What is fellowship management software?
Fellowship management software runs a fellowship from the first application to the last alumni survey: it collects applications, supports selection, follows each fellow through the program and keeps in touch afterward, ideally on one record per person. A research award and a leadership cohort can share an application process and still need very different evidence once the fellows arrive.
Unmesh Sheth, Sopact’s founder, hears the same story from grant managers who have run Submittable, Fluxx or Foundant for years. The platform works, yet every cycle still takes too long and the answers still end up in a spreadsheet. Fellowships often sit at the edge of that picture: one foundation Sopact is talking with runs three fellowship programs outside its main platform altogether. That edge position is the opportunity: one cohort is a complete test of an AI-native workflow, from an intake that listens to alumni follow-up on the same ID.
Why a fellowship is the best first AI pilot
Asked where to start, Unmesh is consistent: think about the whole grant lifecycle, but begin with the application process, and begin it on a fellowship or scholarship rather than a large grant that needs heavy due diligence. A misjudged stipend is recoverable in a way a misjudged multi-year grant is not, and a program running outside the main platform has no migration to plan and no configured workflow to untangle. And the cycle is short: a cohort is recruited and selected within weeks, so you see whether the approach works long before the 6 to 9 months Unmesh says a traditional grant-system implementation takes.
One reason is specific to fellowships. A scholarship decision often ends in a payment and a yearly renewal check; a fellowship decision starts a relationship with someone you will coach, observe and survey for a year or more. The application is the first conversation in that relationship, and if the intake is designed to listen, it becomes the fellow’s baseline.
WATCH THE FIRST STEP
See an AI-native application review, step by step
The pilot starts with the review process itself, and this walkthrough shows it end to end. Unmesh designs an intake that asks applicants to explain themselves, lets 80 applications arrive while an Intelligence Cell reads each one as it lands, then asks which ten are strongest and gets back a report the team can share. Watch the intake design most closely: for a fellowship, those open answers are what the committee reads first and what you will compare each fellow’s growth against later.
Jump to: 0:28 — Barriers in applicants’ own words · 1:14 — Designing an intake that listens · 2:42 — Asking questions in Claude or ChatGPT
Use this in your pilot: Write down the three open questions you would most want to ask a fellow again at exit, and put them in the application.
The cohort journey, from listening intake to alumni
Map one complete cohort before comparing products. The form is rarely the expensive part; the cost sits in the handoffs, where scores are reconciled in spreadsheets, fellows re-enter what the program already holds and check-ins cannot be matched to the application.
| Cohort stage | What to collect | Decision it should support |
|---|---|---|
| Listening intake | Profile and eligibility fields, plus open answers about what the applicant wants to lead, what stands in the way and what support they need. | Is the application complete and eligible? What does this person want from the fellowship? |
| Selection | Rubric criteria, the Intelligence Cell analysis with source passages, reviewer scores, interviews and the committee decision. | Why was a candidate selected, waitlisted or declined, and where did reviewers disagree? |
| Onboarding | The application as baseline, goals agreed at kickoff and the mentor match. | What does support look like for this fellow, and what will you check at exit? |
| Participation | Reflections, check-ins, mentor observations, attendance and deliverables. | Who needs support now, and which activities are helping? |
| Exit and alumni | Exit feedback, six- and twelve-month follow-up and contact updates, each with its date. | What changed, for whom, and what can the program credibly report? |
None of this needs one enormous form. It needs one persistent ID per fellow that every contribution attaches to, shared definitions for what you compare, and access rules that show each person only what their role requires.
A leadership fellowship, followed on one ID
Consider a fictional professional association running a ten-month leadership fellowship, with 120 applications for 20 places in two languages. Beyond profile and eligibility questions, the application asks what the applicant wants to lead in the next two years, what has stopped them so far and what they would need from a mentor.
As each application arrives, an Intelligence Cell configured with the rubric drafts an assessment against each criterion, cites the passage it relied on and pulls out the obstacles applicants name in their own words. In the application-review video’s workforce example those were transportation, childcare and no clear goal; for leadership fellows they are more likely to be time or a manager who will not sponsor the project. When applications close, every file has had the same first read; reviewers ask the AI Assistant who best meets the criteria and who sits on the borderline, check the cited passages, and the committee selects.
The twenty fellows carry the same ID into the program, so their application answers become the baseline without re-entry. Reflections, mentor observations, completed projects and the six- and twelve-month alumni responses all land on that record with their source and date.
The payoff shows in the untidy cases. One fellow reports stronger confidence at exit but has not yet taken on a leadership role; her mentor describes a successful project; her twelve-month response mentions a promotion. Those answer different questions, so the record keeps each source and date visible instead of blending them into one impact score, and a public report can draw only on evidence approved for that audience.
Design an intake that listens, then keep the record straight
The application-review video makes a distinction worth keeping: an application looks like a survey, but it is a data-collection design. Where someone lives or their primary language is useful, but it does not build context; open questions and a short proposal do. For a fellowship, the test for each open question is whether you will want to ask it again at exit. If so, word it so the later answer can be compared with the first.
Give each fellow a persistent ID and keep a separate identifier for each application and cohort, because people apply twice, change email addresses and join more than one program. Let fellows correct their profile, record when a field changed, and keep the value at entry beside the current one. For programs across chapters, agree a small set of shared fields in a data dictionary and let local teams add their own. Set follow-up around decisions: a short check-in during the program can surface a support need in time to act, while career changes suit an annual alumni survey.
Customer practice: one bilingual intake for a cross-border leadership academy
The Institute of the Americas is designing the CaliBaja North American Leadership Academy, a planned ten-month program for emerging leaders aged 26 to 40 who work across the public, private and civil-society sectors of the CaliBaja region, on both sides of the Mexico–United States border. The application is built to work in English and Spanish from the start. Reviewers in both countries share one set of selection criteria and one rubric, and each application file keeps its source material next to the evaluation, so the committee can see why a candidate was scored the way they were.
The design goal is put plainly: “One bilingual application pathway can give emerging leaders across Mexico and the United States a fairer, more consistent starting point.” This is a planning story. The Academy is not yet open for applications, so there are no results to report. What it shows is the order this page recommends: agree the criteria and the listening questions first, keep human review at the center, and set up the fellow record so later cohort and alumni evidence has somewhere to land. Read the Institute of the Americas story →
What the AI reads, and what the committee decides
In Sopact, each field that needs judgment gets an Intelligence Cell, a prompt you configure with the criterion, what counts as evidence and what an insufficient answer looks like; it analyzes each answer, essay or document as it arrives. An Intelligence Row summarizes each fellow across fields and, later, across the program. The AI Assistant answers plain-language questions about the pool with sources attached, and the same data can be queried from Claude or ChatGPT through MCP.
People decide who is selected, and AI output has to be checked before anyone relies on it. A cited passage shows where a draft came from, not that it is fair, so test the reading across both languages and different writing styles, check whether missing evidence is being scored as weak ability, and record the reason whenever a reviewer overrides a draft. During the program, a discouraged reflection may signal learning or a hard month, and it should prompt a conversation rather than a label. Connected records show what changed for fellows; claiming the fellowship caused it needs an evaluation design.
Report alumni results with the denominator showing
Choose outcomes that fit the fellowship’s purpose: publications and progression for a research award, responsibilities and skills applied for a leadership cohort. Then report who answered as well as what they said. If 12 of the 20 fellows answer the twelve-month survey and eight report a new responsibility, the finding is eight of twelve respondents, not eight of the cohort. Keep contradictory accounts and missing responses visible; analyzing pre-, mid- and post-program data covers the method.
Unmesh closes the application-review video with how customers describe the change: work that used to take months now happens as daily business. “It’s not impact measurement. It’s a result-based learning process.” For a fellowship, that means reading this cohort’s evidence in time to shape the next one, because the program wants to learn from it.
Compare platforms by the cohort work your staff will own
Established platforms offer much more than submission forms. OpenWater describes fellowship application, review, communication, reporting and integration functions. Submittable serves fellowship programs alongside grants and scholarships, with configurable forms and review workflows. Their public positioning is a starting point; the demonstration should run the cohort journey you mapped above.
| Option | Why it belongs on a shortlist | What to demonstrate |
|---|---|---|
| Sopact | One persistent ID per fellow; an Intelligence Cell reads each application on arrival; the AI Assistant answers across application, program and alumni records. | Your rubric on past applications, a mentor observation, a twelve-month alumni response, a correction and the resulting cohort report. Confirm operational integrations separately. |
| OpenWater | Fellowship intake and review with reporting and integrations. | How program-period evidence and alumni responses connect to selection records, and what staff must configure or maintain. |
| Submittable | Application and program workflows for fellowship and other funding programs. | How recurring fellow-level analysis works across cohorts, including missing responses, qualitative evidence and restricted access. |
| Your current application system plus connected analysis | A viable option when selection works well and the main gap is follow-up or reporting. | Who owns identifiers, transfers, corrections and access across systems, and how failed transfers are resolved. |
Sources: OpenWater fellowship management and Submittable program management. Product scope and configuration should be verified against your requirements.
Test a change such as a revised rubric or a corrected alumni response: can the program lead make it and reproduce the report without a specialist? The story Unmesh hears most often is that the admin who built the workflow left and now nobody dares touch it, so ask who will maintain yours. Keep finance boundaries explicit as well; recording stipend status is different from issuing payments or handling tax documents.
Run the pilot on one cohort, alongside what you already have
Nothing has to be switched off. Keep the current grant system running for the programs it serves well, move one fellowship cycle, ideally one already outside it, and agree what success looks like before the cycle opens.
- Redesign the intake: keep the eligibility fields, add the open questions you will ask again at exit, and assign each applicant a persistent ID.
- Calibrate the reading: write Intelligence Cell prompts from your rubric, run them on past applications, and compare the drafts with how your reviewers scored them.
- Run the selection: let applications be read on arrival, have reviewers work from the analysis and its cited passages, and record every override.
- Carry it forward: move the selected fellows into the program on the same ID and produce one cohort report that shows response coverage.
- Measure and decide: compare staff time for the old process and the pilot, including checking and corrections, before adding the next workflow.
In the roadmap Sopact recommends, that next workflow is onboarding. Unmesh describes a 60–90-minute conversation with each new grantee whose transcript drafts the reporting plan; for a fellowship, the equivalent is the kickoff with each fellow and mentor, recorded on the same ID so the goals set there are the ones you check at exit.
THE BIGGER PICTURE
Fellowships are where many foundations start
This pilot is one step in a longer plan, and the companion video lays out the whole plan. It follows Maya, a composite of the grant leads Sopact talks to, whose foundation gives about $25M a year through a platform configured over more than a decade: nothing is broken, but everything is slow. Its answer is to pick the smallest, lowest-risk program first, prove application review in one cycle, then add onboarding, reporting and compliance on the same ID. Share it with the director or board member who asks why the fellowship goes first.
Jump to: 3:24 — One ID, the whole story · 4:50 — Pick your smallest grant first · 5:11 — Add one workflow each cycle
Frequently asked questions
Is fellowship software different from scholarship software?
Often, yes. A fellowship usually includes an active program period, with mentoring, projects or cohort sessions, so the record has to hold reflections, observations and deliverables as well as the application. A scholarship leans on financial support, renewals and long-term alumni follow-up, usually at higher volume. Many programs overlap, so compare the work you run rather than the category name. The scholarship management software guide covers the high-volume side.
Why start an AI pilot with a fellowship rather than a large grant program?
The money at risk in each decision is lower, the data is often cleaner because the program sits outside the main grant system, and the cycle is short enough to see results within one round. That lets the team learn how AI-assisted review behaves, compare its drafts with reviewer judgment and measure the staff time involved before applying the approach to grants that need heavier due diligence.
Do we have to replace our grant management system to run the pilot?
No. Unmesh’s advice is to keep the current system running and move one low-risk workflow at a time. A fellowship that already runs outside the main platform is a natural candidate. Run its intake, review and follow-up in the new workflow for one cycle, compare the staff effort and the quality of the evidence, and decide from that result whether to move the next workflow.
Does the AI choose the fellows?
No. The Intelligence Cell drafts an assessment of each application against your rubric and shows the passages it relied on, and the AI Assistant can rank and summarize the pool on request. Reviewers check those drafts, override them where they disagree and record why, and the committee makes the selection. Test the reading across languages and writing styles before relying on it.
Does connecting records prove the fellowship caused an outcome?
No. Connected records improve your ability to examine timing, compare a fellow’s later answers with their baseline and read what fellows say about the program in their own words. A promotion after the fellowship is an observed outcome. Attributing it to the fellowship needs an appropriate evaluation design, a comparison group where one is feasible and a serious look at other explanations.
