[thelist] Best Practices for building a zip/postal code based dealer locator

[.jeff]

bryce,

 ><><><><><><><><><><><><><><><><><><><><><><><><><><><><><
 > At 04:50 PM 9/26/2001, you wrote:
 >
 > I have been tasked with coming up with posible
 > solutions for building a dealer locator that is
 > to be searched by postal code.
 >
 > Can anyone who has built something like this share your
 > experience on the best practices on how to go about
 > building this and pitfalls to watch out for would also
 > be greatly appreciated. If anyone has any boxed
 > solutions they can recommend that would also be great.
 ><><><><><><><><><><><><><><><><><><><><><><><><><><><><><

i'll start off by showing you the one i rebuilt recently.

http://www.burley.com/html/locator/index.html

this locator uses latitude/longitude to determine proximity.  for 
simplicity purposes it actually measures it based on a square with sides a 
selected distance from the center point, rather than a radius.  for the 
most accurate results you'd want to do the extra math to do it based on an 
actual radius.  however, if you're searching for proximity in the range of 
mile-plus and not actual distance, the square method is close enough.

the misleading part is that the latitude/longitude are the physical 
location of the post office serving the zip code selected.  this isn't 
likely to be the center of the zip code area, won't indicate the size of 
the area the zip code covers, and won't reflect the shape of the area the 
zip code covers.  this is important to consider if your audience for this 
locator aren't likely to know the region they're looking in -- perhaps a 
motel location or atm location search.  it's not so much of an issue if 
they're likely to know the area very well.

all that aside, the important thing to make sure to do when building a 
latitude/longitude-based solution is to make sure your search engine is 
always using up-to-date information.  you'd be surprised how often zip 
codes, latitude/longitude, and other factors change.  on average there are 
about 2 dozen changes each month.  the solution is to outsource this 
information gathering/management to someone else.  for us-based zip 
searches, we use jay's computer services (http://www.jcsm.com/) for our 
data.  it comes to us in a multitude of formats and is reasonably priced at 
about $250/usd per year.  according to their site they also do canadian 
postal codes as well.  if you need to do this for other parts of the world, 
you're on your own.

with latitude/longitude information it's fairly easy to find other zip 
codes within a given range.  then, take that list of zip codes and search 
against the zip codes of your dealers adding any additional criteria like 
product they carry, product in-stock, etc. and display that to the user.

one word of caution -- doing a wildcard search using the first 3 or 4 
digits of a zip code is *nowhere* near accurate or relevant.  zip codes are 
not handed out in any sort of logical manner.  you won't have 97116 next to 
97115 or 97117.  neither will 97116 be as close to 97100 or 97199 as you 
might hope.  here's an example using the zips noted.  all miles are 
point-to-point and do not reflect the actual driving mileage between points.

97116 - forest grove, oregon

97117 - gales creek, oregon
         6.77 miles from 97116
         5.66 miles south of 97116
         3.73 miles east of 97116

97115 - dundee, oregon
         31.94 miles from 97116
         27.91 miles south of 97116
         15.60 miles east of 97116

97101 - amity, oregon (97100 doesn't exist)
         40.39 miles from 97116
         39.99 miles south of 97116
         5.72 miles east of 97116

97149 - neskowin, oregon (this is the highest 971 goes)
         46.97 miles from 97116
         39.27 miles south of 97116
         25.90 miles west of 97116

and a couple more to show the problem it creates:

97114 - dayton, oregon
         36.05 miles from 97116
         33.70 miles south of 97116
         12.88 miles east of 97116

97118 - garibaldi, oregon
         27.74 miles from 97116
         6.58 miles south of 97116
         26.97 miles west of 97116

97113 - cornelius, oregon
         14.66 miles from 97116
         4.00 miles south of 97116
         14.11 miles east of 97116

97119 - gaston, oregon
         13.92 miles from 97116
         13.50 miles south of 97116
         3.41 miles east of 97116

97112 - cloverdale, oregon
         40.06 miles from 97116
         29.30 miles south of 97116
         27.42 miles west of 97116

97121 - hammond, oregon (97120 doesn't exist)
         47.96 miles from 97116
         36.95 miles north of 97116
         30.44 miles west of 97116

97111 - carlton, oregon
         27.04 miles from 97116
         26.04 miles south of 97116
         7.31 miles east of 97116

97110 - cannon beach, oregon
         33.72 miles from 97116
         15.48 miles north of 97116
         29.90 miles west of 97116

97140 - sherwood, oregon
         31.82 miles from 97116
         21.99 miles south of 97116
         23.06 miles east of 97116

97146 - warrenton, oregon
         42.40 miles from 97116
         31.90 miles north of 97116
         27.81 miles west of 97116

these are relatively close in numeric sequence, but in exact opposite 
directions from 97116.

the odd thing is that 97062 (tualatin, oregon) and is right next to 97140 
(sherwood, oregon).  it's only 34.78 miles away from 97116 (forest grove, 
oregon) or 5.12 miles from 97140 (sherwood, oregon).  actually, that's how 
far apart the post offices are.  in reality, the edges of the two towns are 
right next to each other -- leave one as you're entering the other kind of 
thing.   you have a dealer located on the south-east side of sherwood and 
another on the north-west side of tualatin and they could only a few miles 
apart.  however, 97062 won't be returned in a '971%' wildcard search in 
favor of zip codes that are as much as half-again as far from 97116 as 
97062.  this inaccuracy is heightened by broadening the search to only 
include the first two digits of the zip code.  what do you do for zip codes 
that currently have alot of holes in the sequences?  finally, using a 
wildcard search it is impossible to give the user an idea of the distance 
from the originating point, even only as an approximation.

good luck,

.jeff

http://evolt.org/
jeff at members.evolt.org
http://members.evolt.org/jeff/