Impact Resources: I started in 1989 when they had outgrown their first office condo. I had to work from home for a few weeks as there was no room for me in the office. I had my trusty PS/2 Model 30 and I was ready to roll.
The company did Consumer Behavior Research by means of surveying people in shopping malls and strips. We collected 400-ish data points per survey and paid an entire dollar for your time. Things like where do you shop for X? Do you prefer value, quality, name brand? What radio station do you listen to? What type of music do you like? What kind of car do you drive? And how much money do you make?
All these data points were hand keyed into PC’s and stored on floppy disks. (We didn’t have fancy LANs at the time). The data from the disks was loaded onto a large (300MB) hard drive and summarized. There were several steps involved in “normalizing” the data and until I was hired, that process involved putting the data on tape, sending it out for processing and restoring from tape when it was returned. It usually took about 2 weeks to turn this around.
When it was all done, you could do some pretty cool queries. You could, for example, show that although the pop radio station was Arbitron #1 in every time slot, it was listened to by 17 -24 year old girls with $0.50 of disposable income per week. Meanwhile, the country station attracted 30-60 year olds with average income > $100,000 and lots of disposable money. As a business owner, you could find out why people shop, or don’t, at your business and update your messaging. Then find an avenue to reach them. As a media outlet you could find your listeners and craft a sales pitch for local businesses to advertise.
We did the top 50 Metropolitan Statistical Areas in the US, so 50 markets to sell local results. Then we normalized those results into a Nationwide view in a product call MART USA. MART: Market Audience Research Traffic.
The “secret sauce” for MART was the adaptable weighting of response records. If you selected a subset of the data, is was weighted on-the-fly to more accurately represent each record. In my testing, the reality was that you could pre-weight the records (weighted averages based on demographic data) and be within 0.05% of the on-the-fly method. The weighting method was built into the query tool and it was written by some Professor in Utah. He wrote it in Pascal, which was very in vogue at the time. Since it was written in Pascal, the entire data refinement process had to be as well. (It didn’t, but the founders didn’t know that.)
So I wrote a series of programs that took the data thru several stages of transformations. I had a read a schema from disk, then read a record and apply as many changes as it required. Most of these were things like change your store of choice from “Moms attic” to “other”. Sometimes we redacted income, if you’re 17, you probably don’t make $2M / year.
We were able to reduce the processing time from 2 weeks to 2 hours. Not bad.
In addition to developing the database processing, I got to play network admin. We bought a “Big” Server and installed Netware 2.15c. It was a forward thinking version as it came with both 5.25″ and 3.5″ floppy disks. We spent about $10k on the server as it had 8MB RAM and 2 300MB SCSI disks. By the time I left we had 2 servers, 1 Dev and 1 Prod. Each had 3x 300MB Disks for nearly 2GB of data.
What happened: We got a buy out offer from Advo (https://www.referenceforbusiness.com/history2/5/ADVO-INC.html) and everyone was pretty excited. The weekend before the final signing, the Venture Cap guys decided to dilute the founders stock and fire them. When the Advo team showed up to sign, they asked about Gary and Phil. When they were told about the firings, Advo pulled the deal. 6 months later, Impact Resources was out of business.
I was able to jump out before the collapse and ended up at a Second Startup.