Wireless and wired telecom providers use geocoding and location analytics for network planning and subscriber acquisition. Location data becomes even more critical during 5G deployments. For example, for fixed wireless deployment using millimeter wave (mmW) bands, which provide extremely high speeds and capacity in focused areas, operators must accurately know what the best locations are for deploying 5G small cells. Just as important is knowing what the customer locations are that will be inside the coverage area.
How well it works depends on hyper-accurate geocoding and a rich set of data attributes augmenting a correct residential or business address. Typically, however, engineering, marketing and customer service use siloes and inconsistent in-house datasets, myriad data products and formats, and disparate analytical tools.
The result can be costly mistakes across a range of use cases, including misplaced 5G infrastructure that blocks signals and investments in neighborhoods that aren’t a good market for 5G services. For cable operators looking to expand network penetration, errors include inefficient routing of fiber drops and overlooked occupants of near-net buildings. In this webinar, we will examine a different approach to linking latitude/longitude with an accurate address and data attributes such as building footprint, height and occupancy as well as neighborhood-level demographics.
Topics covered in this presentation will include:
- Challenges CSPs face today in using data for network planning, including new 5G infrastructure and optimizing existing wired networks for both end users and backhaul customers
- Understanding and implementing accurate address and geocoding data across multiple use cases, including customer acquisition
- How a Master Location Data approach can efficiently augment lat/long and address data with detailed attributes needed by engineers and marketers
4. has never mattered more.
Where to place towers
and how to structure
macro and micro cells, and
to assign data,
5. It doesn’t take much to
figure out what
happens if it goes
wrong.
5
Are you building in
the right place?
Are you marketing to
and servicing the
right people?
Are you properly
provisioning the
service?
9. Superior understanding of
your customers helps you
to identify and engage
ideal customers and
prospects, learn their
behaviors, in the places
that matter to them
10. A location-centric approach to MDM puts
network providers in control.
03.
Operationalize
your addresses
Use the ID to build a
contextual view of a location
for better insight.
02.
A trusted ID
Assign a trusted ID that is
unique and persistent to each
address.
01.
Precision
Geocoding
Achieve the highest level of
address integrity and
positional accuracy.
12. Master Location Data with PreciselyID
WHAT IS IT: Best-in-class Address Point geocoding dataset that is built, owned and maintained by Precisely
WHY IS IT SO ACCURATE: HIGH MATCH RATE + BEST IN CLASS POSITIONAL ACCURACY
Build Data
Merge
De-dup
Standardize
Link to our data
Build linkable data
Pre-score the country
Discover additional
addresses
Manage data
synchronization
Data Transfer
55% Building Centroids
(or better)
38% Parcel Centroids
3% Interpolated
Locations
4% Backfill Centroids
207m Addresses
50m Secondary
Addresses
193m
Addressable
Locations
(PreciselyID)
Support for
POI’s in Address
Matching
Assign
PreciselyID
13. Matching: How well does your geocoder
handle incomplete input data?
13
Mesilla Valley Mall 88011
Input Address Output
700 S Telshor Blvd
Las Cruces, NM 88011-4669
W399S6012 CRZ WI53118
Input Address Output
W399S6012 COUNTY ROAD Z
Dousman, WI 53118-9543
1053 Thornton Lake 97321
Input Address Output
1053 NW West Thornton Lake Dr NW
Albany, OR 97321-1352
1 100 #1 10025
Input Address Output
1 W 100th Street, Frnt 1
New York, NY 10025-4857
14. What point does your business demand?
Unit Address Point
Building Centroid
Parcel Centroid
Street Centroid
Street Interpolated
16. IDs are better than multi-field labels
4750 WALNUT ST STE 200
BOULDER CO 80301-2532
Developer & Local Addressing Authority
Local Addressing Authority & USPS
Developer & Building Owner
Building Owner
USPS & City Government Federal & State Governments USPS
Let’s dissect this addressAn address is a multi-field labelling system that is
utterly unsuited for operationalization and automation.
Each field is owned and updated by different entities.
17. We’ve assigned a Precisely ID for every address
so you can stop relying on complex multi-field
addresses.
Remember this slide?
17
Mesilla Valley Mall 88011
Input Address Output
700 S Telshor Blvd
Las Cruces, NM 88011-4669 9S6012 CRZ WI53118
Input Address Output
W399S6012 COUNTY ROAD
Z Dousman, WI 53118-9543
1053 Thornton Lake 97321
Input Address Output
1053 NW West Thornton Lake Dr NW
Albany, OR 97321-1352 1 100 #1 10025
Input Address Output
1 W 100th Street, Frnt 1
New York, NY 10025-4857
P0000G4I6A4F P0000PB9B3L2
P0000GL1BBGUP0000IVQH0XI
18. Trusted IDs
are essential to
operationalizing
addresses
Operationalized addresses have been:
• Cleaned, standardized, and validated
• Precisely located with geographic
coordinates (geocoded)
• Assigned a unique, persistent ID
• Enriched with additional information
19. PreciselyID
enables data
automation
that adds
context and
improves your
down stream
analytics.
PreciselyID
ID Lookup
Extended
Attributes
Parent /
Child
Type-
ahead
Hard to
locate
addresses
Connected
Data
Portfolio
Address
Fabric
20. • All of the best address locations for use in any
database, business intelligence, or location
intelligence platform
• This dataset is built from our Master Location
Data and contains accurate and
comprehensive address locations.
• Each address record as been pre-verified, pre-
validated, pre-standardized, pre-geocoded,
and previously assigned a PreciselyID.
• Additional attributes include property type,
parent/child relationships.
Address Fabric
23. Straight-though-Processing for Operationalizing Addresses
Address
Geocoding
Interactive
Geocoding
(Type-Ahead)
Reverse
Geocoding
Operationalized Address
• Cleaned / Standardized (postal
format)
• Validated (deliverable)
• Geocoded
• PreciselyID added
Enrich with data
• Risk Data
• Property attributes
• Points of interest
• Demographic data
• Building footprints
• Boundaries
ID lookup
Business impact
• Network performance
• Sales and Marketing
Optimization
• 5G buildout
• Subscriber analytics
24. Leveraging location enrichment and ML to pre-qualify
every location in a market
24
Geocoding, Enrichment, and Scoring
Elastic Autoscaling
Microservices
Store the PreciselyID
+ Network ID
+ Signal Quality Score
+ 911 data
+ Lat/Long
+ Tax Info
Amazon
Cloud
Toolset
Mobile Data
Hotspot Data
Public Data
Property Data
Model best signal by property
type and location
Geo-enrich with Property attributes
Geocode and generate a PreciselyID
Score the
entire market
Geocode
Geocode
candidate address
and generate a
PreciselyID
DBLookup
Gather candidate
details by PreciselyID
ApplicationLayer
There are two reasons for this. The first is technical. The wavelengths that operate as part of the 5G specification, especially in the ultra-dense 30-100GB range, are more attenuated by physical obstructions. As a result. Microcell placement needs closer proximity between devices in order for 5G to deliver on its expected network speeds.
For the millimeter waves in the ultra-dense spectrum, this requires attention to physical realities – building material, distance from cell, knowledge of what people will do with the improved bandwidth, even the weather, as rain can interfere with wave transmission.
You’re going to have more frequent handoffs as a result. And each handoff is a potential lost call or delayed signal.
ABI Research just noted in their Reality Check on the 2019 Mobile World Congress that “everyone” agrees” that antennas are key to successful 5G deployment. In addition, despite some vendor claims that existing base stations can be upgraded to 5G, it’s far more likely still be deployed as an overlay, as existing stations will still be utilized for 4G.
When it comes to getting this all to work, location, knowing where to place a cell and how much bandwidth to supply, will be how Telcos gain a strategic advantage in the new world of 5G.
There’s a shift here between writing in the third person to writing in the second person. Why?
Two reasons: first, it’s useful in challenger to have a tone that isn’t abstracted, and it’s impossible to talk about the state of the industry entirely in the second person. So state of the industry is third person, impact on person being pitched to is second person. Second, it’s a technique called interpellation – it works by describing the subject position that the writer/speaker/salesperson wants the audience to see itself in. The simplest way of doing it is selective use of second person. I don’t recommend it for a blog post, but it works well as a technique for a pitch or keynote (the latter depends somewhat on audience).
We might be tempted to think this is business as usual, just another incremental upgrade in mobile technology. It’s not.
As Yogen Patel, head of Amdocs open network marketing, pointed out at this year’s Mobile World Congress: previous upgrades have largely been about capacity. But with 5G, the real value will be quality of service. The higher speeds and lower latency will combine with new services to allow providers to differentiate based on quality of service in more obvious ways more obvious ways, and in ways that matter to consumers..
This is a big change, and whenever there’s a shift like this in the marketplace, new winners emerge, and old winners fall through the cracks.
This happens with every major technology shift. HD changed who dominated the television market. The shift to mobile changed who made money by supplying information and services
Now a similar change is happening in TELCO. And it’s happening amidst massive capital expenditures for infrastructure like fiber, and the cost of securing access to bandwidth.
As a result, the question of where to place towers and how to structure macro and micro cells, and to assign data, has never mattered more.
The building of infrastructure, like the laying of fiber optic cable, or the placing of 5G cells, will determine signal strength and separate the products and service that matter from those that don’t.
In effect, 5G transforms a tactical issue into a primary strategic concern. If you’re still treating location intelligence as a tactical issue, or as an operational concern, you’re doing 4G thinking in a 5G world.
Every TELCO is going to be making huge investments in infrastructure. You’re going to make those investments. The question is whether you’re going to make the right ones. The answer to that will depend on three things: are you building in the right place, are you marketing to and servicing the right people, and are you properly provisioning the service?
It doesn’t take much to figure out what happens if it goes wrong. You build in the wrong place and a building interferes with your signal, and you lose customers.
You don’t understand your audience and how they use their devices, and as a result you don’t have enough bandwidth in the right spot, or you have too much where you don’t need it. You lose money.
You provision poorly and waste a 200-300 dollars every time you send a crew out to see why a wireless signal isn’t working inside a home as advertised. You lose customers and money—one service call at a time.
Again…the issue with using second person seems accusatory. It’s the wrong tone.
I’m happy to turn this over to the wisdom of the committee. I absolutely understand the concern here. On the other hand, rational drowning is the moment in challenger when it pays to be a bit more confrontational.
When we’re talking billions of dollars of investment, even a small error racks up a pretty disturbing bill. Even if .5% of placement is done poorly – and I think we can agree that’s an artificially low number – it means that a 10B investment results in 50M of waste.
It doesn’t happen all at once. It happens tower by tower, $2-3 million at a time, station by station, cell by cell. If any of these are placed incorrectly, if they can’t provide the right level of service, if they have to be replaced or supplemented because of inefficient coverage, the costs will pile up, and pile high.
It doesn’t end there. Those unforced errors go hand in hand with lost opportunities. An apartment building that gets insufficient coverage, and costs you 50 new subscribers at $600 per year, or an office park with insufficient signal, that costs you 10 small businesses running at $5000 annually.
Now imagine that happening over and over again. Imagine losing 4G LTE customers because their friends aren’t happy with your 5G coverage.
Those aren’t rounding errors. They’re the annual margins. They’re the time to ROI.
The good news is that there’s a better way to do this, a smarter way, a way that takes advantage of the latest technologies and years of experience and knowledge.
You need a comprehensive approach to location intelligence, one that can understand location in three dimensions and then also understand the people who live and work and move through those places, and that can map the activities those people undertake, device by device.
You need to know the shape of building, type of building, topology, the digital traffic passing through that location, the likely demographic and psychographic factors that will influence present and future data consumption in that location. In a 5G world, where signal must be closer to the people who use it, you really need to know the people, where they are, and what they do.
Location intelligence is more than just about looking at a map. Real location intelligence is knowing where to build, who to market to, and how to get it right.
If you can take all that information and integrate it together, then you can build a digital twin before you invest in the build-out; you can create a simulation that can validate or modify strategy and inform execution.
You can learn, and adapt, without wasting funds on bad builds or the wrong audience.
[---- Solution / Marketing ----]
That superior understanding of your customers helps you identify who to market these new services to, who can afford them, who wants them, even who thinks they need them. It can target growth in areas or populations in IoT or new streaming services. It can look at city zoning, economic development data, school locations, and so much more.
Product: Master Location Data
….Computerization leads us onto use-cases for addressing and geocoding
An address is a multi-field labelling system where each field is owned and maintained by different entities. These fields frequently change and the address itself has multiple purposes. Addresses are necessarily highly fault tolerant and therefore utterly unsuited for operationalization.
Now that we have an ID, the real fun can begin. Let’s start to operationalize your addresses.
Touch on all items on the slide. We have additional slides on Connected Data and Address Fabric.
ID Lookup: The ability to lookup an ID (preciselyID) and return the address
Extended Attributes: Easy to provide additional property centric information that informs decision making
Parent/Child: The relationship between primary and secondary addresses
Type-Ahead: Human interactive lookup of addresses
Hard to locate Addresses: Known addresses with poor locations
Connected Data Portfolio: Rich data portfolio connected with a high degree of integrity (see later)
Address Fabric: A raw version of the data inside the geocoder
Coming Soon
History: Currency of an Address and whats happened to it so far
Additional Points: Different types of locations, not just the best we have
World Premium Point of Interest: Included as part of the geocoding process to match business name not just street address
Genealogy: A linkage file that describes the addresses relationship to sub-buildings, rooftops (if different) and parcels)
Use Address Fabric to count buildings within an area of interest.
Other pieces of information contained on address records include
FIPS code and Census information
Parent/child relationships for primary and secondary addresses (units, suites, apartments, etc.)
Property type (Business or Residential)
We have off the shelf data (footprints, POIs, boundaries, etc.) available to quickly enrich your addresses. This is enabled by the pbKey feature within MLD and the fact that we have pbKeys on our enrichment data.
Enriched address provide more context and serve as a better starting point for business operations/challenges/analytics/etc. that appear on the right side of this slide.
We have off the shelf data (footprints, POIs, boundaries, etc.) available to quickly enrich your addresses. This is enabled by the pbKey feature within MLD and the fact that we have pbKeys on our enrichment data.
Enriched address provide more context and serve as a better starting point for business operations/challenges/analytics/etc. that appear on the right side of this slide.
Device IDs
Signal strength
Predicted coverage
Data consumption by device
Average monthly bill
Competitive fiber
Challenge was to classify (Fiber-lit Predict vs. Non) a highly multi-dimensional and imbalanced information of location-enriched 1000-plus-col dataset. So we two approaches to predict next-gen fiber-lit addresses. NeuralNet based classification and SVM-based novelty detection.
NN Classification
Dropped numeric columns with high missing rate and Dimension Reduction (DR)
Performed RFE-based DR again on categorical variavles
Concatenated reduced numeric columns with categorical variables
Split the data 60-20-20 (Training-Validation-Test)
Applied 3 layers of NN (Relu on 1st and 2nd) and Sigmoid on 3rd
Applied Binary Cross-entropy as the Loss function
Batch size of 1K/epoch for 100 epoch
Result as displayed - For negative class we get the f1-score of 0.98 (expected as this is the majority) and positive class a score of 0.92 which is a good result (Google precision, recall and f1-score)
Precision is perfect but recall is a bit lower 0.85 i.e. 85% of the actual ones are captured by this prediction. And 100% of the prediction are the true ones.
SVM
Repeat steps 1-4 above followed by a one-class SVM since this was a PU task by nature since the -ve were unlabeled building rather than true negatives.
Future:For NN try multiple layers with activation function combination
Try ensemble of models