WhoshouldIsee Tracks

Automated Valuation Model (AVM)

Find out how we maximise the accuracy of the Homeflow Valuation tool

The secret behind Homeflow Valuation’s accuracy

Homeflow Valuation is built on a “Automated Valuation Model” (AVM) which is statistically more accurate at producing valuations in aggregate.

What does this mean?

Terraced street

Imagine a long terraced street, with numerous near identical house, many of which have sold in the past year. Every Instant Valuations Tool on the market will nail a valuation request within a few percent.

Now imagine a unique property in a sparsely populated area, where few homes have been sold.

In this case, Homeflow Valuation has a magic little extra step which both gives more accurate valuations and wins the confidence of the customer.

Once the customer has entered their contact information, Homeflow Valuation displays 6 similar properties that are currently on the market and asks the customer to “Select any which are similar” to their own.

First, this improves accuracy – if they select higher valued properties, we adjust the estimate upwards. If they choose lower valued property, the estimate is revised down. This reduces disappointment and avoids building up unrealistic expectations.

Second, this involves the customer in the valuation process. By getting their input, the customer is more likely to trust the valuation.

If you’d like to know more, ask for a demonstration of the science – we have test graphs across 2000 properties showing Homeflow Valuation’s valuation accuracy.


Here is a brief overview of how the Automated Valuation Model (AVM) works:

  • It gathers the postcode, then tries to identify (through API look up and form address questions), the exact property.
  • It also gathers the bedroom count
  • It pulls back all the historical sales data within 5 or so miles from Land Registry
  • If it can latch onto the specific property as having been sold in the last few years, then it will place a LOT more confidence on this data point, but uses some influence from other properties too
  • Otherwise it will select a group of sales, from nearby properties
  • It rejects outlier data points (a key statistical step, that increases accuracy, not all competitor tools do this)
  • It has a couple of other statistical modelling tricks that we don’t disclose, as we are ahead of our competitors in this regard too
  • It then uses regression analysis modelling to come up with a valuation
  • It puts more weighting on more recent sales data than older data
  • It then projects all those data points for the various years, forwards, by the relevant house price growth indicies
  • And it then applies some adjustment variables for bedroom counts, depending on how tightly packed the datapoints are
  • And finally it applies some bespoke tuning variables which you can set at your own agency level (some clients prefer to systematically overvalue the AVM values a bit, others to systematically undervalue them
  • And then it presents the user with some “similar properties” currently on market in their area – it picks a few above the AVM estimation, and a few below. This step involves the vendor prospect INTO the process of creating the valuation – it makes them complicit (so they share responsibility), as well as adjusting the value it finally spits out to make it more accurate. This step is unique to Homeflow.
  • It also gives you something more to talk to the vendor about when you pick up the phone.
  • And finally it spits out a confidence interval… was it working with tight, recent data, and the actual house (high confidence), or scant, old data, and no latch to the specific house (low confidence). LATER we will be adding this to our handling of the response, backing away on low confidence, tightening the range with high confidence.

Related articles


Read what our customers say about us

See our website templates

Get an industry leading website, fast.

Sign up to our newsletter