You may not have heard of “The Very Group” but as one of the UK’s biggest retailers with annual revenues of over £2.15 billion, the company has a heritage stretching back a century. It now operates solely online, selling over 500,000 different items to over three million customers.
Chief Data Officer Steven Pimblett is the man in charge of making sure that the company is using its rich datasets (spanning insight on these 500,000 SKUs for 2,000 brands across fashion, home, electrical and more; customer behaviour and augmenting third-party data) to optimise performance for customers and the company’s shareholders; no mean feat.
Brought in three years ago as the Very Group’s first CDO, he has an expansive remit spanning everything from data platforms and infrastructure (“all the technical side, which can sometimes live with the CTO”) through to BI and analytics, with “teams embedded across the organisation”, as well as all things data science, data governance and operations; he also looks after the marketing technology stack, he says.
It’s a broad set of responsibilities, and sitting down to chat with Pimblett, The Stack’s first question is how he manages co-owning the tech stack with the company’s CTO without stepping on each others’ toes.
CTOs "happy to have an executive partner" on data...
He says: “I've been both a CDO and a CIO over my career and what I find is that a lot of CTOs – because the landscape is so complex across the data environment, the analytical environment, the risk environment – are happy to have an executive partner that can take a lot of that heavy lifting.
“But you really have to make sure that you've got a great relationship, because data without tech wouldn’t be a great strategy” Pimblett adds.
When it comes to the data side, The Very Group has a lot: “You've got every digital touch point imaginable, from offline media, to online media, to your own website, app, mail; it’s a big B2C digital estate” he says.
See also: The Big Interview: Goldman Sachs’ Chief Data Officer Neema Raphael
“But what I didn't really know, until I was going through the interview process, was just how large it is, as a financial services organisation.
“Outside of the banks, it's the biggest lender in the UK. So there's a whole financial services dataset – fully regulated, fully FCA compliant, my role’s an SM&CR” he notes – and “I think the combination of all those touch points with that rich data, was really a job that I couldn't turn down; the scale of the data and the opportunity to leverage it is fantastic” he adds.
On joining the company, his first job was to pull together some disparate data specialists, he says. Whilst the company was data-rich, it “definitely didn't have an end-to-end data strategy” he says, so the CDO he “took all the diverse teams from across the organisation and bought them together.
Bringing data teams together
"They were dotted round a bit, like the platform team was with the CTO, the analytics team was embedded in marketing, some of the governance teams were in risk. I pulled them all together and created what I branded a ‘DNA’ team [short, loosely, for Data, Insight, Action], created a hub-and-spoke model” (now numbering some 150 people)” he explains.
The “hubs” are his “heads of…” who manage “chapters” (or families of specialists in an Agile structure) but, as he explains to The Stack, “the rubber hits the road when we partner with the ‘spokes’: business verticals or performance units, as we call them at Very – where we help them leverage data and analytics to drive tangible actions in their area.”
Some examples: Understanding customer acquisition spending in more granular detail on the marketing front (“We've built a single customer view in the last three years that helps us understand both attitudinal and, stats about all our customers for a richer customer centric-marketing strategy”) as well as supply chain and warehousing optimisation.
He says: “We’re building predictive models with seasonal curves on all these different dynamics; multi-category retail is really complex as well: We do fashion, sports, beauty, home electrics, children's wear… so the complexities of each subcategory make the analytics pretty complex.
"We've leveraged SageMaker..."
“We've leveraged AWS SageMaker, built our own statistical and predictive models, and we're doing more and more deep learning in that space.
“But the biggest challenge”, he adds, “is actually explaining it all, to a business that's one hundred years old, with lots of [product] buyers who have been there decades – and we've brought in our brand new AI-led approach to forecasting. How do you explain that?” he reflects wryly.
Pimblett hastens to add that he’s not playing down existing teams. “They are not ‘finger in the air’. They've been doing it for years and years; they have pretty robust, mathematically-driven [approaches]; Excel, etc.
“But we know, we believe, and we are now proving it, that we can be much more accurate – and use advanced analytics to better predict…”
Getting the company fully on board has meant creating unusual cross-functional teams, he says, like a “PhD data science mathematician, and a buyer who's been there for 20 years, and getting them to co-create; that’s been the challenge! So my teams are not sat in a [silo], they are embedded with lines of business, understanding the domain and getting the feedback [like] ‘Oh! I actually didn't know you could help us with seasonality curves and the role of price’, which is extremely dynamic and dependent on the category.”
That buy-in, he emphasises, is absolutely critical a CDO’s success: “Co-creation, [driving] adoption, value-creation… you can have the best model in the world. But if no one's actually using it, because it's a recommendation and ultimately the buyer is going ‘that's brilliant, but I’m going to buy my own stock’, then you won't create value.”
The Very Group: More AWS, more GenAI
In terms of underlying technology platforms, The Very Group had already been on an on-premises to AWS migration before its first CDO joined, he says, which it has now accelerated and its data team has benefited from.
“We’ve most of our biggest datasets into the cloud. Most of that is on AWS. We’re also quite a big Teradata shop so we’ve moved that to their cloud offering, on top of AWS. But the biggest thing I've probably moved the dial on in the last three years is the AI stack. Prior to that, it was very much a bit of SaaS, a bit of R, but nothing at scale. But over the last couple of years in particular we’ve partnered with AWS to leverage their AI and ML stack; I mentioned SageMaker, but we have now also launched an incubator to use generative AI capabilities with Bedrock” he explains.
How does it hope to benefit from generative AI?
“There are three outcomes that we see that we can leverage it [for] as a retailer” says Pimblett. One he picks up on is product descriptions.
As he explains, owing to the group’s thousands and thousands of SKUs “40% of our products haven't really got in-depth product descriptions.
“One of our first use cases going through the incubator is twofold. Really, it's ‘can we generate full product descriptions from a short description?’ An even more interesting one, and really where it's starting to take it to the next level, is in order to get on the site, you have to have an image. So can we generate a full product description from the image?” he says.
“Early indications to both of those are ‘yes’, and ‘yes’.”
“That's stood up: Bedrock running, with AWS and the business and the copyright team trialling it at the moment on dresses, just to prove the tech [that lets us explore] Bedrock; what LLMs we can plug in, what’s the cost, what about the business change; what are the rule the business would have to sign off. But it’s a great example of a use case that has productivity and cost gains and can give a better customer experience.”
Under the hood, there’s a lot of wrenching to be done to make this happen. As he explains, “some of the heavy lifting is having the good foundations of your data. In this example, it's where do we get our product headers from? Where do we get our product images from? Can we give access to the models for that? Whether it’s classic ML or generative AI a big foundational element is your data capability and cataloguing, metadata management and so on. We've been investing pretty heavily in that in the last couple of years” he explains, including through the use of Alation (a “really positive” data cataloguing platform).
See also: The Big Interview: Alation CEO Satyen Sangani on AI, data, disappointments, distributing power
There is, in short, a whole host of innovations happening.
With the company having gone from paper catalogues to “in the last month 90% of our sales via mobile and at least 50% of them have been on a native app” The Very Group’s digital activity is clearly not going to be slowing down. Its CDO meanwhile, has the spotlight on him to demonstrate that all these technologies are able to keep delivering value in what is an exceptionally challenging and complex retail environment.
Pimblett says he’s enjoying the challenge. “Cloud technology, data analytics, digital are a big part of the transformation of the business. They play a big, big role.
"This is all about outcome: If you get it right, then you have the best availability for customers, based upon the products they want to buy, so your conversion goes up, which means normally your sales and margin will be pretty good. Then you can use analytics to distribute it across full price, promotional, and clearance. If you get a good grip on that, happy customers, happy shareholders. If you get it wrong, you are sat on millions of pounds-worth of stock in a warehouse that you have to write off. Customers not happy because you've bought the wrong stock. The shareholders aren't happy because you've eaten into sales and margin!”
No pressure then…
