Running a multi-round seed-fund call

In short. A multi-round call is an exercise in arithmetic as much as judgement. Decide the funnel width before you open, write the promotion rule before the first score lands, size the jury from reads rather than from who is willing, and tell applicants the dates you will actually meet. The programme that does those four things finishes on time; the one that improvises spends February in a spreadsheet.

Seed-fund calls tend to fail in predictable places. The screening round runs long because nobody decided how many applications move forward. The jury round runs long because fourteen evaluators were asked for "a few reads" and received twenty-nine. Pitch day overruns because eighteen teams were shortlisted for a four-hour slot. None of these are judgement failures. They are planning failures with a numeric shape.

The shape of the call

Start from the end. How many teams can the fund actually support? If the budget disburses to twelve, work backwards: twelve funded, roughly forty at pitch day or the final jury round, and whatever the open call brings in at the top.

A useful worked funnel, and the one used through this article: 181 applications to 40 to 12. That is a screening round that removes about 78 per cent of the field, a jury round that removes about 70 per cent of what is left, and a final stage that is a decision rather than a filter.

The proportions matter more than the absolute numbers. A first round that removes only a third of applications has not done its job and has simply pushed the work into the jury round, where the reads are slower and the evaluators are more expensive. A first round that removes 90 per cent is usually screening on criteria that belong in the jury round.

A timeline that survives a committee

Eleven weeks, with the committee dates fixed first because they are the only dates you cannot move.

Two weeks of the eleven are administrative rather than evaluative, and they are the two most often left out of the plan. Week 5 in particular — eligibility, duplicates, missing attachments — is the week that decides whether evaluators spend their time reading or chasing.

Round one: screening

Screening is not a small version of the jury round. It answers one question: is this application complete, eligible and plausible enough to deserve three expert reads?

Use a short rubric — three criteria at most — and a coarse scale. Screening scores exist to sort, not to rank. Two reads per application, with a third only where the two disagree by more than a set margin, keeps the round cheap. Programme staff and junior evaluators can screen; there is no need to spend a jury member's attention here.

The output of screening is a promotion list, not a ranking. Resist the urge to carry screening scores forward into the jury round; they were produced against a different rubric by different readers, and mixing them makes the final number impossible to explain.

Round two: the jury round

This is the expensive round and the one the committee will interrogate. Three reads per application against the full weighted rubric, blind if the programme's rules allow it, with conflict-of-interest declarations collected before assignment rather than after.

Two things to fix before the round opens. First, the assignment rule: by expertise where the programme has sector tags, randomised within expertise so that no evaluator receives a batch that is systematically stronger or weaker than another's. Second, the disagreement rule: what happens when three reads on one application span more than a set range. A fourth read from the programme lead, discussed rather than averaged, is the usual answer.

Round three: pitch day

Forty teams do not fit in a day. Either the jury round shortlists to sixteen or so for a single-track pitch day, or pitch day runs in two parallel tracks with separate panels and a calibration step afterwards. Both work; the second needs the normalisation discussed below, because two panels are two scales.

Whatever the format, publish the slot length and the question time in advance, run a visible timer, and keep the standings board on a projector that the room can see. A pitch day where the schedule is visible finishes within ten minutes of its plan.

Evaluator workload, as arithmetic

This is the calculation that decides whether the round lands on time, and it takes one line.

181 applications, 3 reads each, 14 evaluators. That is 543 reads in total, or about 39 reads per evaluator.

Now convert reads into hours. A thorough read of a seed-fund application with an attached deck takes 20 to 30 minutes. At 25 minutes, 39 reads is roughly 16 hours of work — two full working days from someone who has a full-time job elsewhere, spread across a fortnight. That is at the upper edge of what a volunteer jury will absorb, and it explains why the last 15 per cent of reads always arrive after the deadline.

The levers, in order of usefulness:

1. Screen harder first. Three reads of 40 applications is 120 reads, not 543. Most workload problems are funnel problems.
2. Reduce reads per application in the early round. Two reads with a tie-break is 80 per cent of the cost of three and loses very little.
3. Recruit more evaluators. Effective, but each addition dilutes the reads per evaluator and makes calibration noisier; below about eight reads each, normalisation stops working well.
4. Shorten the application. A 90-field form produces a 30-minute read. Every field you remove is multiplied by 543.

Build in a buffer of one week and a reminder schedule — at the halfway point, at three days out, and on the deadline morning. Expect to chase two or three evaluators personally; it is normal and it is faster than redistributing their batch.

Promotion rules, written first

Write the promotion rule before anybody scores, put it in the call document, and do not change it afterwards. A rule that is chosen once the scores are visible will always look like it was chosen to produce a particular answer, whether or not it was.

Workable rules, in rough order of defensibility:

• Top N by calibrated score. Simple, explicable, and the default for most funds.
• Threshold plus cap. Everyone above a stated score, up to a maximum of N. Honest about the fact that a weak field should not be padded.
• Top N with a per-category floor. Where the fund has committed to sector or category coverage, state the floors publicly with the rule.

Say what happens to ties, and say who signs off. The committee approves the promotion list; the programme team produces it. Keeping those two roles separate is what makes an appeal survivable.

What applicants hear, and when

Four messages, each on a date published at the start:

1. On submission, an acknowledgement with the case number and the decision date.
2. At the end of screening, a result. Applicants not promoted are told then, not at the end of the whole call. Holding 141 people for six weeks to tell them something you knew in week 6 is the fastest way to lose next year's applicants.
3. At the end of the jury round, a result with the applicant's own scores by criterion and the range for the round.
4. After the decision, the funded list and the disbursement timeline.

Publish the results page and keep the status timeline in the applicant portal current between those messages. Most support email in a call is a status question; a visible status answers it before it is asked.

IdeaScore models each of these as a round with its own rubric, assignment and promotion step, and its reports view will show the funnel and evaluator throughput while the round is still open — which is when the numbers are still useful. The planning, though, happens in a document before any of it is configured. Eleven weeks, 181 to 40 to 12, 543 reads, 39 each.