Andrew Stephen holds the L’Oreal chair in marketing at Oxford’s Said Business School, where his group works on the intersection of marketing and technology. His framing for what a marketing leader has to track is borrowed from the theater: there is the front of house, which customers adopt and pull marketers after them, and the back of house, which is harder to see and harder to time.
His advice on the learning curve is deliberately modest. A leader does not need to become technical. A leader needs to know which questions to ask, and to be able to tell a real answer from a pitch.
Key findings
- Stephen’s central practical point is that most of these systems cannot be bought and turned on. A customer care chat bot has to learn the questions, the answers, the register of an answer, and when to hand the conversation to a person, and all of that is learned from data the company must already have.
- In the cases he has looked at, accumulating enough of that data took two to three years, and the inputs are call center transcripts, email logs and website answers rather than anything purpose-built.
- That produces the leadership problem he names directly. The investment has to be made now for value two years away, it is unglamorous, and it is hardest in exactly the organizations that are judged quarterly.
- His way of making that investment defensible is to refuse to treat it as one bet. The same processed data that eventually feeds the chat bot can be read now for sentiment, emotion and topic, which is the job a social listening dashboard already does and already has a budget.
The questions to ask, and the ones to ask about people
Stephen is not cynical about vendors. His objection is that a pitch is not aligned to your strategy by default, and that a technology project that sits beside the strategy rather than inside it will not produce anything.
The question he says he would always ask is about people rather than software: what do my people need to be able to do, where would those people come from, and what else would they work on once they are here.
One answer is a reason to keep the work inside. Where the algorithms are trained on your own customer data, he suggests, the result may be the advantage itself and worth not showing anyone.
Where to seat a data scientist
Stephen draws the parallel with embedded insights teams, and then breaks it. A data scientist is not a market researcher with new tools. They are computer scientists, and they do not share a vocabulary with brand people.
That gives two ways to fail at once, and he names both. Put them in a distant silo and nothing reaches the brand. Embed them alone in a group where nobody speaks their language and they will not survive it.
Start with what you want to be able to do
The interviewer recalls a researcher at Unilever asking him, after every question he brought, what he would do differently once he had the answer. Stephen agrees that the same discipline applies here, and reduces it to two cases.
He prefers to hold the whole question inside innovation rather than technology, because that keeps a customer in the objective and keeps the shiny object out of it.
Culture does not carve up the way marketers expect
Asked how these interfaces land in different markets, Stephen is careful about the limits of his own evidence. His chat bot studies cover Western consumers in German-speaking Europe, the UK and the United States, and found no large differences among them.
Beyond that he offers two observations rather than findings. There are real considerations, including religious ones, about speaking to something that is not a person. And the familiar cultural stereotypes marketers reach for do not predict any of it.
He adds the correction that matters most to the audience: parts of Asia are further ahead on this than either of the markets in the conversation.
Technology in the service of belonging
Asked what actually makes a brand more human, Stephen starts from what everyone in the conversation is doing at that moment: using communications technology to stay connected while confined. His formulation is technology enabling something people already want rather than technology for itself.
His marketing example is the shift in what a brand community is. It used to mean owners meeting physically at a festival. It now means people who will never meet finding each other around a shared interest, with influencers as a further kind of participant.
The human tenet he attaches it to is belonging, and he is explicit that the same technologies have enabled harm.
Who should own a digital transformation
Stephen’s group studied more than forty chief digital and data officers. Very few came from marketing. Most came either from another business function or from an engineering and technology background.
His answer to who should own the work is neither the CMO nor the digital officer. If a transformation is genuinely about how the business operates, it cannot sit with one function, and the chief executive has to hold the top of it.
What the marketing leader supplies is the argument, and he reduces it to the arithmetic of a customer base: get customers, keep them satisfied, raise their value over time, and then ask where technology enters that.
Mass upskilling is the start of the journey, not the journey
Stephen sits on a standards board that shapes enterprise digital curricula, and believes in them, on the grounds that a transformation needs a bottom-up half and a common vocabulary is what makes the rest easier.
His caveat is sharper than the endorsement. Skills that come back to unchanged work decay, and the illustration he picks is deliberately absurd: teaching someone a new discipline and then handing them the old brief.
The interviewer adds the comparison that stings. Accountants requalify. Supply chain has been rebuilt repeatedly by its own methods. Marketing has largely been allowed to treat a degree as the end of its training.
The algorithm that would take diversity off the page
The ethical case Stephen works through is concrete. An advertising system optimizing for attention can learn, from clicks alone, what kind of face a particular viewer looks at, and generate people accordingly.
The reason this is not a neutral optimization is the finding underneath it. People tend to attend to people who look like them, so a system maximising attention converges on sameness, and he calls the result the visual equivalent of a filter bubble.
The direct harm is that this runs against the only mechanism known to reduce the bias, which is being shown people who do not look like you. Stephen sets that beside a second question he does not resolve: whether people who do not exist should be in advertisements at all.
His remedy is policy rather than guidance, because a guideline assumes a shared moral compass that a large industry does not have. He points at industry-level work then beginning with the International Chamber of Commerce, and concedes it will not stop a determined bad actor.
Do not run it as a sprint
Asked for one thing to hold onto, Stephen picks the risk that the crisis mode itself becomes the strategy. His warning is about exhaustion and about a horizon shrinking to what can be done this week.
The interviewer’s closing reading of the conversation is that these systems, left to themselves, will reproduce and intensify the world they were trained on, which is the opposite of what the moment requires.
Scope and limitations
This is a conversation recorded in June 2020, and Stephen marks the boundaries of his own evidence as he goes: the chat bot studies cover Western consumers only, and the account of rising brand trust is offered as an interpretation of somebody else’s survey rather than as a finding of his own.
Source
This page summarises HGS #5: Andrew Stephen (Professor of Marketing, Oxford), by Marc de Swaan Arons, 2020-06-08.