top of page
Search

How to sell tuna

  • 2 days ago
  • 6 min read

Wattle Labs has created a "Large Behaviour Model" (LBM) of the human population. The LBM, an analogue of LLM's, which seeks to understand human behaviour, rather than language.


Our AI/ML driven psychological and behavioural profiling can help you catch fish!




Transcript

Hello everyone in Australia who really desperately needs to sell tuna. My name is Aaron and this is Wattle Labs. Today I'm going to show you how to sell tuna using AI and machine learning. Wattle Labs has created a large behaviour model of the Australian population. This covers 100% of the population — that is, all 28 million people who currently reside here. We have been verified to be 92% accurate with all our predictions, which is 60 predictions per person for a total of some 800 million predictions of the population of Australia. This demo is very rudimentary as far as our technology goes, but it could have profound effects for business or even charities.


So I'll just get into it. We're going to show you how our customer profiling technology works — that is, profiling each and every individual's behavioural, psychological, socio-economic, and demographic attributes based off very simple information. Our model is keyed on simple information that should be available to every company in Australia about their customers: their age, sex, and location. We're going to take that and do a simple segmentation to help you sell tuna.


In this example, imagine that on Monday morning your boss comes in and says, "If we don't uplift our tuna sales by 50%, you're all sacked." So the first thing everyone does on Monday morning is panic. Then they talk to the data scientist and say, "We need to sell more, therefore we must run a campaign." So what's the best type of campaign we can come up with?


Modern data science would probably say: you've got a bunch of customers and you've got sales data — this is how all your data science really works. What we need to do is segment the customers into four groups. We find the customers in group one who last purchased Simply Tuna, or most often purchase Simply Tuna, or purchased it within the last seven days. We do the same for John West, Sirena Tuna, and Wild Tides. So you're segmenting all of your customers into these four buckets. If you've got anyone left over who has never purchased tuna, you might just randomly assign them to a bucket. And you go, "Great, we're going to run this massive campaign this week and make 50% by Friday." So you run the campaign — and then on Friday your boss comes in and simply says, "You are fired."


The reason you're fired is that most data science leads to this type of outcome, where you're simply advertising to people who are already interested. Google does this all the time — advertising something to you that you're just not interested in, or something you've already purchased. In this case, everyone panics and you run the Simply Tuna campaign for people who already purchase Simply Tuna. It's not generating any new revenue. The marketing campaigns will be extremely successful in click-through rates — the person already eats tuna, they're going to purchase it anyway, you put a 10% off offer in front of them and they click the ad. It gets a huge click-through rate, but it gets no new revenue. And because we needed revenue to not be fired, we all got sacked. That's a really bad result.


Using psychological and behavioural techniques — AI and machine learning — we can discover a lot more about customers to greatly enhance which ad we put in front of which person. So let's go back to the beginning. What we notice about these four tuna brands — and it doesn't have to be tuna, it could be hi-fis, shoes, anything at all — is that Simply Tuna is relatively $1.00, John West is twice as expensive, Sirena Tuna is approximately $4.00, and Wild Tides is approximately $2.00. We've got a really cheap one, a really expensive one, and two in the middle.


We can also look at the positioning of these brands. Simply Tuna is positioned as the budget option. John West is positioned as "the best" — "John West, buy the best." Sirena Tuna is positioned as premium; by virtue of being four times more expensive than the budget version, it positions itself that way. And Wild Tides positions itself as the green, environmentally friendly option.


Now, to make money and save our jobs by Friday, we're not just going to run campaigns to get clicks — that's not going to help anyone. We're going to have to move people who are currently buying Simply Tuna to one of these three alternatives. The problem is, your data science departments and third-party vendors have no understanding of which alternative to push a Simply Tuna customer towards. That's where our behavioural modelling and psychological profiling comes in.


What we can do is take the cohort of people who currently always buy Simply Tuna and segment them into three buckets using our behavioural modelling techniques. First, we find the high income earners — there's a high correlation between high income and Sirena Tuna, because it's a premium-priced product. Next, we find the people who have a product and quality focus — these are your John West customers. Finally, we find the people who have a green, environmental focus — these are your Wild Tides customers.


So what we're doing is taking a person who would have spent $1.00 and, if our profiling technology holds — and it's been verified at 92% accuracy — we have roughly a one-third chance of moving them up to $4.00 (Sirena), a one-third chance of moving them up to $2.00 (John West), and a one-third chance of moving them up to $2.00 (Wild Tides).


So our expected revenue per customer is $2.66, compared to the original $1.00 — an increase of 166%. It's never going to be exactly that good; some people simply won't move. But at least from a theoretical position, as long as the psychological profiling holds, we're looking at increasing sales by 166% on this campaign. And I'm not talking about click-through rates that aren't making you money. I'm talking about psychologically profiling the people in one cohort and nudging them into another cohort that delivers a higher return for the retailer. The same approach works for donations, or across any sector. So by Monday lunchtime we're starting from a position where we have a theoretical return of 166%. We can tell the boss, "We're looking at a 166% increase in revenue this week." It'll never be quite that much, but at least with the previous model you were looking at a net loss.


Now I'll show you how this works in our software. We've profiled the entire Australian population — Australian charities, Coles locations, FMCG, political parties and electorates, companies, charities, schools, and even small businesses. In this example, we've taken all charity donors — this is data from 2019 from Tasmania. Greg Simmons is a fake name; I've replaced all the names so there's no personally identifiable information. But at an individual level we can profile and get very deep socio-economic, demographic, psychological, and behavioural information about each and every person in a cohort.


Taking our Simply Tuna cohort, our first step is to find the high income earners. That's as simple as querying, "Please find high income earners." There are 29 of them, and we can download that as a CSV file — that's our first campaign. It's worth noting this cohort is a little skewed because, as it turns out, high income earners tend not to donate to charity. Next, for John West, we find people with a focus on products and facts — that's 370 people out of the original cohort. We download their data as a CSV, which just contains the customer ID you can pump straight into your marketing engine. Finally, for Wild Tides, we look for the environmentally conscious — the greenies. Who in the Simply Tuna cohort is actually a closet environmental crusader? Filtering for a focus on environment and community gives us a whopping 94,000 people — roughly two thirds of the original cohort. There's a huge correlation between people who donate to charity and those who care about community and the environment. Who would have thought?


And that's it. It's not even midday on Monday and we've done it. We've got three campaign lists — campaign one, campaign two, campaign three — ready to go as CSV files with your customer IDs. The front end of the process is the same as normal: identify who's in the Simply Tuna brand. The back end of executing the campaign is the same as normal. What we're adding is the psychological and behavioural profiling in the middle that greatly enhances your sales outcomes. It's obvious, really — if you're trying to sell a premium product that's four times more expensive than anything else, the greatest predictive factor for that sale is income. The highest income person is most likely to purchase the most expensive product.


Sure, we're not going to hit the theoretical 166% uplift in revenue — but the result by Friday will be "not fired," and maybe even "promotion." That's a much better outcome. Anyway, that's all I want to say on that. This works across all sectors — charity, FMCG, luxury goods. Let's all not get fired and get promotions instead. Thank you, have a great day. Goodbye.


 
 
bottom of page