INTERVIEW: Louis H. Ray, the practical coder

Most of my career has been in information technology, although recently it's also been in engineering and logistics services. The most important lesson I've learned in software development is that functional knowledge'understanding the customer's business in detail'is more important than technical IT knowledge. If you don't understand the business, how can you be sure you're solving the right problem?Most customers are not perfect at describing what their problem is. They can describe where the pain is, not the problem. I don't believe you send IT people to determine it. I think you have to team IT people with people who have knowledge and experience. They have to work on teaching each other how to talk because they don't by nature communicate very well. It's a team-building, personnel management and education process. Where we're successful, it's very powerful. We're not always successful. Depending on the level, it can be two or three or four months, or two or three years.For example, we did a major redevelopment for the Agriculture Department's Risk Management Agency. We sent in a senior program manager, and he spent the better part of two years in joint application development sessions. The agency has a complicated business'insurance ratings, underwriting, actuarial processes. Little by little, we developed a complete client-server application that replaced all the mainframe systems.A recent job we did for the Risk Management Agency was to create a data warehouse and develop fraud identification techniques. That's a work in process. As you find and close a door, someone finds a new one. You have to use artificial intelligence; there isn't enough of the real thing.After we find a potential problem, we look for data anomalies to find out if they constitute fraud and a mechanism to close it down. If there isn't, we implement techniques to monitor and report on it. More often than not, customers already are into platforms. We're not particularly concerned with the platform. We do a lot of comparative analysis of software packages and cost-benefit trade-offs. Where a new solution is required, we'll try a prototype.We like to test up front, not at the end. We're radically opposed to large-scale development efforts. The failure rate is way too high. And it's going to get worse because as the world speeds up; by the time you actually get a solution, it's solving the wrong problem. The problem has changed underneath you.If you can't deliver something useful in three to six months, you're tackling too big a piece. You need to make regular deliveries at small intervals. You can't afford to freeze specifications'the world that drives the specifications is moving too fast. And you have to stay close to the customer during development. If you turn your back, his problem changes and you're irrelevant. DLA was charged with taking over all the military supply depots. We analyzed the software that was running at the different depots and picked what was most modern and easily maintained, which was the Army's.We implemented that at the largest DLA depot, in New Cumberland, Pa. Then we began incorporating the service-unique requirements, and as we completed them, they went out to the depots.The primary functions are receiving, stowing, picking, packing and shipping. There are unique automated handling systems at each depot'conveyor belts and supply carousels.So we had DLA-unique plus service-unique plus depot-unique systems, all of which had to be integrated into the [$15 million] Distribution Standard System. We put together a team of people with lots of experience operating depots, and we paired them with computer types who had experience with IBM Corp. mainframes and Computer Associates International Inc.'s CA-Datacom running under CICS. The lower-tier material handling stuff was VAX-based C++.The difficult problem was the dozens of interfaces to DLA and service information systems for ordering, logistics and shipping. All this had to be maintained while we were incorporating the new requirements and installing. We got the last depot installed late last year. The Federal Retirement Thrift Investment Board, which manages the Thrift Savings Plan, had an IBM mainframe with CICS and CA-IDMS maintaining the records. Employee access was via a voice-response unit.What we did for the board was develop a front end so users could come in with Web browsers. We did it quick and dirty and cheap [$125,000] with Amazon Integrator [from Intelligent Environments of Burlington, Mass.], which let us generate CICS queries and repackage the response screens and reformat them for active Web pages. We got that up and running in a couple of months.It's currently receiving about 1.3 million queries a month, and the data is only updated monthly.For the Justice Department, we developed a multimedia training package called Safe Schools to help teachers and school administrators recognize various risks of violence. It runs through scenarios, records the responses and evaluates them against an intelligence database to see how effective they are likely to be. We expect to complete that [$400,000] Small Business Innovation Research project in the next two or three months. We've done workflow automation for the National Institute of Standards and Technology. It requires digital certificates for legally acceptable sign-offs. We're using BizFlow 2000 from HandySoft [Corp. of Falls Church, Va.] and Entrust/PKI [from Entrust Technologies Inc. of Plano, Texas]. So far we've automated more than 200 forms for 3,000 employees in administration, contracts, wages and salaries. The forms are routed around the country for completion, approval and tracking.For the Securities and Exchange Commission, we did a project related to year 2000, but I'm proud of the fact that we never did Y2K work. We didn't get involved in a bit of Cobol.We developed an optical character recognition system to allow companies to report to SEC on their Y2K status. We accepted forms, OCR'd them and built a preparedness database for SEC and the White House. We used Teleform from Cardiff [Software Inc. of Vista, Calif.] and managed the whole process including OCR, verifying the database and doing error correction.I think we got the first round of data online in 60 days. Having the right commercial software makes it a lot easier.