Hi Rob, nice to meet you. Can you start us off by introducing yourself and why you joined BlackCurve?
I’m Rob Horton, I’m Product Director at BlackCurve which is a price optimisation company for online retailers. We basically take a load of data from our retail customers and we use it to calculate a price that we then feedback to the website.
My background is fairly traditional for a Data Scientist, I did an undergraduate in Physics, then did a Ph.D. in the theory of condensed matter, a couple of years in the city and then ended up here.
The underlying theme from the data modelling perspective is equilibrium, I did thermodynamics in my Ph.D., then marketing in the city and then pricing here.
The careers have been different but the actual research piece is fairly constant.
You joined BlackCurve about a year and a half ago, tell us a little bit about your journey so far.
I joined a year and a half ago as a Data Scientist and at that time we had a very powerful tool but we were yet to find a real product-market fit. We had customers who we worked with but it was very much on a consultancy basis.
The last year for me has been working very closely with Emmanuel Aremu, our Marketing Director, Paul Rowley, our Sales Director and Philip Huthwaite, our CEO, to create the story around BlackCurve: why it works and who we should be applying it to. Essentially, turning it from a very clever enterprise-grade solution into a more mass-market SaaS-based solution.
The dream for us is, if you’re an online retailer, you can rock up to Magento or Shopify or whatever your store is, click a button on the app store and you’re away with our pricing software to do all the heavy lifting for you, we set it up and you start seeing immediate value.
We’re seeing 10x returns for our customers, which is really nice. The nice thing about working for this company is that the product actually really works, which shouldn’t be understated from a Data Science perspective.
Yeah, especially in a world where there’s a lot of Data Scientists who can say they provide a real return on investment.
I think I’m slightly abnormal for a Data Scientist as I’m incredibly commercially minded. My default is not for complex solutions, I call it “business first data science”. If it’s the case that a bar chart or simple segmentation solves the problem, my bias would be iterate with that and then work out if you need to do anything more complex.
There are scientific problems and you can make it more complicated if need be. That’s why I think you get a lot of failure in Data Science project and a lot of turnover in Data Scientists because you’ve got to keep business at the core of what you do and it very easy to get lost in the data and solve the fun problem but not the business-critical one.
I do agree with that. There are a lot of projects, especially in bigger companies, where they hire a team of Data Scientist but don’t have a project that a Data Analyst can’t. They try and solve problems with a really complex machine learning solution that’s not that useful.
Exactly, and they might not have specified the correct problem in the first place. Is your problem a machine learning problem? What is the way to solve the problem should be the question you’re asking? Is this the right tool for the job? You shouldn’t be wandering around with the hammer looking for nails.
I’m quite lucky here in that the Co-Founder, Charles, who built the first generation of the software, is the complete other end of the spectrum so we keep each other honest. I would say my bias is too commercial while he can be too theoretical so the pair of us we complement each other well.
In the last year you have done a fair bit with marketing, getting the product-market fit, so what would you say has been your biggest challenge?
Well, we’ve gone through a massive period of growth, we’ve raised 2 seed rounds, £500k and completed it with a £1.5 million recently. We’ve hired a lot more people so we’ve had quite a big cultural change, especially in the ways of working.
The biggest challenge to date has been integration, which is why we’re guns blazing. I don’t think I need to tell too many data scientists about the horrors of cleaning data or the rest of it. The real question is how do we get our developers out of that? How do we automate our process so they can spend as much time writing code, improve the product or optimising things?
For me personally, it been interesting joining a business at the stage as we had a working product but how do you building a unique system that works with data science and research? The technology decisions you make at the start massively impact you at the start.
So if I was a Data Scientist at BlackCurve, what would my day look like?
We like to keep people quite customer-facing, because of the size of the business, there’s 12 of us so we have to be. However, we like to encourage our data scientists to do 3 days a week of research.
That’s quite a lot.
And then aside from that, we have customers reporting and the rest of it. The ultimate goal for us is to automate and optimise everything, to take the human out, so the question for us at the moment is how much is blue skies vs real business impact.
We are data and analytics company so you need data scientists from our background, i.e. research, to get into that business wheel to work with the sales team, to build rapport and to be customer-facing.
We try and keep a balance between adding actual business value and research next generation stuff so we don’t get left behind. I think it important to ensure that people are continually doing research and blue-sky stuff because you need new ideas to avoid problems in the future.
It’s also self-learning and what they can pick up. We interviewed Miguel Martinez from Signal Media who’s recently written a few blog posts about his experience of hiring and losing data scientists. His biggest thing is learning, development and research which he thinks are super important.
I agree, I do think because I’m commercially minded, if someone is paying you, for example, a graduate Data Scientist, which is £50 – £60k plus you’re going to be expected to deliver commercially. So there’s keeping people engaged through R&D and that’s great but it has to provide value to the business. For a company of 12 people, you’re a risky asset. I could hire 3 BDR’s or 2 developers for the same money.
I think ROI on a Data Scientist is well and truly there but you can’t be precious about it.
Do you think that why there’s such a high turnover with Data Scientists? Being paid a lot of money without seeing a proper return on investment?
Yeah, completely. I think that it’s that and the fact that data scientists come from an academic background and want to do research. I don’t want them to lose that and that’s why I try and do two days a week, otherwise, you lose your edge. You have to have one eye on the present and one eye on the future.
Back to BlackCurve, what would you say in the last year and a half your biggest success?
Definitely, product-market fit, getting to the point where we have customers who would happily champion our product.
We have Seamus Óg McGilligan, who was one of our customers from Donaghy Brothers, come over recently and we did a pricing breakfast and he spoke to a number of prospective customers about the platform.
Startup life is hard and it’s really up and down and being up in the trenches and coming out the other side with someone happily fly over, sing your praises to a room of strangers is rewarding.
Make’s all of the hard work worth. We’re in start-up mode at the moment and we work super long hours but when a company says “you’ve done a great job”, it makes it all worth it.
Yeah, I think that for me has been the best thing. The process we went to get there is a culmination of a lot of work across the whole company to figure out what’s the product, who we should be selling to, where does it work best, who understands it and all of that profiling that was a really interesting process to go through.
I learnt more through that process than anywhere else. As a Data Scientist, you can rarely work cross-functionally. In academia, you sit in your little box doing your research and I loved it but it’s very different. In the City, I did portfolio optimisation and built & ran tools but I didn’t interact with sales and marketing on a strategic level.
For people reading this who have or are considering moving to a startup, what advice would you give someone?
You need to be involved. I think people underestimate how risky of a move it is. We know the company could die if we don’t hit our targets. I don’t think that’s a problem because the benefits that you get in terms of culture, career progression and experience are huge.
Here at BlackCurve, you have to be interested, you have to want to be collaborative, you want to work with people and rolls your sleeves up.
Wear many hats, as cheesy as that is.
Well, it’s true. In the last year, I have been technical sales, database cleaner, package installer, data scientist and product director, you really have to do everything. That’s how you learn and how you get better, especially from a Data Science background.
If you can develop commercial insight and continue to develop it, that’s the killer.
“The Unicorn Data Scientist”
Yeah, that’s it but if you can develop commercial awareness and can pre-empt solutions, that’s gold-dust. You can save a week in the business cycle, which is huge.
Be open, presumptive and be interested. I think it does suit people who want to be more generalist than a specialist.
Really? Because they have to do anything?
Right, I’ll go back to you with horrible corporate expression. You want to be T shaped, you first get your seat at the table with the knowledge. I started working for BlackCurve because they needed this skillset and this has allowed me to broaden out.
It’s always interesting speaking with founders, data scientists, whoever but it’s always the same, “wear many hats”. If you start saying it’s not my job then it’s probably not for you.
That’s not really an option! If you read a lot of literature around it, there are phrases like fire fast and hire slow. If you find the right fit for the company that fine but you look into yourself and ask “is this a company I actually want to work for?”.
Don’t work for a start-up for the sake of working for a start-up or if there isn’t a good personality fit. You’re going to be with those 5, 10, 20 every day in a tiny room day in day out.
Personality fit is absolutely key, and one thing start-ups struggle with it being able to define that, and, especially during scale-up, keeping a positive culture.
There’s about 10 – 15 working for us at any one time and we’re always having conversations about diversity and culture fit.
And I completely agree with you, if you don’t bake it in at this stage, it goes and then you have 60, 100 people in a toxic environment and you didn’t think about it from the start.
I think it comes from the top down.
Absolutely, its generated from the bottom up but it has to be projected from the top.
In terms of your product, what is price optimisation and are retailers using to achieve success?
This is the million-dollar question for us, what problem is BlackCurve trying to solve?
If I think of a traditional, online retailer there generally managing their pricing through Excel spreadsheet and they have 2 to 20 people who’re jobs are, for a day or two a week, to crunch through and change the price on the website, upload a CSV, and off you go.
You can already see, first and foremost, that a horrifically manual process and it often not what they should be doing, their marketers, buyers or merchandisers who are spending their time managing pricing and not going out doing the job that they are paid to do.
What this also means is that, because this is a weekly cycle and the way the internet works, it is that they don’t keep up with the competition.
For example, it’s summer and everyone in an electronics retailer warehouse shifting fans out because and it happens every year. Suddenly, your demand for fans go up, how do you respond quickly to your demand?
Realistically you could double the price of the fan and you’ve massively increased your margin. That where we come in, we automate that pricing process and using data around what your competitors are doing.
Using all of the data, we can suggest where your prices should be or push you toward where it could be, test on the marketplace and looking at the response. As a business, we are optimising or maximising the margin that you can make. We’ve seen on average 9x ROI on the software, and it’s nice to work with a company where it really works.
It’s worth stressing that it’s big data that we work on and part of what we’ve found it that we’re better off saying, if you think of the physiologically of somebody running an online store, you’re going to focus your winners, the 20% of products that you make your livelihood on and they kind of forget about the rest.
It’s results orientated thinking and actually your leaving the bulk of the margin, so we take the boring 80% and harvest the margin there and that’s we’ve found the most success.
Do you have a specific case study?
Yeah we do, with Donaghy Brothers, a leading electronics company based in Ireland (you can read it here: https://www.blackcurve.com/case-study-donaghy-brothers). Plus, many more here –https://www.blackcurve.com/customers.
AI is quite a vague term, but there are retailers out there who have implemented “AI” better than most, I’m thinking of Tesco and Marks and Spencers as examples.
I love the Tesco example. I sat through a talk about Dunnhumby a few years ago before I got into the space.
If I was to ask a normal person what three things that they are likely to forget, what would you forget, butter, milk, and bananas?
I’d say orange juice.
Well, the best thing about it was they performed super advanced data techniques to find out super simple answers. It would still come up with butter, milk, and bananas; my nan could have told you that.
I find that there’s a bit of backlash against AI and ML from investors, people have got burnt and it’s an interesting space to be playing in. At BlackCurve, AI and ML for us is an interesting one, the core engine in rules based so this is “1970’s AI” and the reason for that is explainability because if we’re setting a price to a customer, that customer needs to understand exactly why that price was set.
With a rules-based system, it’s straightforward. With a neural net or whatever, it nigh on impossible, people say that they can do it but I don’t believe them.
When we start doing more complicated stuff, it’ll be around data processing. Cleaning the data, coming in, looking for features in the data that can inform product decision or other stuff like that and that actually starts to become a much more tangible, commercially focused data.
It’s derisked by that and it’s more of an ML focused problem. We like to talk about explainable A.I. and really what it is the culmination of rules, heuristics, and modelling techniques so we can pick the right tool for the job.
You said a lot of VC’s have been burnt by AI? A lot of companies called themselves something AI, get millions in investment but never release a product. What do you think is the main issue with the AI industry?
AI problems are really, really hard. I would love to go away and build really robust, predictive AI models, but it’s incredibly difficult because you’re always thinking about the problem.
I’ll give you an example with us, it’s a new example of why predictive modelling in pricing is hard. If I were to go to an academically minded data scientist and “I’ve got this dataset and I wanted to predict price”, the problem is the dataset I’ve given you is not representative of how the marketplace and where we want them to be, it’s in the now.
That’s fine, I can build an ML model and learning algorithm that can learn the marketplace. That’s great, but actually, they haven’t changed any of the product prices in the last five years, where do you go from there?
What we do is, in theory, to analyse and segment the data to provide the jumping-off point so we can overcome that problem. We look at sales data and competitor data to inform where we should be going and then that starts a positive feedback loop, generating prices and data sets.
That is an answer where you need a huge amount of knowledge about your marketplace to solve that problem.
I think that’s the biggest problem is there are a lot of people developing AI and throwing it at a lot of problems without thinking if it’s an AI problem or not.
I’ve recruited for Data Scientists and have worked with a lot of AI companies, and I think they don’t always think of real-world solutions before actually implement super-advanced solutions.
As a final question, where do you see BlackCurve going in the next few years?
We’re in scale-up mode and we took investment earlier this year. The first question, is how do we scale? If we go from 10 to 50 to 100 to 1000 customers, what would that trajectory look like and how do we provide additional value for our customers?
Nothing is defensible forever, so one of the really nice things about this job, usually it is hard to find data-driven people. But because we need that data to show we are adding real value, we have a playground for a data scientist like me to start playing and then continue to add value.
A lot of the challenges that our customers face is that, fundamentally, they can’t afford a data scientist.
Or they are risk-averse if they want to hire somebody and they don’t work out.
Yeah, exactly that. Are you going to pay £30k a month for something that you don’t really understand?
You’re trusting this person to make all of these changes and when you don’t have that expertise already onsite, then how can you benchmark them?
The nice thing for us is we have that data and we can branch out and we can provide great insights and price recommendations, and collaborate with other people to make this service better for years.