.svg)
Software Engineering
We can turn innovative ideas into world-class products and services. Working iteratively our specialist software engineers build robust systems to bring your vision to life, no matter the medium.
Case studies
Software Engineering
Timeform (Flutter)
Making expert racing insights accessible with AI
.png)
Background
Following our successful AI partnership, Timeform partnered again with Parallax to create QuickCard
→ Results
No items found.
The Brief
Timeform, part of the Flutter group, provides predictive insights, ratings and analysis for horse racing, analysing around 11,000 races a year in the UK and Ireland and over 130,000 individual horse performances.
They wanted to create a new, simplified race card (known as QuickCard) for casual fans – one that is powered by their data and strips out jargon to make it easier for those less familiar with the nuances of horse racing to engage with the sport.
Having already partnered with Timeform to build a bespoke generative AI platform, this was a natural next step for us. We’d built the functionality to predict positioning and provide commentary; now we needed to translate this into meaningful and, crucially, accurate summaries for novice audiences.
No items found.
→ From the client
The Challenge
The goal was for QuickCard to be entirely automated, to be available for all UK and Irish racing with no human input. The data it is based on (the positioning and comments from our previous phase of work) is checked by a human analyst before it’s published. But the output for QuickCard needed to be automatic.
We could not, as we originally tested, set the AI model free to generate phrases for QuickCard based on what it identified as positive and negative traits from the existing commentary because there is so much nuance in this sport. A horse’s recent surgical procedure for example, could be interpreted as either a good or a bad thing depending on wider context and could wildly skew the reliability of our summary cards.
To ensure accuracy, we had to come up with a way of restricting the model to a more curated set of sentiments and phrases, while still allowing it to make selections at scale.
The Solution
To address this, we worked closely with Timeform’s analysts to define the kinds of insights QuickCard can present and when they are appropriate. The system assesses each horse using the available form data and can leave an insight blank when the evidence is not strong enough, keeping the output concise and grounded in Timeform’s expertise.
We built the solution using Claude Sonnet 4, running through AWS Bedrock. We initially benchmarked the approach against Claude Opus, but at the volume required, Sonnet 4 provided a more cost-efficient option on a per-race basis.
Timeform’s analysts prepared example QuickCards to represent appropriate outputs. These examples support ongoing evaluation of output quality; against the benchmark set, our system achieved an F1 score of more than 95%.
No items found.
The Impact
The result is a fully automated pipeline capable of generating QuickCards for every eligible race, with the underlying racing analysis still grounded in Timeform’s expert judgement. Analyst guidance and ongoing evaluation help keep the automation reliable.
Since launching in summer 2025, QuickCard has been used at several large race events. Timeform is now exploring a digital format to make the QuickCard more accessible to the general market.
→ From the client
Working with Parallax has helped us better understand the art of the possible in unlocking the full value of our data, analytics and how we scale our AI capabilities to engage, inform and be trusted by customers.
Michael Williamson
Head of Racing Analytics and Product at Timeform
