Chapter 4 of 13 · Funder road · About 15 minutes

Check each partner report against the agreement

Partner reports are arriving as PDFs, spreadsheets, survey exports and emails, and no two look alike. Before anything is added up, you need to know whether each number means what the agreement says it means.

On this page: When reports arrive · Row by row · One question per gap · The gap log · AI first pass · Limits · Try it · Questions

What should you do first when partner reports arrive in different formats?

In short: Before you read the story or add anything up, lay each report beside the agreement and check what every number means. The format does not matter; the definition does.

Reports arrive the way each partner works: a PDF narrative, an Excel sheet, a survey export, numbers pasted into an email the night before the deadline.

The temptation is to copy the numbers into your portfolio sheet. That is where most portfolio errors start: a placement counted at six months sits next to one counted at 90 days, and the total looks precise while meaning very little.

The agreement you wrote after the onboarding call is the yardstick. It names each metric, how it is counted and when it is due; the shared data dictionary holds the full definition behind each row.

How do you compare a partner report with the grant agreement?

In short: Go metric by metric and give each one of four statuses: matches, different definition, missing, or not split. Only a match can go into the roll-up as it is.

Start from the agreement's list of metrics and read the report against it, not the other way round. Read the report first and its structure sets the agenda; an absent metric is hard to notice while you read what is there.

None of the three gaps says anything about performance. Partner C may have placed more people than anyone; you cannot tell yet, because its count uses a different window.

How do you turn each gap into a question the partner can answer?

In short: Write one plain question per gap, name the agreed definition in it, and send all of them together within days of the report arriving. The partner should be able to answer each one from its own records.

A good question is short, specific and answerable. It names the metric and the agreed definition, and asks for something the partner can produce, like the three for Partner C in the slide above.

Send them while the partner still remembers. A week after submission, the person who compiled the report knows which spreadsheet each number came from; three months later, they may have moved on, and the answer becomes a guess.

Keep the tone clarifying, not auditing. You are checking that both sides mean the same thing by the same word. A funded partner often reports to several funders, each with its own template, so a different definition is usually a trace of another funder's form, not carelessness.

The partner's side of this exchange is in Map every funder's ask to one evidence base.

What goes in a gap log?

In short: One row per gap: the metric, what the agreement says, what the report says, and the question you sent with the answer you received. It turns scattered emails into a record you can reuse.

Keep the gap log next to the agreement, one table per period. It answers the question every portfolio manager hears at board time: why is this partner's number not in the total?

Gap log template, filled in for Partner C (fictional). Copy the four columns and add a row per gap.

When an answer arrives, write it in the log, update the partner's number, and only then move it into the roll-up. If the partner cannot give the agreed count, record that too; a held number is still information.

Can AI do the first pass of the comparison?

In short: Yes, for the reading and the lining up. Give it the agreement table and one partner report, ask it to list gaps without filling in any values, and have a person review every row before a question is sent.

The first pass is repetitive: find each agreed metric in a long PDF or spreadsheet, quote it and compare the wording. AI tools do this well when you give them the agreement as a table and forbid guessing. Keep participant names and emails out of what you paste.

You are checking a funded partner's report against our reporting agreement.

Agreement table (one row per metric: name, definition, period, breakdown required):
[PASTE AGREEMENT TABLE]

Partner report:
[PASTE OR ATTACH THE PARTNER REPORT]

For each metric in the agreement table, in the same order:
1. Quote what the report says about it, with the page, sheet or line where you found it.
2. Give one status: matches, different definition, missing, or not split.
3. If the status is not "matches", write one plain, friendly question the partner can answer from its own records.

Rules:
- Do not calculate, estimate or convert any value. If a number is not in the report, write "not found".
- Do not treat a similar metric as a match. If the definition differs in any way, say how.
- Return a table with four columns: metric, agreement, report says, question.

A person reviews every row. The AI can misread a table or miss a footnote; it cannot know what the partner told you on a call; and the question goes out under your name, to a relationship you manage. Treat the output as a draft of the gap log, not the log.

What does this check not tell you?

In short: It tells you whether a number can be compared, not whether the partner is doing well. A clean match can sit in a weak program, and a gap can come from a strong one.

Nor can it confirm that a count is correct, only that it is defined as agreed; tracing a number to its records is covered in Make every number in your report match its source. Some gaps are timing: a 12-month retention figure cannot exist before 12 months have passed.

Once each report is checked, the matched numbers are ready to add up: Roll up and benchmark portfolio results shows how to handle the ones still held.