Though I’ve posted on this topic recently, I’ve had requests for a more specific look at how to start and maintain data today for future analytics endeavors.  So, some tactical and specific advice:

1) History: How far back?  It’s important to balance three competing factors:

1a) Observations: Your stats guys will always want more observations because it allows them more precision in their results.

1b) Change:  At some point the world is different enough that any results based on that world will be irrelevant.  How fast your world is changing in your corner of the world is up to you, but 3 years is a reasonable minimum.

1c) Of course, Cost: Though storage costs are falling extraordinarily quickly, there are still costs in setup time, systems work, perhaps consultants, etc.

2) Granularity: In this case, the stats guys are right: more granular observations are almost always better.  Unless something like cost is getting in the way, store it as granular as possible.

2a) Geography: Obviously keep it only at the level in which it’s relevant to your business: if you’re an online-only retailer with no regional pricing or advertising differences, there’s no need to keep geo data based upon IP.

2b) Product: Rarely a bad idea to keep as granular as available, and even invest the time in mapping data to the products.  For marketing or advertising sources, involve your agency.

2c) Channel: If you have more than one channel, segment, or set of customers, keep the data mapped to those levels.  Don’t be afraid, by the way, of using codes like “All” or “None” if it makes sense.

2d) Time: Unless you’re doing very detailed online tactic optimization, no need to keep anything deeper than daily, and in most cases weekly is just fine.

3) Measures: Ok, so what data sources should I keep?  For lack of a clean-cut answer, anything that matters.  “Matters” could be defined as having a reasonable expectation of moving the needle in your sales by 1% or more during the time it was executed.  Because that’s likely an unsatisfactory answer, here are some examples:

3a) No need to keep it:

i) An out-of-home billboard in Tucson: Not likely, unless you’re going to make billboards a larger part of the portfolio
ii) The iPhone app with 2,000 downloads: Not likely large enough (unless it’s direct response and is driving significant sales)

3b) Yes, keep it:

i) The 3-day inventory clearance event we held at the end of last season: Yes- price moves the needle.
ii)  Website layout or navigation changes, or even changes to the IVR system at a call center: Yes, these drive surprisingly large changes
iii) Anything done by an agency:  Yes- you’re spending on it, keep it, as relationships (and agency contracts) change over time

Hope this helps.  And when in doubt, don’t hesitate to contact your analytics partner.