The programme had been running for three years and six cohorts. Cohort quality, as measured by eighteen-month company outcomes, had been inconsistent. Some cohorts produced strong companies. Others produced companies that looked strong at demo day and then stalled. The managing director could not identify a pattern in who had been selected. The programme’s evaluation rubric was the same across all three stages of selection: application review, screening call, and final panel.
In year four, they brought in an external consultant who asked one question: what decision are you actually trying to make at each stage, and are your criteria designed to answer that question? The answer was that they were not. The application screen was asking questions about market size and product vision that were better answered in a conversation. The screening call was asking questions about traction metrics that were better answered by looking at the application data. The final panel was asking questions that were appropriate for Stage 1 rather than Stage 3.
The redesign took four weeks. Criteria were separated by stage and redesigned to answer stage-appropriate questions. Application screening was restructured to filter on two binary signals: does this company fall within the programme’s thesis, and is the founder capable of clear written communication? The screening call was restructured to assess coachability and problem clarity. The final panel was restructured to assess programme fit and the specific development gap the accelerator could address.
The next two cohorts produced the programme’s strongest eighteen-month company outcomes to date. The selection pool had not improved. The selection process had.
Most accelerator selection processes produce inconsistent cohort quality not because the best companies do not apply, but because the selection process is not designed to find them.
Three Process Design Failures, a Five-Stage Selection Framework, and the Evaluation Dimensions That Distinguish Strong Companies
The three selection process design failures
Uniform criteria across stages. Research on accelerator selection processes found that accelerator managers’ implicit decision criteria naturally shifted from criteria related to whether an opportunity is real or capable of winning during initial screening, to different criteria related to whether an opportunity is worth doing during final selection. When programmes apply the same rubric at all stages, they are forcing evaluators to apply criteria that are appropriate for one decision context to a different one. The result is that the screening stage does not filter efficiently, requiring evaluators to spend time on companies that a well-designed screen would have eliminated, and the final panel stage does not discriminate effectively, because the criteria it is using were designed for volume filtering rather than quality selection.
Panel scoring without pre-registered evaluation rubrics. Most programme panels score applicants using criteria that are described to panellists at the start of the session rather than pre-agreed and defined with specific observable indicators. Panellists with different reference points interpret the same criteria differently. A criterion called “team quality” means something different to a panellist who has evaluated fifty founding teams and something different to one who has evaluated five. Without pre-registered rubrics specifying what a strong, adequate, and weak response to each criterion looks like, the panel score reflects the composition of the panel as much as the quality of the company. This produces cohort inconsistency across cycles even when programme criteria are nominally stable.
Reference checking done after selection rather than before. Most programmes conduct reference checks, where they conduct them at all, after the final panel. By this point, the selection decision has effectively been made and the reference check is a compliance step. References conducted before the final panel, when the question being answered is “does this company belong in the final pool?” rather than “does this company deserve the place it has already been offered?”, produce materially different information. A reference that reveals a founder has a pattern of not implementing feedback is useful before the final panel. After, it is an uncomfortable data point that rarely changes the outcome.
A five-stage selection framework
| Stage | Decision Question | Criteria | Decision Makers | Time Investment per Application |
| 1. Application screening | Does this company fall within the programme’s thesis, and is the founder capable of clear written communication? | Binary: thesis fit (yes/no); written clarity (is the problem statement specific and the solution described without jargon?). No scoring of market size, team credentials, or traction at this stage. | Programme staff; no external reviewers needed | 3–5 minutes per application |
| 2. Written response review | Does this founder demonstrate the ability to identify what they do not know, and can they describe a specific operational challenge they are currently navigating? | One open question added to shortlisted applicants: “What is the most important thing you need to figure out in the next 90 days, and what have you tried so far?” Scored on specificity (not aspiration) and evidence of active experimentation | Programme staff with a defined scoring guide (1–3 scale on each dimension) | 10–15 minutes per shortlisted application |
| 3. Screening call | Does the founder demonstrate coachability, and can they distinguish between what they know and what they are guessing? | Coachability indicator: how does the founder respond to one specific challenge of an assumption in their application? Problem clarity: can the founder describe the problem their customer has, in the customer’s words, without pivoting to the product? | One programme staff member plus one mentor from the network | 20–30 minutes per call |
| 4. Reference check | Does the founder’s track record across prior experiences suggest they will act on programme feedback and apply it to decisions? | One reference from someone who has worked with or invested in the founder; one open question: “Tell me about a time when this founder received feedback they disagreed with. How did they respond?” | Programme staff; reference call conducted before the final panel | 15–20 minutes per reference |
| 5. Final panel | Does this company represent a programme fit: is there a specific development gap that the programme is designed to address, and is the founder likely to use the programme’s resources effectively? | Programme fit score (1–5) on three pre-registered dimensions: clarity of the development gap the programme can address; founder’s stated priorities for the programme versus what the programme actually offers; evidence from prior stages that the founder learns under pressure | Panel of 3–4 members with pre-registered rubric and independent scoring before discussion | 30–45 minutes per finalist |
The critical structural principle is that each stage answers a different question and uses criteria designed for that question. Stages 1 and 2 are designed to reduce volume efficiently; they should take under five minutes per application at Stage 1 and under fifteen at Stage 2. Stages 3 and 4 are designed to assess the qualitative dimensions that cannot be assessed from written materials. Stage 5 is the only stage where programme fit, not company quality in the abstract, is being evaluated.
The evaluation dimensions that distinguish strong early-stage companies from impressive-looking ones
The most common selection error in accelerator programmes is selecting for pitch quality rather than company quality. Pitch quality, meaning the coherence of the narrative, the specificity of the market size claim, and the visual quality of the deck, is learnable in six hours with a good coach. Company quality is not. The two are correlated only weakly at early stage.
The evaluation dimensions that most reliably predict whether a company will benefit from and build on an accelerator programme are not the ones that appear most prominently in most rubrics.
What founders do not know and can articulate. A founder who can name, in specific terms, what they currently believe to be true but cannot yet confirm, is demonstrating epistemic clarity that predicts programme benefit. A founder whose application contains no uncertainty is either not being honest or does not have a working model of their own knowledge gaps. Both are programme risks.
How founders respond to challenge. The screening call’s most useful data point is not the answer to the challenge; it is the founder’s behavioural response to receiving one. A founder who integrates a challenge, even partially, into their next answer, has a learning orientation that predicts accelerator benefit. A founder who becomes defensive or pivots away from the challenged assumption is demonstrating a pattern that the programme will spend significant time navigating.
Whether founders can describe their customer’s problem in the customer’s language. This is the single most reliable indicator of whether a company has genuine product-market understanding rather than product conviction. Founders who describe their product’s features as the solution to a problem they have inferred are different from founders who can quote the specific language a customer used when describing the problem. Research on impact accelerator selection found that all team-related criteria were significant for selection except teamworking, communication, and leadership skills, suggesting that programmes systematically underweight the qualitative dimensions of how founders actually work in favour of structural credentials.
The coachability reference. The question to ask a reference is not “would you recommend this founder?” Almost every reference provided by a founder will say yes. The question that produces useful information is behavioural and backward-looking: describe a specific instance where this founder received feedback that challenged something they believed. What happened next? The answer reveals whether the founder’s pattern of behaviour under challenge is one the programme can work with or one it will spend twelve weeks managing.
The Implication
Selection process design is the highest-leverage investment a programme can make in cohort quality because it determines the quality of every programme output that follows. Curriculum quality, mentor quality, and programme duration all matter. They all operate on the cohort that selection produced. A selection process that admits companies that are impressive-looking but not coachable, or companies whose development gaps are not the gaps the programme is designed to address, ensures that curriculum and mentors are deployed on the wrong problems.
Programme directors who focus on quality over quantity, choosing companies that solve real problems and demonstrate both coachability and commitment, build cohorts that produce outcomes rather than cohorts that generate visible activity. The redesign required to achieve this is not expensive. It requires separating the decision questions across stages, pre-registering evaluation rubrics before panels, and moving reference checks to before the final panel rather than after. Each of these changes is operational, not strategic. Together they produce selection processes that find the companies the programme can genuinely help rather than the companies that most effectively navigate the application process.
The programme that does this consistently will produce stronger cohort outcomes than a programme with better resources and weaker selection discipline. Selection is the leverage point. Everything else is downstream of it.