The accelerator gets smaller
Batch sizes grew for years on the theory that more companies meant more outliers. The largest programme has been trimming them.
Patrick Perkins via UnsplashAn accelerator is a portfolio strategy wearing a programme. Standardised terms, a fixed cohort and a demo day produce enough shots that a small number of outliers carry the fund, which is why batch sizes grew for a decade.
They have stopped growing. Descriptions of the current model note batch sizes moderated well below their peak, with the largest programme running smaller cohorts more frequently rather than very large ones twice a year.
The arithmetic that justified growth still holds and ran into a constraint. More companies produces more outliers only while the inputs to each company hold constant, and attention does not scale with cohort size.
What a founder gets from an accelerator is mostly attention and access: partner time, warm introductions, and a demo day where investors are actually looking. All three dilute as the cohort grows, and the last one dilutes fastest.
Demo day is where the dilution became visible. An investor who could once review a whole batch cannot, so they filter, and the companies below the filter get less from the programme than the brochure describes.
The move to more frequent, smaller batches addresses that directly. It keeps annual volume roughly intact for the fund while restoring the density of attention per company, which is the product being sold.
The standardised terms remain the most underrated part of the model. A known instrument at a known price removes weeks of negotiation for founders with no leverage, and that is worth more than most of the curriculum.
Competition has also changed what the programme has to supply. When capital was scarce the introduction was the value; with more sources of early capital available, the accelerator competes on selection signal and network rather than money.
Selection signal is the durable asset and the most fragile. It rests on outcomes from cohorts admitted years earlier, and it degrades quietly if admission standards move to fill a larger batch.
Which is the most plausible reading of the reversal. A programme whose product is a signal has to protect the signal, and the cheapest way to weaken one is to issue more of it.