Event Location Analytics
Strategy for cost efficiency & improve effectivity of event marketing campaign
Overview
Who is the target user?
One of the automotive company in Indonesia (specifically sales & marketing division) who facilitates hundreds of dealer, spread all over Indonesia.
Background
Dealer often held event in some location, and user usually will subsidize the marketing budget for it. Then, we want to evaluate and analyze the location of event whether : 1. they held event at strategic location or not 2. they already covered all potential location or not 3. they created cannibalism among one another or not, etc.
Expected Benefit
This story map will become a decision support system, and future marketing event could be more planned and become more : - effective = by held event at more strategic location where number of potential buyer is more than any other place and generate higher revenue - efficient = by not spend too much cost at non-strategic location
Network Analysis : Location-Allocation
Which event locations are considered better?
Location-allocation helps find the best locations for event to serve a set of demand locations (in this case, suspect address).
For this analysis, the locations are trying to maximize coverage and minimize facilities. In order to dive deeper, you may swipe right to see which event locations (indicated by green dot) are recommended, and which are not.
Legend of Maps Beside
Please see graph below to see the summary of location allocation analysis result
Only around 34% of event coordinate is good, and the rest might be overlap and cannibalism
Network Analysis : Service Area
Service area are commonly used to visualize and measure accessibility.
In this case, I create 3-colored polygon (green, blue and red) that represent 10, 20 and 50 km drive distance toward a specific event location respectively. In short, it can determine which residents can reach the event within less than 50 km and more likely to see car there.
Output from this analysis will be used as input for the next process, which is clustering. (Note : this 50 km is based on assumption that we are willingly to drive 50 km from Bekasi to BSD to attend GIIAS event)
Unsupervised Machine Learning: Clustering
There are some variable to be analyzed: - service area coverage = how near/accessible the event location is (minimize polarization, nearer better) - Market share = how strong our position in that area (maximize polarization, bigger better) - Total demand = how potential this location is (maximize polarization, higher better) - Total visitor quantity = how is the crowd if we held the event in this area (maximize polarization, higher better) - Cost/SPK = how efficient the event that were held in this area (minimize polarization, smaller better)
Since there are some variables and it might be more complex if we conduct data analysis manually, so we use clustering to get the event location segmentation
Legend of Clustering after interpretation
To know how we understand the clustering result which originally only show cluster ID from 1 to 4, please continue reading the next story.
Power BI for Data Visualization Monitoring
From ArcGIS pro, we could extract the data and create monitoring dashboard if needed.
And with these data visualization, we could interpret the clustering result.
For example: - based on average distance : cluster 1 is the nearest to event location (in average, people in cluster 1 only has to drive 15 km to reach the event). This is much better that cluster 2 which in average, people should drive around 50 km to reach the event. Therefore, we will give score 4 to cluster 1, and score 1 to cluster 2 - based on total demand : cluster 4 show the least demand the , while cluster 2 is the highest according to total demand. Thus, we will give score 4 to cluster 2, and score 1 to cluster 4 - and so on. Please scroll down below to read the summary of clustering interpretation
Explanation of Clustering Result
To make it easier to be interpreted, I visualize the result as heatmap where : 4 (darkest green) is the best score while 1 (white color) is the lowest score
We can understand the characteristic from each cluster, and if we see the map again with clustering result, we could understand which area is included into cluster 1, 2, 3 and 4.
Therefore we can decide the event marketing campaign strategy, such as cost allocation and event location selection.
Conclusion
- By using analysis provided in ArcGIS pro (Location Allocation and Service Area) and machine learning, we could understand the segmentation of event location.
- There are 4 event location segments : - Maintain - Potential - Can't lose them - Need further investigation Each of them has their own characteristic, then we could plan the strategy when we want to held the next event. If not, then it may lead to ineffective and inefficient operations such as allocate huge cost to non-prospect area, not invest in potential area, and other risk.
Future Improvement
- Consider another scoring method since current method might be not fair enough. For example : - highest market share number (such as 75%) will be given score 4 - lowest market share number (such as 15%) will be still given score 1. The gap between 75% and 15% is much wider than gap between 4 and 1.
- Analyze time-series data to get insight when is the best time to held an event. So we could held event at the right place and the right time.
- Analyze deeper with data from Living Atlas, such as Indonesian Purchasing Power per Capita.
- Enrich the analysis by analyze the dealer coverage and other geospatial data.
- Use other geoprocessing tools in ArcGIS, such as Calculate Market Penetration in Business Analyst Tools, etc.
- Create a feature in mobile apps which is used by Salesman, to save the customer location directly in coordinate. So we will get cleaner data, don't need geocoding, and the analysis will be more accurate.
- Create primary key for each province, city, district and sub-district in order to get better result from join (or spatial join), since we currently use city / district name as they key to join.
- Automate the process by utilize Model Builder.
Tool Documentation
This solution uses :
Software :
- ArcGIS Pro Network Analyst extension
- ArcGIS Online as Web GIS Platform
Data :
- Internal data (market share, cost, event visitor, total SPK, etc)
- External data (regional administrative boundaries, Indonesia purchasing power from Living Atlas)
You could see below workflow as the summary
