Work we've done
Better solutions to real-life problems
Our view is that ‘AI is dead – long live AI’.
To us, AI means using the best of Artificial Intelligence, Augmented Intelligence, Automated Intelligence and Accessible Intelligence. We build bespoke products that combine data science, engineering and product design to get the best of human and machine intelligence. Here are a few things we’ve done – but the options are infinite. What’s key is to find where AI is the right solution for your business’s specific challenges and ambition.
Contact usInteract with machine learning (PlayML)Challenge
A major UK airport had a manual method of updating car park prices. This meant prices for long-stay, short-stay and valet were not optimal and could fluctuate for customers.
Solution
Alongside engineering a new cloud data platform, we worked with the commercial and digital teams to establish a new price testing capability, including new processes and tools to enable multiple pricing strategies to be deployed and compared. Our data scientists designed pricing strategies based on booking curve and product mix analysis which increased booked revenue.
Result
- 7%+ revenue increase based on new pricing strategy identified and tested
Challenge
The content brand of a global industrial product and electrical component supplier have a diverse customer base which they are trying to engage online through targeted technical articles and relevant content. In order to maximise advertising potential on their website and drive usage across group retail brands, it was necessary to enhance the User Experience (UX) through relevant content and advanced content discovery to increase engagement.
Solution
Datasparq designed, built and now support an end to end cloud data service which incorporated a reinforcement learning framework to provide personalised recommendations based on each individual user’s taste profile and reading history. The service is currently live and incorporated through relevant APIs in to the website.
Result
- Increased customer engagement
Challenge
A leading national retailer wanted to establish a new service offering within existing stores. But which locations would generate the highest revenues for this new service, and which should they open first?
Solution
Within just 3 months, we used anonymised customer data to build an accurate predictive model of how customers would spend. The output was a list of priority locations ranked by predicted year 1 revenues with clear explanations to augment the final real-estate investment decision. This was built to be flexible and deployed across different cloud provider platforms.
Result
- Optimised retail footprint
- $12m+ annual revenue increase
Challenge
A leading Car Park Management company using Automatic Number Plate Recognition (ANPR) was issuing many Parking Charge Notices a year but fewer than half were being paid. How could they predict payers and non-payers?
Solution
We built machine learning models that combined factors like weather, vehicle data and local services to confidently identify which notices should be sent and which shouldn’t. Integrating this prediction into the issuing system significantly reduced costs without affecting revenues, and was integrated into the daily issuing process, processing over 120tb of data every day.
Result
- Improved customer payment collection rates
- £1m+ reduced costs annually
Challenge
Our client is a high turnover/low margin food supply business.How could they improve their demand forecasting to increase profit through reduced waste and optimised warehouse and haulage use?
Solution
Over just three weeks, we worked closely with the client to understand how factors like holidays, events, weather, trends and news affected demand. We then used this data to train a machine-learning model for each product sold. Performance was measured against metrics allowing direct comparison with their current forecasting.
Result
- More accurate food demand forecasting
- 25%+ increase in number of days 95% accuracy achieved
Challenge
A major government department was managing delivery of large capital investment projects by complex stakeholders where risks were either invisible or not raised in time. How could they reduce overspend and delays?
Solution
In under 14 weeks, we created a multi-cloud architecture using both Google Cloud platform and Microsoft Azure while a multi-disciplinary product team built a data platform and an application. 250 users were able to use the product to better mitigate risks.
Result
- Mitigated investment risk
- 100+ person-days saved a month through automation
- 1 day - high risks now flagged in 24 hours (previously 4 weeks)
Challenge
Reed submitted a challenge to the Emergent Alliance, asking how skills data might be used to help people get back into work after Covid, especially those whose sector or industry has been significantly disrupted. We contributed engineering, data science and product management effort towards the challenge.
Solution
Our solution used thousands of job descriptions and skills to create a graph of canonical job roles, connected to other roles through a skills similarity measure. This provided the basis for a search engine which accepts skills as an input from which we’re able to produce a probabilistic rank of job clusters and a ranked results set based on continuously updated historical data.
Result
The Job Finder Machine is live. This proof of concept demonstrates a powerful, intuitive and different approach to finding jobs. If you’re interested in discussing how the work done here might support challenges you’re facing with recruitment, then please get in touch.
Doing things better
Our innovative team are always exploring new ways for AI and machine learning to work better for businesses
