The correspondence problem
The diversity and volume of each representatives’ incoming messages have greatly increased in the lifetime of our republic. Since 1789, the number of constituents have increased 20-fold per House district, and roughly 100-fold per Senate district and for the White House. Digital technologies have reduced the costs of mass communications from representatives to constituents, and they have reduced the personal costs to constituents of sending messages to their representatives.
This is the correspondence problem: how can a representative effectively process and respond to communications from her constituents? The fundamental difficulty of understanding and responding to the many-to-one communications from constituents have not changed. Four things have always made the correspondence problem difficult:
1. Translating:
In general, every constituent will have her or her own mental image of how the government functions. So does everybody in the government. Somehow the thoughts and wishes of the constituent must be translated into some description of what the constituent is saying in a language that the representative and/or staff can understand.
In Chris Casey’s classic, The Hill on the Net, there is a picture of a postcard sent to Senator Kennedy’s office from an angry constituent, saying simply, “If you vote NO, I will remember you”. There is no way for the Senator’s office to know what vote this refers to or even if the card is expressing a positive or negative sentiment toward that vote.
No matter what technology is used to transport the message, the representative and staff are left to divine the meaning the constituent intends to convey. As with any communication, the author puts together some message, the recipient reads (or watches the snapchat, or whatever) and comes up with some belief about the meaning of the message. Sometimes the intended message and the received message are similar, and sometimes there are misunderstandings.
As citizens we can sidestep the translation problem by working in teams. Sending the message recommended by an advocacy group (such as https://5calls.org/) gives us the advantage of combining our efforts to submit multiple messages that were developed by professionals to convey what we (presumably) intend.
2. Categorizing:
Get Eric Loeb’s stories in your inbox
Join Medium for free to get updates from this writer.
Once the constituent’s message is understood, representatives have always had the subsequent need to perceive the message in the context of the broader constituency. Is this an isolated request? Are people knocking down the doors clamoring for the same change?
The perceptions of the representative are routinely gamed by lobbyists and organizers as they find ways to make their messages more salient to the representative. Jamming the phone lines with 100 calls is more effective than sending 10,000 emails. They notice the phone lines but they just laugh at the IT guy. Or so I hear.
Categorization can be done intuitively, but with large enough message volume (i.e. at most if not all US Congressional offices) categorization is managed by Correspondence Management Software (CMS). Leading CMS systems use keywords to match constituent messages to established categories. For example, the translation function may determine that a constituent email has asked for a yes vote on some issue and requested the representative’s attendance at the local high school graduation ceremony. To correctly manage the message, these sections of the email need to be categorized into the two pertinent buckets.
Categorization is needed for automated or semi-automated constituent responses and it is also often used to provide the representative with a snapshot of the issues concerning the constituency. Current CMS systems are deeply inadequate for informing the representative. So far as I am aware there is no CMS package that compares messages on a given topic to the projected rates derived from other sources (e.g. twitter traffic) and the political demographics of the constituency. The representative may learn that the American Health Care Act had the largest number of calls, but not that those calls were lower or higher than expected given other predictive factors. The representative’s natural human tendency to incorrect reasoning is exacerbated by standard CMS systems.
3. Responding:
It is difficult to respond to constituents’ communications in a way they will appreciate. Standard CMS systems map a communication to one or more categories for which there is some prepared response ready. However, constituents will not feel heard if there is no human element. A purely automated response will almost invariably be recognized as such.
To make matters worse, constituents expect to receive a response in the same medium they used. If they call they expect a person to answer; email should answer email; letter should answer letter. But what if they sent a text message or a video. Is the office prepared to respond in kind? Should it be?
Customer interactions should be viewed as opportunities. Constituent contacts should be no different. The social good in a constituent contact is a quantum of citizen support. “Somebody gave a damn! I’m in!” In my view, this means that responses to constituents should include paths to further involvement in whatever issues they have shown interest in. Unfortunately, representatives are in a bad line of business: more customers means more costs and no additional revenue.
4. Re-conceptualizing:
Constituent requests will sometimes be completely off the wall. They may want the representative to fix some problem that is an obscure function of law, regulation, culture, and economics. They may ask to change “the law” to do something, where there is no single law governing the matter. They may bring up new issues or maybe even provide some great ideas. How can representatives’ offices expand their staff+automation interface to the public to respond to off-the-wall ideas, new issues, and/or great ideas? Is this simply a matter of expanding the taxonomy of issue categories that drive the available constituent relationship management systems, or is there some fundamentally better approach?
Next: How does AI help with the correspondence problem?
