Evaluate an accelerator by connecting the founders and companies you admitted with what happened during and after the program. Decide which changes matter, collect an appropriate starting point and plan follow-up before demo day. Keep delivery, founder capability and company performance separate so a successful event does not become a claim of business growth.
This lesson is for accelerator managers, program evaluators and teams reporting cohort results. Bring one cohort’s goals and the information you already collect. You will produce a collection-and-review plan covering founder learning, company milestones, follow-up coverage and a decision for the next cohort.
Begin with the decision you need to make
Choose a question such as “Which support should we change next cycle?” or “What can we report about company progress six months after the program?” These questions need different evidence. A founder satisfaction response may help improve a mentoring session; it cannot establish business survival or the program’s contribution to revenue growth.
Keep counts of applications, admitted companies, attendance and completed sessions. They explain reach and delivery. Add selected outcome evidence rather than dismissing these operational measures or collecting every possible business metric.
Separate founders, companies and cohort participation
A company can have several founders, and a founder may be involved in more than one company over time. Use a person record for the founder, an organization record for the company and a dated participation record for the cohort. Link responses and milestones to the unit they describe.
| Evidence | Record it belongs to | Why it matters |
|---|---|---|
| Founder’s confidence applying a skill | Founder and observation date | Two founders can have different experiences. |
| Company revenue for a quarter | Company and financial period | Do not count it once for every founder. |
| Participation in a workshop | Person, session and cohort | Attendance is a delivery record, not demonstrated capability. |
| Company status at follow-up | Company and checkpoint | No response is not evidence that the company closed. |
Choose a useful starting point
Reuse reliable application data when it fits the evaluation purpose. Refresh changing information at program entry where necessary. Do not add commercially sensitive questions to selection simply because you might later want them for research. Explain the purpose and who can access the information.
For each selected measure, record the definition, unit, period and source. Revenue must distinguish currency and reporting period; funding should distinguish announced, committed and received amounts where relevant. A self-rating needs a clear question and scale. It measures a reported perception, not demonstrated ability.
Across regional programs, agree a small shared core for the comparisons leadership actually needs. Local programs can retain their own mentoring questions and sector milestones. A versioned dictionary maps compatible measures and identifies which results should remain separate.
Plan collection around real milestones
Use a few purposeful collection moments: entry, a useful in-program checkpoint, completion and an appropriate later follow-up. Choose timing from the program’s aims and realistic business development cycles. Six or twelve months may fit some programs; they are not universal requirements.
Ask founders for information they can reasonably provide, reuse existing milestone documents and clarify who supplies company-level information. If two founders submit different revenue figures, retain both sources and resolve the period or definition before approving one company result.
Record missing, not yet due and disputed evidence separately. An inactive portal account does not establish business failure. A missed milestone invites a conversation about context, changed plans or support; it does not automatically identify a failing founder.
Review learning and traction with different evidence
Founder learning may be supported by a reviewed task, reflection, mentor observation or reported use of a skill. Company traction may involve customers, revenue, product adoption or another goal appropriate to its stage. One does not automatically establish the other.
Analyze open comments using a defined codebook and review the source passages. Keep barriers, requested support and reported benefits distinct. A theme can suggest what to investigate; it does not prove the cause of a company’s performance. Protect confidential business information in cohort summaries.
Work through a small cohort example
A fictional cohort includes 12 companies and 18 founders. At a later checkpoint, 10 companies respond: eight report operating and two report having closed. Two companies’ statuses remain unknown. Report eight known operating companies, two known closures and two unknowns. The operating share among respondents is 8/10, or 80%; the known operating share among all companies is 8/12, or 66.7%. Neither number establishes what happened to the missing two.
Suppose 14 founders have comparable entry and completion skill self-ratings, and nine report a higher rating. Report 9/14 among matched respondents with 14/18 coverage. Do not present that as nine companies improving or as evidence that the accelerator caused a gain in business performance.
If several founders say customer interviews helped them revise a product assumption, use reviewed excerpts to illustrate that learning. Keep the company outcome question open until evidence supports it.
Compare cohorts only after checking context
A cohort can be evaluated without another cohort. When comparison is useful, check admission criteria, company stage, sector mix, economic conditions, follow-up timing and response coverage. Identical survey wording does not remove those differences.
Conversely, a wording change does not automatically destroy a series. Review whether the construct, scale and collection method remain compatible. Document any mapping or break. Do not silently combine incompatible results to preserve a trend line.
Compare starting points before interpreting later differences. A higher funding total could reflect one large round or a different company mix. Show the distribution and relevant context rather than judging the program from one headline total.
Use Sopact to maintain the evidence across the cycle
Plan connected founder, company and cohort records, then bring in surveys, milestones and supporting documents. Configure analysis around the agreed definitions and review rules. This makes the evidence available for recurring review without asking the team to rebuild the relationships each reporting cycle.
Test one company with two founders before scaling the workflow. Confirm that company figures appear once, individual perspectives remain distinct and permissions protect sensitive information. AI can help organize source evidence; the team approves interpretations and any external claims.
Practice: prepare the next cohort’s evidence plan
- Choose one program improvement decision and one reporting question.
- Define a founder measure and a company measure at their correct record levels.
- Identify the shared core and useful local questions.
- Plan collection moments, contributors, due dates and missing-data handling.
- Write the outcome statement you could make from the fictional example, including its limits.
Frequently asked questions
Must all accelerators use the same metrics?
No. Use measures relevant to the program and company stage. For comparisons across a network, agree a limited compatible core and retain useful local evidence.
Does no response mean a company has failed?
No. Record the status as unknown and follow the agreed verification process. Absence from a platform is not evidence of closure.
Does a higher result prove the accelerator improved?
No. Review starting points, coverage and context. Observed improvement and a causal program effect are different claims.