Models

The Tendril Analytics Tool includes proprietary models developed to be used in conjunction with the raw data, further enhancing its value. These models - as briefly described below - allow the user to go far beyond what is achievable via population segmentation alone, and are tailored to address issues commonly faced as energy providers look to expand their interactions with their customers.

Existing Models

Weather Sensitivity: Heating & Cooling Days

The Tendril Weather Sensitivity Model is a heuristic which examines how a dwelling’s electricity consumption varies with weather, yielding a score each for sensitivity to hot and cold weather, respectively. The basis of this model are the bills spanning a year-long period for a given dwelling, combined with hourly-resolution data for the zip code in which it exists. This particular model does not attempt to determine the cause of a dwelling’s sensitivity to weather. For instance, an older home with poor insulation may have the same score as a newer home occupied by residents who prefer the interior temperature to be abnormally cool in summer and warm in winter.

Propensity: Smart Thermostat

To offer a single field that captures a dwelling’s occupant’s openness to smart thermostat technology Tendril has leveraged its extensive experience in energy efficiency programs of all types, in particular the Orchestrated Energy product. This database of program performance is combined with Experian demographic and physical dwelling information in a supervised machine learning algorithm to produce a score, which is then conveniently binned into leveled categories. Smart thermostats are one of the most cost efficient energy savings technologies available, and this model is designed to help get them in the hands of people who will install them in their home.

Seasonal Home Likelihood

The Seasonal Home Likelihood models builds upon the weather sensitivity model to develop a heuristic that identifies dwellings which are occupied only part of the year. By examining how the energy consumption of a dwelling varies over the course of the year, as well as how that consumption compares to other dwellings in the same region, periods of occupancy can be differentiated from when the dwelling is empty. Because of geographic differences in seasonal home patterns, the model is also calibrated on a region-by-region basis producing a more reliable prediction. Note that because it relies on a full year of billing history newer customers may not receive scores.

Upcoming Models

AMI Predicted High Bill: Based on a user's consumption so far, do we predict their bill will be higher for the upcoming month?

AMI Unusual Usage: Which homes have abnormal usage curves, ie they peak more than others in the AM or PM?

Digitally Engaged Users: Which users respond well to digital channels?

Demand Response Propensity: How likely are users to enroll in a demand response program?

Electric Vehicle Detection: Which users seem to have electric vehicles?

Propensity: Electric Vehicle Purchase: Who is likely to purchase an EV?

Propensity: HVAC Service/Upgrade: Which customers look like they need to service or upgrade their HVAC system?

Propensity: Marketplace Purchase: Who is likely to purchase an item from a marketplace?

Propensity: Smart Speaker: Which customers are likely to purchase voice technology, such as the Amazon Echo?

Propensity: Green Energy: Who is likely to participate in a community solar or wind energy program?

Propensity: Solar: Who is likely to purchase solar panels?

Propensity: Utility program participation: Who is likely to participate in "other" utility programs (fridge trade-in, insulation, rebates for "x", etc.).