Build the Identity Foundation That Pays Off

JACK FREEMAN:

Thank you all for coming to join me today. I'm Jack Freeman. I'm in charge of customer analytics at New Look. We are a UK fashion brand. We were founded in 1969 and have about 300 stores in the UK with a big online presence — so a big omnichannel presence in the UK. We're the third largest womenswear retailer, focused on affordable fashion for women aged 18 to 45, with a smaller men's category and also 9-1-5, which is our nine to fifteen year olds. We try to get the full lifecycle of our customers. We also extend our reach through other digital providers such as ASOS and Zalando, which is really big in Germany. So we have extended reach to other markets, and we've also recently launched the New Look Club.

Thanks to things like Amperity, where we can better store our customer data — we launched properly in November and we've already got a million customers signed up, which is fantastic. We've got roughly 8 million active customers, with about 7 million opted into marketing, and about 85% of our gross demand is linked to PII. Basically, everyone that goes into a store gives us their email for some reason, and it's fantastic. From that we run great marketing for them and really bring them into the brand.

So I want to talk you through the vision of New Look and how we've got to where we are, and why customers are so important to us. Our vision is to blend the art of fashion retail with the science of data. This is really important to us because we have colleagues that have spent 20 or 30 years of their career searching for garments and finding the perfect dress for the perfect person at the right moment in their lives. If we come through as a bulldozer saying data is the cure, that's not going to work and that's not our business. We want to blend it together and we want to put the customer at the center of everything we do — and everything means everything. We're not just using customer data for our marketing activities; everything we do is through our customers. And I should really state that we can get here thanks to our partnerships with Databricks and Amperity.

It took us about eight months to get fully live with Databricks. We've migrated about 90% of our data into Databricks, which is fantastic — so we can activate across those channels. With Amperity, it was a matter of weeks to get live. We went really fast, and because of that we can make a difference at pace. This has all happened in the last three years, with Amperity in the last year. Pace is key. And the size of our teams — analytics is about 12 people, data science is a couple of people, engineering is about 10 people. Roughly 1.6 billion dollars in revenue, big scale, small teams, fast pace. Hopefully that resonates with many of you here.

Our lifecycle — we need to put the customer in the middle of everything. How we plan, what strategy we're doing — we need to ensure that any business strategies we have running are informed by an understanding of what our customers want, when they want it, and how they want to shop. As little as three years ago, New Look was like a market trader: oh, that's pretty, let's buy 10,000 units and hope it sells. Oh, it's not selling, let's discount it. How can we get the customer at the center of that and understand their promotion sensitivity, their churn predictions, and what trends we can see through our data? One example: we've noticed that if students buy a dress early, it's going to be a hit, so you've got to restock it as quickly as possible. How we distribute is really important, as you can imagine, because there are different trends across different parts of the UK.

And equally it's about the supply chain — I'm going to go into a bit more detail about how we've used customer data to change our supply chain. Again, this is all at pace, all in the last year or two. Most of the people here have got marketing backgrounds, so I won't go into too much depth — but we send five emails a week to our customers, which is quite a lot. In fashion, though, people want a lot of emails. That's about 1.5 billion emails a year we're sending, so we've got to really understand what our customers want and how to personalize those emails. And then what's really key is closing that loop — looking at your aftercare, your returns rates, understanding who, why, and when they're contacting customer care. Can we get ahead of problems? Can we pull those signals into the individual platform, understand our customers better, and change how we're developing as a business?

That's quite a wordy slide, but there's a lot going on. I think it's really important to take you through the transformation in a slightly more technical way as well. Three years ago we started with the modernization of our data platform with Databricks and leveraging Azure. We migrated from an IBM DB2 solution — the Databricks team was fantastic in helping us move everything across — and now we have a scalable platform that enables us to leverage future technologies within AI, as well as the speed of what we can do. I'll come back later to some of the projects we're doing and how we can test and learn at pace and deploy to our colleagues in record time.

It wasn't cheap. We've all talked about ROI — if we're going to do a million-pound migration, you've got to show value quickly. At pace, we did a two-week POC to start with. We looked at supply chain — 300 stores, that's a lot of supply chain work. That whole supply chain, as little as two years ago, ran off one spreadsheet. One giant spreadsheet, and the columns were just scary. We just took it into Databricks, organized it, cleaned it, changed the pipelines. The team were getting data once a week to run the entire supply chain across the entire globe. Now it runs every 20 minutes and they get much better quality data, pace, and speed. And again, we can do that because we've got all our data in one place, and it all deploys now via Power BI.

So anyone can access the data. As we gain speed and get more demand for data — because we're starting to get these wins and we know we want to put the customer at the start of the business — we also recognize we need to bring our colleagues on that journey. Several people have already talked today about how change management of colleagues is really important. So we launched DAVE — Data Advocate, Visualization Enthusiasts — our colleagues across the business that we want to bring along. The key thing is that these people are smart and they know our products better than anyone else, but if we don't bring them up to speed on data, analytics, and visualization, two things can happen: one, they get left behind as the business changes. And I also think there's a bit of compassion here — the shape and structure of our business is going to change in the future.

I want these people to be ready for their future, whatever career that may be as we change and adjust our business. Bringing it back to partnership as well — next week we're going down to Weymouth, the seaside town where the business started, where the legacy parts of our business are, and Amperity are joining us, Databricks are joining us, and some of our other marketing partners such as our digital media agency and Kantar, who we do a lot of work with for our market data. Under that umbrella we can bring everyone along to upskill through data literacy training, Power BI training, and so on.

So the buy-in is there, people understand data is important, and they can get the training to activate it. We have all the data in one place and we've had some quick wins and we're driving efficiency — but it's not good enough. We need our customer data, we need to understand our customers better. I was looking at our customer identifier, which was done via some very old fuzzy logic engine, and it wasn't good enough. I had something like a thousand customers spending 50,000 pounds a year on clothing, and I was like, that is not even slightly possible — that's a lot of dresses and t-shirts. We looked into it and it just didn't make sense. That's when we got talking to Amperity. With Amperity, our PII is still sitting within our Databricks instance, within our sphere of influence. I'm not sending data to a satellite and having lag issues, deployment issues, and worrying about what happens if we change our provider in the future. None of that. It all sits within the environment, which gives us better governance, security, and understanding of our data. That's just the starting point — then there's the quality of the stitch, and I'll go into the architecture and how we use this data in a moment.

So we've got our customer data into Amperity, stitched it better, and then we face into the future around AI use cases. I just want to touch on something we're doing with Databricks right now using Agent Bricks and Databricks Apps. Some of these are buzzwords, but I won't go into too much detail — I'm loving it. It's incredibly interesting and it's changing our colleagues' roles in the business. Imagine that every piece of clothing you're wearing, someone somewhere has put that garment on a model and measured every single stitch, seam, fit, and curve. Is the bust correct? Is it in the right place? That process at New Look would take two weeks, because you get notes, you send to a supplier who's in India, you wait for the time difference, you send it back — you get the idea.

We've now got an app, a standardized form where our colleagues put a headset on and just talk through the fit room process. They talk to the machine and it collects all that raw, unstructured data and standardizes it. And it does one better — if you're in the dress department, you know about your supplier, but you might not know that supplier also does shoes, jeans, and something else, and we might have problems in other areas that you're not checking. So we can use Databricks to append more data. And what we're going to do in the future is use our Amperity ID to add even more data, such as returns rates and what customers are saying through the customer support center, enabling that merchandiser to change the product before it's even been mass produced. That whole two-week process is now a day. Our speed to market has gone through the roof, the quality of our products has gone up, and we're expecting our returns rates to decline, which is really exciting.

Okay, technical diagrams. I know we're all here for technical diagrams. So — we've all got loads of data: data, data, data, data. Marsys is our main email deployment platform. We've got SAP E-commerce Cloud for the website, Adobe collecting clickstream data, ASOS and Oracle doing our retail and stock data. All of that, when I first joined, was just scary. What we've done is sucked all that data into Databricks into our lakehouse — it's all there, available, ready for us to interrogate, all governed with Unity Catalog. Then we take all that PII data — emails, phone numbers, everything we've ordered, everything in-store — and put it into Amperity to stitch it all.

I think we had something like a hundred million profiles originally. Working in partnership with Amperity, we then identified quirks in our data we didn't even know existed. For example, if you're a store manager, you might get a bonus for collecting lots of PII. So we can now see who's been entering their own email repeatedly, and exclude them — because that's impacting our promotional activity and where we're going with certain high-value segments. But what we also do is leverage Amperity's out-of-the-box solutions such as churn predictions and product affinity scoring. This is new for us. From a small data analytics team perspective, I would love to build this stuff — I have so much fun doing it — but I don't have time and my marketing team are screaming at me.

They want these models, so we can provide them. The marketing team can have these models and start running campaign tests, then come back to us as a center of excellence to check: did this work, didn't it work? We can go to market at pace with out-of-the-box solutions. But where I wanted to go further is to take the Amperity ID, pull it back into Databricks, and enrich it further. So we take the extended attributes from Amperity, we take all that transactional data, behavioral data, all those signals, and we start building out even bigger models. My data science team are focused on recommendation models and substitution models at the moment. Those are all powered by the Amperity ID. We've also got a lot of returns forecasting going on.

We asked the business, what's one of your biggest pain points? The distribution center said they can't plan their workforce correctly because they don't know when returns are going to hit. So we built a returns prediction on the customer, a returns prediction on the product, a returns prediction on the season. All those data science models built off the Amperity ID, all within the same environment, all then pumped back into Amperity as well. We're using returns predictions to build a returns forecast, and it's going to save 2 million pounds in workforce costs in the next year. All of that was made possible because we have a single unique identifier with the extended attributes already in place, so our data scientists could just pick it up and go.

We've got all those additional models predicting CLV, behavioral segmentations, mission-based segmentations. We can build out basket missions, customer missions, and deploy them to our different outputs. Because Amperity has native links into paid media providers, we can remove that friction for our other colleagues. Our digital media agency was taking data, putting it somewhere, moving it somewhere else — PII data makes me very nervous, I do not want to be doing that — but through Amperity we can just plug it straight through. And where we're getting to is talking with our digital media team and saying, okay, we've had a certain strategy because we were limited by resource. Now we can be smarter — we can run A/B testing, control groups, fallow groups, and holdout groups across these channels. Future us can run better MMA modeling and marketing incrementality. It's about setting us up for the future.

The big win is in the extended attributes in Amperity — treating everyone as unique customers every time. What we found from the data is that a significant portion of our high-value, high-frequency customers have two emails with us: an email they use to buy products, and an email where they receive their marketing. When we surveyed them, it turns out they like to have one email for all their marketing and offers, and then their normal email for, well, proper stuff. Using Amperity we can link all those transactions with their opt-in and then extend it with more attributes, so we can be there at the right time, in the right place, with the right offer for our customers. And it's early days — because this is a lot of change. That's one bit of feedback I give to everyone: change management. Bringing your colleagues on the journey is so important. Yes, we're having wins, yes, we're moving things forward, yes, they're running more tests — but there's a lot more we can do, and there's a long way for us to go.

So we've got this ecosystem, we're building out attributes, and it all sounds great. But I really like this next bit and I think you will too — it's about what this actually means to us, what we've collected, and what you could be collecting and doing with your customer data.

We've pulled in browsing data — how customers have traveled through the website. We've also pulled in transactional data: billions of rows, every single SKU. We've got about 11,000 SKUs live on the website at any one time, in a very competitive marketplace with lots of new product development all the time. We've got transactional data from stores, online, click and collect, order in store — it's complex — and all the card tokens attached to that. We've got returns data — not just the fact that you've returned, but the reason you've returned. We partner with other partners who send even more data into Databricks and Amperity on when, why, how, quality, and so on, and we can follow up and understand what's going on.

Opt-in data is obviously important, and we've also been able to link the Amperity ID and pull in all the PII from all our CSAT information. All our surveys — on the website, in store, online, at marketing events — all stitched into one profile, so you can know this person is happy or unhappy. In the future I'd love to present a story using the Amperity ID in this data to show a customer's journey as they become more and more unhappy, and how we could have done something about it much sooner.

1.5 billion emails a year, all that data stitching — our Club proposition is brand new, so you can see some of the data is still coming in, but that's bringing us a lot of fresh new data. Customer care — every communication, every time a customer has contacted us — pulling that in alongside social behaviors: have they come from Instagram, do they prefer Instagram, Facebook, TikTok? TikTok is huge for Gen Z and in fashion it's absolutely massive. Showing how our customer base is changing is really important. All this comes together.

And that's how we build Phoebe Williams — she's a fictitious customer of ours. We want to show how much information you can pull together simply by having a solution that stitches everything accurately and in one place. The classic PII — and for me a big win was card tokens. I don't know if any of you have tried to build your own card token logic to match customers together, but if you haven't, don't. It is quite painful, and Amperity does a much, much better job. Demographic information is also quite interesting. I worked at New Look for three years and was told we had no date of birth, so I couldn't do any age profiling at all. Once the Amperity stitch was finished, I was like — there's date of birth! Where's that been? We had data we didn't know about and couldn't find, and through this process we found it. So now we can do age group targeting.

Profitability around CLV and AOV — yes, we can pull that out of the platform, great. We can better understand our customers. We can also leverage predictive CLV models within Amperity, and our digital performance marketing partner is building a predictive CLV model that takes this data and the digital signals they receive to build out a marketing strategy around how much CPA they should spend. And all of this sounds complex, but we've been building all of this in the last six months. You can do this stuff at pace with Databricks and Amperity.

Particularly the returns propensity model took us six weeks and is instantly making a difference to the business. Engagement scoring — I think it's really interesting. We have all these signals, but how do we translate them to our colleague base? How can we score our customers and say, okay, we are really pushing this customer segment but they aren't buying more — why? If we look at an engagement score, it turns out they're all using the app, they're all using the website, they're coming to the store all the time. Maybe they just don't have the share of wallet. Times are tough, and particularly with the cost of living crisis in the UK and everything that's going on geopolitically — when we survey our customers, we see that about 30% of our base have about 30 pounds of disposable income for the month.

So if they buy that dress from us, that's their one luxury for the whole month. They can't go out in that dress; they have to buy that dress and then the next month they can go out in that dress. It's important that we show up for them in the right way, with the right product, at the right time — because it's really important to them, and it's important to us. Because of that behavioral engagement data, even more information — all of this is just changing the perceptions of our colleagues. The marketing team suddenly said: if you can tell us a churn prediction and then look at the frequency of emails they receive, can you say we're maybe sending too many emails, which is making them opt out, which is making them churn? And I'm like, yes, absolutely. There's a limit of five emails a week and this customer cohort is getting seven a week — something's not right, let's change what we're doing. And equally, for the first time, we can coalesce CSAT and NPS information with these customer segments.

This is all good — but you're telling me you know who your customer is, great. What are you doing with it? How are you going to make money? How is your CFO going to turn around and say yes, rubber-stamp, you can hire another analyst? So these next elements are where we're starting. And then the next couple of slides I'm going to take you through how we're fundamentally changing the business even further, which I think is quite exciting.

Marketing strategy, multi-touch attribution models — we're starting that now. We need to understand how much we're spending on our marketing and how much money we're making from it, because right now it's a bit blind.

Paid media suppression — Amperity has been really helpful here. We have a target to save two to five million pounds in paid media this year, and it's got to come out of budget. How can we be more efficient with our marketing? Simple things: if someone has just purchased, do we send them ten adverts on Meta, or do we wait a week? That kind of thing sounds simple, but that's where we are in our journey.

Marketing engagement can then be further enhanced with next best action models, cross-selling models — all the fun stuff. We can run it through these platforms and personalize our experiences better with better recommendations. One of my data scientists ran a two-week POC — in part thanks to Claude Code, which is fantastic if any of you haven't tried it yet — around product substitutions, and has proved that his model in two weeks is better than our current website model. We're going to look to deploy it as soon as possible. You can go to market with this stuff so quickly.

But we're not stopping there. A lot of this has been marketing-focused because that's an easier route to market where we can prove ROI. But the bigger wins are around product development and sourcing. If we can get customer data into the hands of our buying and merchandising colleagues, the difference we're going to make is incredible. Right now they're buying products based on aggregated spreadsheets. You'll see them with ten spreadsheets, looking at them all day for the products they're going to buy. These buyers should be looking at fabrics and the quality of the products. How do we get them away from Excel? How do we get them onto the DAVE programs that automate the Excel processes, and how can they start using customer data to look at trends, who's buying, why they're buying, and how they're buying the products? This leads into demand forecasting — that's the big one for the business. Every team spends one to two days a week forecasting. We can get rid of all of that and center it not on an aggregated spreadsheet running a polynomial regression line from Excel, but on data science models that work for everyone, grounded in customer data.

So — bear with me for these next few slides. This is the first time I'm presenting this because it's hot off the press from the last six weeks. We were challenged by our investors and the banks. They said: okay great, we've drunk the Kool-Aid — you have customer data and you're telling us it can essentially run the business — but your plan for the next five years is based on aggregated finance spreadsheets. So they challenged us to build a customer-centric model that plans the business for the next five years. Taking customer data and truly putting it at the heart. And it was done with a team of five people.

Slim and lean — we did about six to eight weeks of work on it. And now we have a model that is going to the banks this week, with KPMG supporting from an auditing perspective, to say: this is what New Look will look like in five years. We take 8.3 million customers, break them into nine segments based on whether they're retail only, online only, omnichannel, are they opted in, are they part of the Club. We look at that behavior, then break it down further into recency, frequency, monetary, and trading periods — are they new, are they retained, what's the probability they'll change? We add in an additional 117 variables per cohort on transaction frequency, UPT, AOV, retention rates. All of that pulled together, attributed to 300-plus stores, and we let the model run. It comes out as a ledger-based model that iteratively says: this many people will lapse, this many will buy this, this many will do this.

What's important is that all these teams historically will build out their own strategies, their own Excel spreadsheets, attach them to a business line, and go to the bank asking for investment. CapEx will now be decided by this model. It's not perfect, but I'm not sure it was perfect before. I'll call out a few — the omnichannel activation side of things: digital CapEx, the store portfolio, the category headroom models that individual teams are running — they conflict, and they all want CapEx. This model arbitrates that, because we know certain customers can't increase their headroom or conversion above a certain limit. It's all done in the data, all centered on the customer instead of aggregated spreadsheets.

The store portfolio element is particularly interesting. We looked at our portfolio, at stores we were going to close, at the customers there — but also at the customers who live near a store but aren't shopping it. We ran analysis to see: when a store closes, what happens to them? Where do they go? Do they go online, to another store, or do they lapse? We can now calculate the omnichannel impact of a store closure. Through this work, the number of stores we were going to close is down by about 20 to 30% because of this analysis, which then supports the omnichannel workstream, which then supports Club. You get the idea.

What this looks like is a model that, within five years, sees our retained base grow by 16%, as we focus on our core high-value customers. Club becomes a key driver of this, and there's significant growth in revenue and omnichannel cross-shopping behavior. It's been a rollercoaster. And I should say — it really put pressure on our colleagues to have better strategies focused on customers. Could they talk about their strategy around a customer? Could they say: we should open a store here because we don't satisfy the customer segments in this area of the UK, and it will drive this much money? It really pushed them to ask: is this the correct strategy? Does it align with my colleagues' strategies? And does it make the difference we need to make at New Look?

Which leads us to the future. We're working at pace, and right now this model is going to the banks this week to get ratified and then have CapEx distributed so we can face into the future of New Look. We're really doubling down in the next two quarters around paid media optimization. We've made a really good start, but I think there's a lot more we can do, and a lot of it is around change management — getting our colleagues to use customer data, not just understand it. We need to work with them, celebrate their successes, and push on this journey. Demand forecasting: can we move off Excel spreadsheets and actually say these customers will buy this much in the future? And then recommendation engines and personalization — we can put the recommendation engine with the Amperity ID in Databricks at the heart of our business.

We don't need what we currently have: a recommendation engine on our website, one over here for paid media. We can have it all centralized in one point, deploying recommendations everywhere — from our buyers all the way through to our marketers. This will be a huge success as Customer 360, and then we can start looking at Product 360. That means: our buyers are thinking about this as a product, but can we think about her? Who's buying our product, is she happy with what she's getting, and are we showing up for her in the right place at the right time?

So I really think this is all about putting the customer at the heart of your business, and, like I said at the start, blending the art of fashion retail with the science of data. I recognize that was quite a whistle-stop tour, but if you have any questions, the microphone can be passed around — and equally I'll be around later. Thank you for your time.