Spare a thought for marketing bods — who are, after all, human, even if they don’t always act like it. You know, those folk with the unenviable job of ensuring brands stay on message; outperform the competition; and are simultaneously communicating effectively with customers of all stripes across multiple ‘touch-points’, as they like to call the business end of their output.
Since language is subjective then measuring brand messaging — as marketers apparently must — is akin to putting your hand into a snake pit in the hopes of pulling out a docile python. Which is to say both darn hard and sweating toil.
Bottom line: Language is not the sort of stuff that lends itself to easy performance review via quantified dashboard. Clicks yes; complex sentences not so much.
But never fear, marketers! UK startup Relative Insight (previously called Relative.ai) reckons it has come up with a fix for your linguistic horrors — by figuring out a way to turn language into data sets that support comparative analysis. So, in other words, it’s saying its all-seeing language-crunching algorithm is better than your expert marketing messaging opinion.
Relative Insight is launching in full today, having been applying its technology with a small group of beta customers for the past nine months. Clients have included ad agencies, such as Ogilvy, and brands themselves, such as Microsoft. It’s aiming to work with both.
“Essentially what we do is we take text, that’s language in text format, and we turn it into quite complex data models using a combination of technologies. Think of it as just turning language into data. And then what we specialize in doing is comparing those data files that we produce — comparing two data files with each other — and then seeing what the similarities and differences are between the two data blocks,” says CEO Ben Hookway explaining how Relative Insight transforms marketing missives into standardized data that can be usefully analyzed.
“The crucial bit is understanding the differences between two relevant bodies of text,” he adds. “For example, one thing we do a lot of for clients is brand benchmarking. A brand will put a lot of effort into its outbound communication. Its style of writing and what it tends to talk about but that’s a lot of hard work and in the process of getting that done and doing all the copywriting brands quite often forget that they are perceived in a context which is their competitors and the market.”
What this argument boils down to is that brands can easily end up being so focused on their own output that their messaging ends up sounding identical to other industry rivals — e.g. retail banks parroting the importance of trust or security, for instance — ergo the brand fails to gain any competitive advantage because it’s not standing out from the competition.
Relative Insight’s ‘fix’ for this problem is conducting algorithmically powered “contextual comparison” of brand marketing output — to allow one brand to measure its own marketing messaging against rivals.
The technology also lets a brand compare its own messaging across multiple channels to ensure its own output is on-message. Or look at how effective its messaging is at talking to different groups of customers. Or how different groups of customers are responding to its products. And so on. Once you have a standard model to compare use of language there’s clearly no shortage of areas to put it to work analyzing — not just the brand’s own output; its customers’ direct communications with it are also up for grabs.
“Our clients have never been able to measure language — in marketing it’s a very subjective thing… Everyone’s got an opinion. It all tends to be very instinctive,” says Hookway. “Trillions of dollars must get spend on a subjective opinion. That is ‘what language is the best to use?’.
“What we do is come along and measure all language using one standardized model which then allows our clients to make objective comparisons between language — and we provide them data. So we turn language, which is a subjective expression, into an objective data source.”
As a wrangler of words myself, that, frankly, sounds terrifying. So I for one am hoping this startup realizes there’s no money to be made from trying to flog its services to publishers to quantify journalists’ outputs — and rather stays its course targeting marketing’s trillions. Truly that’s where the gold is in them thar hills, trust me.
Relative Insight (née ai) was spun out of Lancaster University back in 2012 but the underlying research into building language models powering its latest venture dates back around a decade. The technology was actually initially developed as a cybercrime-fighting tool (under the name Isis Forensics) to help identify criminals in policing areas such as child protection and counter-terrorism.
But applying the model in the marketing space clearly offers orders of magnitude more revenue-generating potential, especially given the fragmentation of marketing across myriad channels. Hence Relative Insight’s new focus.
Its technology can be used to analyze and quantify any textual output — whether that’s the text on a billboard ad, the brand backstory on a website, the updates on your corporate Facebook page, your outbound customer-facing emails, whatever it is so long as it’s text-based — which it argues sets it apart from rival tools that focus specifically on social media monitoring, for instance.
It also argues that sentiment analysis tools can’t handle the same level of linguistic complexity as its language model can. Specifically, this model looks at a combination of word use, semantic use and grammar to create its standardized model, says Hookway.
“It’s quite a complex mechanism that we use. This is borne out of 10 years of R&D in the language model,” he says. “The technology is actually used in law enforcement, and we do a lot of work where — particularly in cyber crime — where people are frequently pretending to be somebody they are not… We can tell in a chat room whether you’re a 12 year old girl, or whether you’re a 50 year old man doing a very good impression of a 12 year old girl. And we do that by the comparison of the language sets.
“What we do is complex, in that we take account of all kinds of factors. And to do this automatically at scale is actually quite tricky. For example, the word ‘break’ has got something like 76 different possible meanings in the English language, and we take account of context — how it’s used in context… in semantics and so on — and we model at that kind of complexity.”
Another differentiation point then: Relative Insight can work at serious scale.
“We frequently analyse something like a million word and more bodies of text,” he adds. “And to put it in context, the Harry Potter series of novels is about a million words.”
While it’s focused on text (and currently only supports English), the technology could also be applied to contextually quantify and analyze video content, according to Hookway, provided the language used in video content was first transcribed to text. That potentially unlocks another big swathe of marketing content it could be paid to quantify.
The startup has raised an undisclosed amount of funding from Manchester-based VC EV Group. Its initial focus is on the U.K. market but Hookway says it’s likely to move quickly to open up in the U.S., after fielding interest from across the pond. He has a trip to Silicon Valley and New York planned for this summer.
The business is self-described as a managed service — meaning once it has analyzed the brand’s texts it presents them with the raw data, flagging up “critical differences” between the two compared bodies of text to give its clients a clear path forward to improve their marketing messaging (or that’s the promise, anyway).
“Quite often it’s pretty obvious [what a brand needs to do],” adds Hookway. “There will be glaring cases where one brand is for example not using pronouns enough. They’re not including their customers enough, they’re not trying to talk in an inclusive way. Because the brands are experts in their own market as soon as they see the data — and it is data, and it is factual — it’s pretty clear to them what they need to do.”
Of course language is remarkably slippery. It’s alive, mobile, ever-changing — shedding words and meanings, and evolving shiny new ones to take their place. So it remains to be seen how much linguistic complexity Relative Insight can really get a handle on — and how insightful the linguistic insights it delivers really are. The proof of the pudding is in the eating, as they say.
Creative copywriters working in marketing may well feel their envelope-pushing words are incomparable. Relative Insight clearly thinks otherwise.