Reading time only counts when somebody is reading

There is a number in nearly every document-sharing tool that says how long somebody spent with your proposal. It is usually wrong by an order of magnitude, in a direction that flatters you.

The reason is simple. The tab was open for fourteen minutes. For eleven of them it was behind a spreadsheet, and for two more the page was scrolled to the signature block while its reader took a call. Counting elapsed time since load measures none of what you wanted to know, and it fails in the most expensive possible way: by producing a large, encouraging number about a reader who was not reading.

The measurement

In XDocs, time accrues against a section of a document, and only while two things are true at once: that section is actually on screen, and the tab is the one the reader is looking at. Scroll past a clause and its clock stops. Switch windows and every clock stops.

The result is a smaller number than other tools report, and it is a number about a person. A section-level account of attention — this clause for ninety seconds, twice; those four skipped entirely — is a different kind of object from a per-document total, and it supports questions a total cannot.

Why this is only half the work

An honest heatmap is still a heatmap: a picture you have to interpret, every time, with whatever you happen to remember about the deal. The measurement becomes useful when it is joined to an outcome — a signature that has not arrived, a round that is taking longer than usual, a clause that keeps preceding a decline. Those joins are what the assistant does with it.

Diagnosing a stall

"They haven't signed" has several completely different causes that look identical in an inbox. So the diagnosis puts four facts in one picture: when the envelope was sent and reminded, whether the notification email actually left, whether there are viewer questions nobody has answered, and how far into the document the counterparty read.

1. 1

   Never opened it

   That is a delivery problem, not a negotiation. Check the address, check whether the mail bounced, and resend to a human you can telephone.

2. 2

   Read the indemnity twice, then went quiet

   That is a negotiation, and it has a subject. The clause they re-read is the conversation to open, and you are opening it with information rather than with "just checking in".

3. 3

   Has not been asked yet

   They are further down a sequential routing order and the signature in front of them has not arrived. Chasing them is noise; chase the person actually holding it.

4. 4

   Asked a question that nobody answered

   The stall is ours. This is the most embarrassing diagnosis and the easiest to fix.

Is three weeks normal?

Only your own history can answer that, so time to signature is benchmarked across counterparties, kinds of agreement, document lengths and signer counts — and reported with the sample size. "Your median for a mutual NDA with two signers is four days, over eleven envelopes" is a usable sentence. The same claim over two envelopes is not, and should not be dressed up as though it were.

Which clauses cost you time

Reading time is correlated with delay and decline per clause — but as time per hundred words, against each document's own pace. Raw seconds would simply rank the longest clauses in the longest documents. Normalising makes a clause in a two-page NDA comparable with one in a forty-page lease, which is the only way the comparison means anything.

The companion question is where people stop, and whether it is the same place every time. Nine readers stopping inside one clause is a fact about that clause. One reader stopping somewhere different is a fact about that reader's afternoon, and reporting it as a finding would be a way of generating work out of noise.

Scoring a room's visitors, with the weights shown

For data rooms, visitors are scored on breadth, depth, return and commitment. The score is a sorting device, not a verdict, so two things come with it: the weights are published, and the report says what the top visitor never opened. A bidder ranked first who never looked at the customer contracts is a more interesting sentence than the rank.

The question nobody anticipated

Purpose-built reports always run out, usually during the one conversation where the data would have helped. So there is a general query underneath them: a metric, a dimension, filters. Views by day for one room, downloads by visitor domain, median seconds per section for one document over the last quarter.

And every answer — the general one and the specific ones — returns the basis it used and the sample it had. An analytics answer without its denominator is a rhetorical device.

What it adds up to

Measure honestly, and then join the measurement to something that happened. The first part makes the number true; the second makes it worth having. A dashboard that only does the first is a more accurate way of not knowing what to do next.