Information foraging, and why SYNC reads before it writes

The short version.

Information foraging is a theory from Xerox PARC about how people look for information. It says they behave like animals looking for food: they follow a scent, and they leave a place when it stops paying.^[1][2]

We use it as a way to think about researching an account. A source is a place to search, a dated fact is the scent, and stopping is part of the method. This is an analogy of ours. We found no study that applies the theory to sales.

Our own records show where the scent runs out. The web search step of the research found a signal in 283 of 2,529 research results. Fewer than half of the dossiers carry a date we could read, and hiring is the most common kind of case.

What the theory says.

Peter Pirolli and Stuart Card developed the theory at PARC in the 1990s.^[1][2] They borrowed from biology. An animal cannot search everywhere, so it goes where food is likely and moves on when a place gives too little for the effort. A person looking for information has the same problem and tries to get the most value for the time spent.

Three terms carry the theory. A patch is a place that holds information, such as a website. Scent is the cue that tells the searcher whether a patch is worth the visit. A later paper defines it as the searcher's sense of the value and the cost of opening a page, judged from what is visible before opening it.^[3]

The third term is leaving. The theory treats the decision to stop as a calculation, made again and again during the visit:

How it maps onto researching an account.

The closest use of the theory that we know of is in intelligence analysis. There it describes a loop of searching, filtering, reading and extracting, which feeds a second loop in which the analyst makes sense of the material.^[5] We know that paper through other sources only. Account research has the same shape, and the mapping below is ours.

• 01A source is a patch. A careers page, the company's own site, a register entry, a press article. Each costs time to read, and most hold nothing a seller can use.
• 02A dated fact is scent. A vacancy posted last week or a funding round announced this month tells SYNC that there may be a reason to write, and where to read next.
• 03Stopping is part of the method. When the sources give no dated fact, the right result is that the account waits. A search that never stops will always find something to say.

This is why SYNC reads before it writes. The message depends on what the reading found, and an account where it found nothing should get no message.

Where the scent runs out.

Most searches find little. The web search step returned anything at all in 1,096 of 2,529 research results (43%), and at least one signal in 283 (11%).^[6] That step was switched on for some runs only, so the shares mix periods with and without it.

Dossiers are written for accounts that scored well enough to be worked, so these 1,303 are a selection. The figure of 566 does not say that 43% of companies have a dated reason.

One finding goes against our own rule as it stood. When we measured, this site said that an account without a dated reason gets nothing from SYNC. Of the 570 dossiers marked ready, 303 were judged ready on a signal with a source and a date. The other 267 were not: 230 were judged ready on a signal with a source and no date, and 37 have no reason recorded.^[6] Ready means that the account may be written to, after a person approves the message.

Today the check asks for a cited signal and its source, and does not require a date. This was a choice made when the check was built, because requiring a date would have left most accounts waiting. We decided to keep this: a cited signal without a date can still make a dossier ready. The reviewer sees that the date is missing, and we are making that more visible in the review screen. The wording of the rule on this site now says a reason we can cite, not a dated reason.

The count of 334 above is a different count. It looks at whether the case carries a readable date today, and the 303 is the reason recorded when the dossier was judged. The two differ because a case can gain or lose a date after the verdict was written.

For the 334 ready dossiers with a readable date, the signal was a median of 8 days old when the dossier was judged. A quarter were 25 days or older, and a tenth were 235 days or older. These ages describe what is stored, and they do not show that the signals are fresh. For 99 dossiers the age is zero, which usually means a vacancy read on the day of the research: it tells when we looked. For 27 the age is negative, which means the date lies after the day the dossier was judged. And 50 are older than 60 days.

Nobody checked the dates against the source. A date is what the page stated or what the model read from it. Each case also carries a label for strength, and we do not report it as a measure. Of the 516 cases labelled strong, 515 have a date, against 27 of 256 labelled moderate and 24 of 280 labelled weak. The label mostly repeats whether a date exists.