AV (Automated Valuation) is the second of our four topics in the discussion on The Big Four:  AI, AV, DS, and BS.  (Read the earlier blogs here.)

Why these four topics?  Because they cover the whole of the mechanical part of appraisal.  It is my position that when good mechanics are in place, issues of integrity, ethics, and even bias come clear and visible.

This series is intended to encourage new thinking about old topics — in particular, as the Appraisal Foundation is currently pushing forward, seeking updated ways.  This includes how updating of appraiser qualification knowledge with currently available, established data analytic technology.

Automated Valuation Models (AVMs) are the business monetization of models carried out by algorithms (logic and instruction set).  “Automated” here means carried out by computation without direct human control.  Automated things are algorithms.  Models require decisions, by someone!

AV is any part of an analysis which is carried out by computation.

AVM is the industry.

This is the core of our discussion here:  While there is no human control within an algorithm – some human must make the decision of what/which model is the basis for selecting that algorithm. (Or any needed tweaks!)

In part 1, we noted that all systems of valuation must follow the same simple basic steps:  1) Problem ID; 2) Data selection; 3) Prediction/estimation; and 4) Delivery.  Identical.

What is different, for each of the four steps – is who does it, and how well is it done.

The competing trade-offs are a) reliability, b) cost, and c) speed.  (A fourth constituent might be additional results or services.)

Following our thinking above, we can see that “AV” can simply be a stand-in word for the algorithm applied.

So how does AV apply to a commercial AVM?  First, we set out step-by-step parts:

1. Problem ID – This decision involves several judgments.
   • Property type
   • Depth (scope) of work needed
   • Reliability (credibility) level desired
2. Data selection
   • What is similarity
   • Overall data frame
   • Selection algorithm(s)
   • Data sufficiency levels
   • Outlier and influential data pints handling
   • The level of data wrangling needed, including confirmation and verification
3. Prediction
   • Which of the three “adjustment” algorithms are applied
   • Select the source for market trend (or data for the price index)
4. Report
   • Mechanism for delivery
   • Level of interaction by the user
   • Level of interpretation of analysis
   • Additional result information
   • Level of explanation, to suit user competency

This detailed view of “automated valuation” shows the importance of the expert’s role in today’s and future valuation processes.  Someone makes these modeling decisions – whether the independent analyst, or the user, or the original computer programmer – by writing code and making decisions.