review queue with prompts for each reviewer.
Scoring rubrics and panel calibration
A rubric is the document your reviewers should be looking at while they read each application. Without one, every reviewer scores against their own implicit model and the final meeting becomes a debate about what "good" means rather than which talks make the lineup.
A workable rubric has three to five dimensions, each scored 1–5, with one sentence describing what each score level looks like. Common dimensions:
• Relevance to track — does this talk belong in the track it was submitted to?
• Depth of expertise — does the applicant have the standing to deliver this talk?
• Novelty — is the perspective fresh, or have we seen versions of this talk at three other conferences this year?
• Clarity of takeaways — will the audience walk out with something concrete?
• Speaker craft — does the prior-talk video show command of the room and the material?
Before the panel reviews independently, calibrate. Pick three sample applications (one strong, one borderline, one weak) and have everyone score them as a group. Discuss any score that varies by more than one point. Twenty minutes of calibration up front prevents three hours of debate later.
Keep the rubric visible alongside every application during review. Reviewers should be able to score and add a one-line note without leaving the screen.
Reviewing applications efficiently
At fifty applications, three reviewers, and the rubric above, you are looking at roughly six to eight hours of review time spread across the panel — plus a final selection meeting. The structure that keeps this manageable:
• Two reviewers per application minimum, three for borderline cases. One reviewer is the assignment; a second pass catches the cases where a single reviewer was having a bad afternoon.
• Review in track-batched order, not submission order. Comparing fifteen distributed-systems abstracts back-to-back makes scoring far more consistent than reading a distributed-systems abstract, then a frontend one, then a leadership one.
• Cap review session length at ninety minutes. Score quality drops sharply after that. Three ninety-minute sessions over a week outperform one all-day marathon.
• Use a tool that lets reviewers see each other's scores only after they have submitted their own — anchoring bias is real and well-documented.
• Flag, do not debate, in the review phase. Reviewers add a short note when an application needs panel discussion. Save the debate for the calibration meeting where everyone has the rubric in front of them.
The difference between a panel that finishes the round in two weeks and one that drags it out to six weeks is almost entirely about whether the review tooling lets people focus on judging rather than on logistics.
Purpose-built speaker management software handles reviewer assignment, blind scoring, calibration meetings, and the decision-letter workflow in one place — far less brittle than a shared spreadsheet.
Communicating decisions
How you communicate decisions sets up your speaker relationships for the next five years. Senior speakers remember rejection letters. They especially remember silence after a submission.
For accepted speakers, send a single email that contains: confirmation of the slot, the next steps, the content deadlines, the contract or speaker agreement, and a single point of contact. Don't split this across five emails. The most common cause of late content delivery is buried deadlines in scattered messages.
For rejected speakers, send a short, respectful note within forty-eight hours of the final decision. Three sentences is fine. State the decision, thank them for the submission, and — if true — note that you would welcome a submission for a future edition. Do not write a custom rejection rationale unless you genuinely have time to do it for every rejected applicant. Inconsistent feedback (some get rationale, most do not) generates more goodwill loss than uniform brevity.
Do not leave anyone in limbo past the notification date you published. If decisions are slipping, send a holding email with a revised date. Silence past the published date is the single fastest way to lose your alumni network.
Building a reusable application pipeline
The reason most organizers rebuild their CFS process from scratch every year is that nothing was saved at the end of the previous round. The rubric lives in someone's Google Doc. The form lives in a Typeform that someone duplicated and edited. The reviewer list is in last year's WhatsApp group.
At the end of every round, archive: the call text, the form structure, the rubric, the reviewer scoring history, the decision letters, the accepted-and-rejected applicant lists with light tags (track, format, region, seniority), and a one-page retrospective on what went well and what to change.
Next season, the round opens with a duplicate of the previous one, edits to themes and dates, and a one-week setup instead of a one-month setup. The applicant database compounds: rejected-this-year speakers can be invited proactively for next year if their abstract still fits. Three rounds in, you stop sourcing speakers cold and start curating from a pool that has already shown intent.