Additional course links
You leave with: A launch checklist and a five-failure test log for the next program, with an owner for every unresolved item.
Where this fits: Lesson 5 ends with a keep, extend or replace decision. If yours is to run another program on the workflow R-2 tested, this page is the launch plan. Bring your R-2 test log; you leave with a launch checklist and a five-failure test log for step 5 of the workbook.
Watch · 6:30–8:14 · start with applications, fellowships first
Unmesh's first best practice: think end to end, start with the application process, and start with fellowships and scholarships rather than large grants that need heavy due diligence. At 7:40 he adds the rule behind this page: do not boil the ocean, and a human still reviews. Watch on YouTube ↗
To onboard a grant or RFP program, translate its approved rules into records, build the smallest intake that serves each criterion, and configure review and clarification separately. Then run a fictional submission and five failure tests before launch, and connect selection to the first reporting period. On a workflow you have already tested, that is a setup of weeks, not a new implementation.
Launch the next program on what you tested
Horizon's next program is a small fellowship for frontline staff at youth employment programs (fictional). It is the kind of program Unmesh suggests starting with: fellowships and scholarships need less due diligence than large grants, with low dollars at risk and a short cycle. The temptation is to rebuild it in the configured system, because that is where programs have always lived.
The better move is to put it on the workflow R-2 proved. In typical timelines from Sopact's work with application programs, traditional application software, non-AI-native tools such as Submittable or SurveyMonkey Apply, takes 2–3 months to set up the form, rubric and workflow. AI-native setup of intake, rubric and prompts takes about 2 weeks.
Copy what the pilot tested: the role table, the clarification rule, the conflict process, the calibration step and the handoff to reporting. Do not copy R-2's form or rubric wholesale. A fellowship asks different questions of a different applicant; the fellowship management page shows what changes when the applicant is a person rather than an organization.
This page concerns applications for funding. Procurement tenders may need different processes, even when the call is labeled an RFP.
Translate the approved process into records
Begin with the opportunity, not a blank form. Record its purpose, who is eligible, dates and time zone, available funding, permitted uses, submission requirements, decision authority and clarification policy. Check that the public instructions and internal rules say the same thing.
| Record | Keep attached |
|---|---|
| Applicant | A persistent unique ID and current contact details; for the fellowship, a person, linked to a host organization |
| Opportunity and round | Published rules, dates and approved versions |
| Application | Submitted answers, files, timestamp and revision history |
| Review | Reviewer, criteria version, evidence, scores and conflict status |
| Award | Decision, approver, agreement, conditions and amendments |
| Progress report | Award, period, agreed measures, narrative and supporting files |
The fellowship adds a wrinkle R-2 did not have. A fellow is a person, and their host organization may be an R-2 grantee such as Eastgate. Link the two records; do not merge them. A later application or report should never overwrite an earlier one.
Build the smallest complete intake
For each field or file, state its purpose: identity, eligibility, evidence for a criterion, administration or an agreed reporting need. If a question serves none of these, cut it. Several answers may support one criterion, and not every answer needs a score.
Write the Intelligence Cell prompt alongside the question, so each answer is read against the criterion it serves as it arrives. If you cannot write the prompt, the question is probably not clear enough for applicants either.
Reuse reliable details where they exist. A host organization already on file from R-2 should confirm its details rather than retype them, while the application keeps the values as submitted, so a later profile update does not rewrite its history.
Set required and optional fields on purpose. Explain file formats, deadlines and what happens if evidence is missing. Test save-and-return, upload recovery and accessibility in the actual setup before promising them to applicants.
Configure review and clarification separately
Assign reviewers and test conflict handling, reassignment and access, as R-2 did with A-17. Put the criteria, anchors and missing-evidence rules where the review happens. When the AI suggests a score, the reviewer should see the passage behind it and be able to record a disagreement.
Keep confirmed ineligibility apart from an unanswered eligibility question. Apply the published clarification policy the same way for everyone, with an owner, a deadline and a record of the response. One applicant should not receive unlimited coaching while another receives none.
A green status is not a verdict on merit. "Required evidence received" or "ready for committee review" describe where an application is, not whether it deserves funding. Selection stays a separate decision.
Run a fictional pilot submission
Horizon example · fictional
The fellowship requires a personal statement, a learning plan and a letter from the applicant's supervisor at the host organization. A test applicant, F-01, submits the statement and plan but not the letter. The published policy allows one clarification request within a set window.
The expected result is a visible missing-evidence status, a focused draft request for staff to review and a due date calculated under the round's rules. The workflow should not invent the letter, mark F-01 ineligible without review, or quietly score the missing evidence as zero. This is A-44's case from R-2, replayed on a new form.
When the letter arrives, keep the original submission and the later attachment side by side. A reviewer checks that the letter meets the requirement, and only then does the status change, with the reviewer and date recorded. Test the case where the letter never arrives, too.
Test five failures before launch
| Test | Expected behavior | R-2 precedent |
|---|---|---|
| Incomplete evidence | Apply the published rule; missing evidence is not confirmed failure | A-44, held for one clarification |
| Duplicate or corrected submission | Identify the current version and keep the earlier one | New for the fellowship |
| Reviewer conflict | Restrict or reassign access and keep the decision | A-17, reassigned before scoring |
| Score disagreement | Keep both assessments and the resolution | A-31 feasibility, 4 and 2, settled at 3 |
| Award handoff | Carry commitments, reporting measures, first due date and owner into the award | Eastgate's first quarterly request |
For each test, record expected behavior, actual behavior, evidence, issue owner and retest result. Include an interrupted upload, an applicant on a phone and a submission close to the deadline. Use test records, and never send test messages to real applicants.
Connect selection to the first reporting period
Award approval should create or link the agreement and the reporting plan on the same record. As with Eastgate, a 60–90 minute onboarding call after award defines goals, responsibilities, metrics and theory of change, and that becomes the plan.
Store the agreed commitment as a versioned definition, not a number to overwrite when the first report arrives. Keep each reported result, with its period, source and review status, as its own record.
A fellowship will measure different things from a youth employment grant. Identify the small shared core Horizon needs across programs, map compatible measures through the data dictionary, and keep the local questions that make each program useful to its own team.
Where money is involved, name the accounting system of record and the matching award IDs. Connected reporting supports reconciliation; it does not replace finance's controls.
Test the configured workflow
In Sopact Sense, start from what R-2 tested and adapt it: the persistent unique ID, the Intelligence Cell prompts per question, the role table and the reporting handoff. Test what your team can configure and maintain without help, including routine changes, source retrieval, exception review and reporting permissions.
AI can draft a review packet or a clarification request. A person checks the evidence and approves anything consequential.
Before each reuse, check dates, criteria, permitted uses, reporting commitments and messages. A template should save setup time without carrying last round's conditions into this one.
Put it into practice
Prepare a launch decision
- Using step 5 of the workbook, list what you will carry over from the R-2 test log and what the new program needs that R-2 did not have.
- Write the purpose of every intake field and its Intelligence Cell prompt. Cut any field without a purpose.
- Trace one fictional application from intake to the first reporting request.
- Run the five failure tests and keep their evidence.
- Ask a colleague who was not involved to explain every status change, then resolve what they could not explain and retest.
- Record the launch decision, its approver and any non-blocking work with an owner.
Download the workbook (PDF)
Practice data: horizon-r2-applications.csv · horizon-rubric.csv · calibration-scores.csv · eastgate-q1-update.csv
Check your reasoning before moving on
A good carry-over list includes the role table, the clarification rule, the conflict process, calibration and the reporting handoff, and leaves out R-2's form and rubric. The fellowship's new needs include a person-level record linked to a host organization and a supervisor letter as evidence.
F-01 should end up held with one clarification request, not rejected or scored zero. The launch is ready when all five tests pass or have an owner and a date, a colleague can explain every status change, and the first reporting request carries the measures agreed at onboarding.
Questions grant teams ask
How long does it take to onboard a new program?
In typical timelines from Sopact's work with application programs, AI-native review takes about 2 weeks to set up intake, rubric and prompts, 1 day for AI to read and score every application, and 1–2 weeks for human judgment and follow-up with applicants. A program launched on a workflow you have already tested should sit at the short end, because the roles, rules and handoffs are proven.
Can one form be reused for every funding program?
Reuse the structure and the common fields, but review each program's purpose and rules. A fellowship and a youth employment grant ask different questions of different applicants. Keep the local questions that make each program useful and map only the compatible measures for portfolio reporting.
Does every direct award need competitive scoring?
No. A direct award can use a different approval process while keeping the evidence, due diligence, authority and reporting requirements that apply. Record who approved it and on what basis, and connect it to the same reporting plan as a competitive award.
What does "ready for review" mean?
Whatever you define it to mean, written down: for example, all required evidence received and assigned to an eligible reviewer with no declared conflict. It is a workflow status, not an approval and not a judgment of the applicant's quality.