Forty-one grant submissions and one index that never breaks

The organisation

The Kettering Institute employs about a hundred and eighty people across four research groups. Its Office of Sponsored Programs is three staff: a director, a proposal manager and an administrator. Between them they assemble, check and submit every federal and foundation proposal the institute makes — in a busy cycle, upwards of forty.

The office does not write proposals. It receives parts from investigators and is accountable for the compliance of the whole. That distinction matters, because it means the office's deadline is always later than everybody else's and always harder.

The problem

A federal proposal is assembled from parts written by people who do not work for each other. Biosketches, a budget and its justification, facilities and resources statements, letters of support, data-management plans, and the research plan itself. On a multi-institution proposal those parts come from six or seven co-investigators at three or four organisations.

They arrive within about six hours of the deadline. They arrive in six formats. They arrive with six different ideas about margins, fonts and line spacing, all of which the sponsor specifies and validates automatically on receipt.

The office's job is to make one compliant PDF, under a hard page ceiling, with the sections in exactly the order the sponsor's guide lists them, with the whole thing bookmarked. Page counts were checked by hand, section by section, by one person reading a printed guide. Twice in two years a submission went in with a section a page over, and both were rejected by the validator without ever reaching a reviewer.

The deeper problem was that the order was knowledge rather than a document. It lived in the proposal manager's head and in a much-annotated printout. When she was on leave during a February deadline, the institute missed it.

What they tried first

The office's first response was a checklist — a two-page document listing the sections in order with their page limits. It was distributed to investigators with every call for proposals. It was, as far as anyone could tell, never opened. Parts continued to arrive named after the person who wrote them.

The second response was a template document with headed empty sections, into which investigators were asked to paste their contributions. This was worse: it produced parts that were formatted, but formatted wrongly, and now had to be un-formatted before being reformatted correctly.

The third was a spreadsheet tracking each section's page count against its limit, maintained by the administrator during assembly. It caught the over-length sections it was updated for. It did not catch the ones added in the last hour, which are the ones that go over.

What changed

The sponsor's section order is now a shared preset. The office holds it; the investigators use it. A part submitted through the preset already knows which section it belongs to, so assembly is placement rather than interpretation.

The compliance check runs against the sponsor's margin, font and spacing rules and reports what fails rather than silently correcting it. That distinction was insisted on by the office: a proposal quietly reflowed to fit is a proposal whose author no longer recognises it, and the author is the one who has to defend it.

The page ceiling is checked on the assembled document, section by section, before the index and bookmarks are generated. Bookmarks come from the section structure, so the reviewer's navigation pane matches the sponsor's guide without anyone building it.

How it was put in

The office built one preset, for its single highest-volume sponsor, and used it internally for a full cycle without telling the investigators it existed. Assembly time for that sponsor's proposals halved before anybody outside the office knew anything had changed.

Only then was the preset shared, and it was shared with the three investigators who submitted most often rather than with everybody. Those three became the reference for the rest, which mattered more than any instruction the office could have written: a preset recommended by a colleague who has used it gets used.

The second sponsor's preset was built by an investigator rather than by the office, from the first, and handed back. The office reviewed it and adopted it. That is now the normal way a new preset appears.

The result

Forty-one submissions went out in the last cycle. None was returned by the sponsor's automated validator, against two in the two years before.

The office stopped maintaining the page-count spreadsheet. It stopped printing the sponsor guide. The February deadline that was missed during a period of leave has been met twice since by staff who had not previously assembled a proposal for that sponsor, which the director regards as the real test of whether knowledge has moved out of a person's head and into a document.

Assembly is now measured in hours rather than in whether it will be finished. The office's stated aim for the next cycle is to stop working the evening before a deadline at all.