Today’s startup thinks businesses should have tidier data. It’s using a mix of proprietary machine learning and data science, alongside a sprinkling of generative AI, to serve its target market.

Fireflai’s story also builds nicely on the successful exit its co-founder achieved with his previous business. Scroll down to read more.

But first:

• Atomico–which yesterday announced it has closed two new funds totalling $1.24 billion to back early- and growth-stage startups across Europe–wants YOUR view on the state of European tech

• Fill in Atomico’s survey to take part in this annual staple of the VC world

– Martin

Fireflai wants to light the way to chaos-free business data

It’s amazing how chaotic many businesses are if you look beneath the surface. The appearance of a flawless operation can often mask manual processes, inefficient software, and reliance on ‘that thing Joan knows how to do but no-one else has ever figured out’.

And that chaos can often be present in a company’s data, too. Take an ERP system, for example which many businesses use to organise records of what could be thousands of products.

“A lot of companies struggle with what they call their ‘master data’,” explains Thomas Gardner, CEO and co-founder of Manchester-based Fireflai.

“There could be multiple ERPs that don't talk to each other. Over time, you can end up with millions of records that have got duplications, that have bad data about products that they either don't serve or don't sell anymore.”

And even when the records themselves are accurate, there can be missing data, such as dimensions or weights.

“That has a huge knock-on effect downstream. Because, for example, if they've got a product that isn't populated with the weight, and that product gets to the docks for shipping in China and comes over, they get hit by massive shipping costs,” says Gardner.

“And it also affects their ability to rationalise pricing and understand where they get the best discounts from, because they're not able to tell which suppliers are giving them the best discounts.”

Just as the saying goes that ‘a tidy home equals a tidy mind’, tidy data can very much equal a tidy business.

It won’t surprise you to read that this is the problem Fireflai has set out to fix. It uses a combination of machine learning and data science, combined with a sprinkling of generative AI over the top, to clean up millions of lines of data for manufacturing, distribution and retail businesses.

How it works

Gardner shares an example of a business working with Fireflai that operates in the construction supplies sector.

“80% of their stock is not online. And it's not online because they don't have the skills, the time, or the capacity–because this data is constantly coming in–to get those products into a position where they can be put on the website.”

And so Gardner says Fireflai has been put to use in this case to perform product enrichment. The software scans through details of 200,000 boilers and radiators, identifying where data like dimensions, weight, and a product description is missing. It then automatically adds these details.

“We're cleansing that data and getting it structured, and then we're passing it through generative AI, which is going to the internet. It's looking at supplier websites, it's looking at customer-trusted sources like competitors, it's taking that data, and it's populating their system with the data so it can then be published onto the website. 

“In effect, we can reduce the 80% [of products not listed online] down to 20%. And all of this is really quick. So unlike traditional consulting engagements, which can be six to 12 months long, we're doing this within weeks.”

But surely, generative AI is a risky tool here? I’m reminded of a post I saw on Threads just yesterday, where an image generator had re-invented the disc brake system completely inaccurately. Surely there’s a risk generative AI could hallucinate product descriptions or basic product metrics?

Gardner says the third-party models they use are more accurate than they expected, out of the box. What’s more, he says that prior to AI getting involved, the startup’s proprietary machine learning and data science model will have done a lot of the work. 

“What we're then asking AI to do is look at the output of machine learning and say ‘do you agree that that is the product we think it is, and is that description correct?’ We're using AI as the validation lever.”

The story so far

Gardner previously founded a business automation consultancy called Robiquity, in 2016. He exited the company in 2022, after private equity firm Growth Capital Partners took a stake in it.

He has co-founded Fireflai with Craig Sumner, who is the startup’s CTO. Sumner’s background includes a stint as Head of Intelligent Automation for CPG at Deloitte, among other digital process transformation and technology consulting roles.

Gardner says the pair began work on Fireflai in October last year. They have since completed a successful pilot programme with a business in their target market, and are currently running five more.

The startup has recently been part of Praetura Ventures’ Praeseed programme, as well as taking part in the Microsoft for Startups programme.

Next up, they plan to boost their resources so they can focus on customer acquisition and product development simultaneously.

“The whole focus of the next year is to take these pilots across multiple industries into production, monitor their behaviours on the platform, bring new customers into the fold and really nail down product-market fit,” says Gardner.

“The go-to-market attraction for using a SaaS-based solution like ours is that we've done all the hard work; all the integration, all the infrastructure, all of the risk sits on our platform. All of the use cases are pre-built…

“We allow customers to access that technology without having to bring in an army of data- or machine learning engineers. They don't have the bad infrastructure, they don’t have to set up the software, they don't have to take on the risk of doing AI and getting it wrong. 

“There's a whole segment of the market that sits in that just under enterprise, but just above SME, call it mid-markets. And we think those are the sweet spot for where we'll position our product long term.”

Subscribe to Premium to read the rest.