At Ideal Postcodes, we are always looking ahead and striving to innovate
Since 2018, we’ve combined the cutting edge of machine learning with the problem of retrieving and extracting accurate address information. Ideal Postcodes was the first to combine authoritative datasets like Royal Mail PAF with the power of natural language processing.
By using the extraordinary capabilities of artificial intelligence in our address cleanse service, we were able to cleanse and resolve considerably more opaque and erroneous addresses with high confidence. Using AI analysis we found we could:
Detect premise identifiers like house and unit numbers with more accuracy
Strip out extraneous information like names and directions
Identifying whole attributes of an address like street, city and postal codes
We quickly found that our cleanse solution could match 17% more addresses to an authoritative address by including our NLP AI solution.
Using Artificial Intelligence, Intelligently
Having used artificial intelligence to drive our cleanse product for over 5 years we’ve also learned crucial lessons when it comes to machine learning and handling the results of any AI solution. These are some hard won lessons we’ve learnt to help us build a better AI solution over the years.
AI is only as good as the data you provide it. It’s not sufficient to feed the model a stream of correct addresses. Years of malformed, incorrect which have been labelled is required to arm the model with appropriate knowledge to handle idiosyncratically wrong addresses.
AI can be wrong, and there’s no easy way to check. There’s no easy way to check the output of AI at scale. This means a layer of monitoring, sampling and testing is required to ensure AI is working.
AI can be slow. The computation behind AI can take an order of magnitude of compute or more compared to traditional cleanse methods. This can add latency and reduce throughput.
Address data is a crucial component of customer and business records, and no profile is complete without it. Quality address data helps prevent fraud, delivers better customer experiences, and enables efficient logistic processes. We’ve constantly worked on our AI solution to ensure that the addresses we provide are complete and correctly formatted, using artificial intelligence and data from authoritative agencies.
The Right Approach is Hybrid
We recognised the importance of improving data quality and increasing confidence in address solutions early in 2018. We believe that our experience has shown the correct trade-off and best way to support our address solutions is through a hybrid approach. We’ve combined authoritative data from national databases with AI algorithms only under the right circumstances.
The authoritative data we use includes information from sources such as Royal Mail's Postal Address File (PAF), Ordnance Survey's AddressBase, Office for National Statistics, and Companies House. These databases contain validated and up-to-date details on properties across the UK. By using these trusted agencies, we can ensure that our customers receive accurate address suggestions.
AI automates and improves the process of address cleansing by using algorithms to standardise, verify, and complete addresses. By parsing unstructured text inputs and cross-referencing multiple data sources in real-time, AI can minimise errors. Our AI algorithms have been trained on millions of addresses and can detect patterns in address data faster and more proficiently than a manual process.
Next Steps with Large Language Models
Fast forward to 2023 and AI is an exciting space to work in again. With the advent of Large Language Models like OpenAI’s ChatGPT, new possibilities are opening up in improving our what's possible with address search and validation.
We’ve already begun to incorporate ChatGPT into our documentation to enable developers and website owners to get to an answer faster and solve issues in real time by conversing with our documentation!
We are currently exploring the use of large language models to further classify and understand address validation, which will enable us to enhance our services.
With the advancements in natural language processing, we believe that large language models will allow us to better understand the various ways in which people enter their addresses online. By analysing the text entered by customers, we can determine the likely intended address and provide accurate suggestions to users.
At Ideal Postcodes, we are always looking ahead and striving to innovate. Our use of AI in address cleansing is just one example of this commitment to providing a high-quality service to our customers. As we continue to explore the use of large language models, we are confident that we will maintain our position as a leader in the industry.
Contact us if you have any questions.