| Are you in the order processing
business? Do you process ten, twenty or a hundred thousand orders a day?
Do you have hundreds of different order forms out in the market? Are they
more or less the same, yet are they all slightly different? Do you need
to keep track of all sorts of promotions the marketing department is running?
Do you have to process old customers differently from new customers? Are
some of the data on the forms too sensitive to be keyed in by just anybody?
If the answer is yes, you need an automated forms processing system that is more than just an OCR/ICR engine with a tool-kit. You need an integrated solutions that addresses all your needs today and in the future. Architectural Components The following are features you should seek in your solution. First, look at the way the solution is architected and make sure that it is decomposed into multiple components. One sensible way of doing this decomposition is shown in the figure. All work starts with the form definition module where the system is told about the forms to process. There is a scanner where the document enters the system. There is a module responsible for identifying the form being processed The OCR/ICR is performed by an independent module. Based on recognition results, forms are sent to keying. The data validation module checks the consistency of the data; the archiver creates a permanent record of the order; and the extract module translates the order for the shipping department to take over. Keying is performed by people. Their access to the system is managed by a user administration module; their performance is managed with a work-flow management module. A report generation module allows you to create custom reports about the orders you have processed, the performance of your keying staff and the overall system. Last but not least a system management module is needed to manage the overall performance of the system. You should expect to see a full database management system (DBMS) behind the scenes where all the data is stored. You do not want a half-baked database loosing your orders. Companies such as Oracle, Informix and Sybase are working full-time to provide you with better and better DBMSs. Your system should be able to plug in their latest release without much hassle. As for throughout, you should not tolerate any bottlenecks other than that of the database server. If the system is built right, you should be able to use multiple scanners, run your recognition engine on multiple workstations over the LAN, keep employing more keyers and grow your solution with your needs. |
If your solution runs on Windows NT or on Unix operating
systems (ideally both) you can pretty much pick and choose any workstation/PC
in the market to grow your system. Although NT is deemed to be the operating
system of the future, this author thinks that Unix still provides a better
platform for implementing robust and efficient enterprise solutions.
What to Expect Let's get into the details of each module and see what you should expect in each one. Drop out ink is convenient, but order forms are not test score cards; they need not be marketable. The form definition module should work with your forms. Make sure it is well integrated with the recognition and validation modules. Recognition can (should!) distinguish the numeral "1" from the letter "l" more easily if it knows that the field is numeric. You could buy a scanner that can implement most of your solution in hardware. This may work, but it is more like buying a one-piece Stereo/TV/VCR unit; it is hard to replace just the TV when new technology hits the market. You can select scanners that can scan one or both sides of a sheet; some scanners can microfilm images and/or print on the forms; some scanners can handle batches of documents with different sizes. Throughput and image quality are other scanner parameters. You should be able to choose a scanner that meets your requirements and budget. If you have a good solution you can always add more scanners in the future. Think of the hassle of sorting the different order forms. Form recognition can save you a bundle if it can intelligently identify the form. Some systems require that you place a marker on a fixed location on all forms. You are better off with a system that allows you to specify markers anywhere on the document. Recent technology is capable of recognizing a form by the data field it contains. In fact, if you can have this technology, you no longer have to worry about creating a new form definition every time marketing moves a data field an inch. Two forms are equal if they contain the same data blocks. The OCR/ICR module should be capable of implementing voting among multiple recognition engines. Tomorrow, if a new OCR engine hits the market, you will want to get the additional two percent accuracy. You need machine print and hand print recognition as well as mark sense - that is, check-boxes - recognition. There are many data types that are unibiquitous to order forms. If your form definition module allows you to distinguish them, your recognition system can perform better. In fact, for data such as order and customer numbers, you should be able to query a database for valid IDs. A signature data type and basic signature verification should be part of the solution. Chances are you do not need to spend much money on a system that can catch excellent forgeries. Data Entry There is a reason why competitive companies pay keyers by correct keystrokes. Make sure your system has an efficient interface for data entry and allows you to define how to arrange your keying screens. Snippets of data fields, rather than whole forms, are faster to key. They allow you to specialize your keyers and isolate sensitive information on your forms. Zoom-in, zoom-out and controlled access to full image are obviously desirable features. |
You need different classes of users such as keyers, supervisors
and system administrators. You want to:
Data validation uses the information you provided in form definitions and makes sure that the data entered is consistent. Otherwise, it routes it back to keying. Customer service wants to have the order information at their fingertips when a call comes in. Even though disk prices are coming down, chances are you need a multi-tier archive system that moves aged orders off to slower and cheaper media. Finally, you want to handle that special promotion where one out of every 50 customers who bought a toaster gets a free loaf of bread. Specifying a query that performs the task with the report generation tool had better be possible. The Payoff Solutions that meet the above requirements do exist. Fingerhut just purchased one from Oyster Software. You may not need every feature mentioned above, but make sure you have the full list that you require, then call solution providers such as Dakota Systems, Oyster Software, Symbus or Wheb and see what they can do for you. You can pay for your solution in several ways. You can buy it outright, in which case the price may depend on the throughput of the system as well as the different features you want . If you are buying it outright, arranging a lease agreement should not be a problem . Alternatively, you should be able to pay for your system as you go along. In this case, you pay a transaction fee for each page the system processes. The terms should include minimum and maximum processing volumes. The bottom line is your solution should pay for itself in less than two years. Don't settle for anything less.
|
|
|
|
