Seven hundred years ago, a golden tablet let a stranger carry the trust of an emperor across an empire. Today, our systems carry information much more easily. Trust is proving harder to move.
The tablet that made trust portable
In 1291, after seventeen years in his service, Kublai Khan finally let Marco Polo go home. There was a condition attached: Polo would escort a Mongol princess to Persia by sea.
Before he left, the Khan gave him one more thing, a golden tablet a foot long and three inches wide, inscribed with a warning that anyone who failed to honour its bearer would be killed.
Polo carried it the whole way home. It was still among his possessions when he died in Venice more than thirty years later. The tablet was called a paiza. A traveller carrying one could arrive at a relay station he had never seen, in a province where nobody had heard his name, and receive horses, food and lodging from an official who knew nothing about him personally.
No letter of introduction. No shared history. No negotiation.
It is tempting to read this as a story about medieval logistics. It mostly isn't.
The official did not need to know Marco Polo. He needed to know what the tablet meant. And that is a much harder problem to solve than it sounds, because meaning does not travel the way objects do.
How does trust cross a distance?
The usual image of the Mongol Empire is conquest: horses, sieges, terror. The more interesting part of the story, the part the historian Jack Weatherford spends most of his book on, is what the Mongols did once the fighting stopped.
They ran the largest contiguous land empire in history across dozens of languages, religions and legal traditions, and they governed it with a strategy that had almost nothing to do with force. Genghis Khan recruited engineers, scribes, administrators and translators from the populations he conquered, regardless of their ethnicity or religion, and promoted them on ability rather than lineage.
Religious tolerance was official state policy, not a courtesy. Whatever administrative or agricultural practice worked, Chinese, Persian, Muslim or otherwise, got adopted and spread across the empire without much concern for where it came from.
The Mongols never tried to make the empire's subjects culturally Mongol. They built a system in which very different people could become legible to one another, and the paiza was one small, physical piece of that system: a signal an illiterate frontier official and a Venetian merchant could both read the same way, without sharing a language, a religion, or a government.
That is the real achievement. Not the tablet. The legibility.
Culture is what turns a signal into meaning
Here is a version of the same problem, seven hundred years later:
• A D2C brand's order is delayed by two days. To one customer, a tracking link with an honest revised date reads as competence. To another, the same tracking link reads as a brand hiding behind automation, and what she actually wants is a personalised communication, before she has to ask.
• A snack brand had a bad batch reach a small set of customers. One customer tagged the brand publicly and stayed engaged only once the brand replied in the same public thread, apology, correction, timeline, visible to everyone who'd seen the original complaint. Another customer, in the same week, DM'd privately about the exact same batch and would have found a public reply mortifying. She wanted it handled quietly, by a person, and never mentioned again.
• A personal-care brand selling largely on COD across Tier 2 and Tier 3 towns tried, twice, to nudge customers toward prepaid checkout with a small discount. In its metro cohort, conversion barely moved, those buyers were prepaid already, discount or not. In its smaller- town cohort, prepaid conversation actually fell for a few weeks. Removing the "pay when it arrives" option hadn't removed a payment method. It had removed the one signal those customers used to decide the brand was real.
• A fashion label's founder used to personally reply to a handful of reviews every week, and for years, that was read as care. Then a repeat customer, a genuinely high-value one, said something the founder didn't expect: the personal replies made her nervous, they read as under-resourced, like there was no real support function behind the brand. What she trusted instead was a refund that arrived inside the stated SLA without her having to negotiate it.
The information can be identical. The meaning isn't. That gap is where cultural intelligence actually lives.
None of these customers is wrong. Each one is reading the same category of signal through a different set of assumptions about what a trustworthy brand looks like, and the founder who notices the difference keeps the customer. The one who doesn't, quietly loses her, usually without ever finding out why, because she didn't complain. She just didn't reorder.
Then the brand scales, and forgets what the signal meant
As a D2C business grows, it starts trying to standardise these signals, and there is nothing wrong with that impulse.
• The customer goes into the CRM.
• The complaint becomes a ticket ID.
• The angry DM becomes a CSAT score.
• The COD customer becomes a risk flag.
• The relationship becomes a set of fields. But what gets lost is the memory of what it meant.
A support lead knows a particular repeat customer will forgive a delay if told before she asks, and will not forgive it if she has to chase. A community manager knows one influencer's audience reads a slow reply as disrespect, and another's audience won't notice for a day. A founder knows exactly which one-star review is really a loyal customer having a bad week, and which one is a customer the brand should let go of.
None of that lives in the system. It lives in a person, often in one person's head, sometimes in a WhatsApp thread nobody else has read. For a small brand, that is enough, right up until that person resigns, goes on leave, or the brand simply grows past what one inbox can hold.
Then the brand finds out it never built a trust system. It had trusted individuals, and those are not the same thing once they are unavailable.
This is the problem AI is about to inherit
Software used to wait for people to operate it. A support agent read the ticket, checked the order history, made the judgement call.
AI changes the direction of the relationship. A support agent can now read a chat history, draft a reply to a complaint, approve or deny a refund, and increasingly, decide on its own what to do within a defined boundary.
That sounds like an enormous leap in automation. It is. But there is a harder problem underneath it.
• An AI agent can know an order is two days late. Can it know whether this particular customer needs an apology or just an ETA?
• It can know the return window closed four days ago. Can it recognise when this is a customer worth an exception, and when it isn't?
• It can see a complaint posted publicly. Can it tell the difference between a customer who wants an audience and one who wants privacy?
• It can retrieve six months of order history. Can it tell which of those orders actually mattered to the relationship?
This is where the distinction between information and trust becomes important. The data may be universal.
The meaning isn't.
The next interface is not language. It is context.
This is why the future of AI agents in D2C will not be decided only by how well they write, reason or retrieve information. They will have to understand the environment in which their actions are interpreted.
An agent replying to a customer on WhatsApp or Instagram needs more than the order details. It needs to understand what this brand has promised, what this particular customer expects, the history between them, and which signals establish confidence for her specifically, not for the average customer in the dataset.
An agent working a support queue needs to understand the difference between what the refund policy formally allows and what will actually keep a good customer.
An agent handling a complaint needs to know when "this is fine, don't worry about it" from a customer means exactly that, and when it means I am about to leave and never tell you why.
This is cultural intelligence at machine speed. And it changes what we should expect commerce software to remember.
Not just: who is the customer? But: how does this customer decide she can trust you?
Not just: what happened last time? But: what mattered about what happened last time?
Not just: what does the policy say? But: how does this brand actually behave when the policy meets a real, upset human being?
The paiza was a primitive form of institutional memory
The golden tablet did not contain Marco Polo's personality. It did not explain his entire history. It carried something smaller but enormously valuable: enough recognised authority for an unfamiliar person to enter a system without starting from zero.
That is what makes the object so useful as a metaphor for a modern consumer brand.
A brand also needs ways for trust, authority and judgement to travel. From founder to support agent. From support agent to customer. From one city to another. From a first-time buyer to a repeat one. And increasingly, from human to machine.
The challenge is not to make every customer read signals the same way. The Mongol example suggests almost the opposite. Their empire worked partly because it could absorb different ways of reading the same signal, rather than forcing everyone into one inherited model of what proof should look like.
The challenge is to build systems capable of carrying meaning across those differences.
Where this actually goes
The most interesting D2C systems being built right now are not the ones that simply automate a reply or approve a refund faster. They are the ones that try to make context portable: the memory of a relationship, the logic behind a past exception, the signal that once told a support agent this customer needs a call, not a template.
Think about the person in your team customers trust most. Maybe it's the support lead who always knows what to say to the customer who's about to churn quietly. Maybe it's the founder who somehow knows which one-star review deserves a call and which deserves nothing at all.
That is organisational intelligence. It is also cultural intelligence. And most brands spend far more on their checkout flow than on documenting this layer, even though it's the one that decides whether a customer becomes a repeat one.
The tablet was small. The problem wasn't.
The Mongol Empire needed a way for authority to travel farther than the people who created it.
A growing D2C brand faces the same problem with trust. The difference is that the distance is no longer measured in kilometres between relay stations. It's measured in customers, cities, support shifts, festive-season order spikes and, increasingly, the replies an AI agent sends on the brand's behalf while nobody is watching.
We built technology to move information across all of that. Now we need technology that can move meaning.
The paiza was never the trust. It was the mechanism that allowed trust to cross a distance. Seven centuries later, that problem is still here.
What do you think?
• What does your brand currently treat as proof that it can be trusted, and would your most loyal customer name the same thing?
• Which of your team's judgement calls, about when to bend a policy, when to call instead of message, when to let a customer go, exist only inside one person's head, never written down anywhere a new hire or a machine could learn from?
• When an AI agent replies to your customer on your behalf, whose judgement is it actually using, and is that the judgement you'd trust with your best customer?
We tend to think technology makes commerce more connected. It does.
But connection is not understanding.
Trust travels only when its meaning travels with it.
Sources and Further Reading
• Jack Weatherford, Genghis Khan and the Making of the Modern World (2004): the central source for Mongol meritocratic administration, religious tolerance, and the relay systems that made the empire governable.
• Rustichello da Pisa, The Travels of Marco Polo (c. 1300): the primary medieval account of Polo's years at Kublai Khan's court and his departure with the golden tablet.
• Ata-Malik Juvaini, The History of the World-Conqueror: a Persian-language administrative account of Mongol imperial practice from within the empire.
• William of Rubruck, The Journey of William of Rubruck to the Eastern Parts of the World: an earlier European traveller's first-hand record of the Mongol relay and travel-permit network.
• Morris Rossabi, "Communications and 'Communicators' in the Yuan and Early Ming," Journal of Chinese History (2025): recent scholarship on communication and administrative systems under Mongol and Yuan rule.
• Erin Meyer, The Culture Map (2014): a modern framework for how identical business signals, feedback, disagreement, silence, are read differently across cultures.
• Geert Hofstede, Culture's Consequences (1980): foundational research establishing that what counts as trustworthy behaviour varies systematically by culture, not individual preference.
• Mordor Intelligence, "India D2C E-commerce Market" report (2026): documents 25-30% cash-on-delivery return rates and the working-capital stress this creates for Indian D2C brands.
• Unicommerce, "India D2C Report 2026": platform data across 6,000+ Indian D2C brands showing COD returns running as high as 58% in the festive quarter against under 15% for prepaid orders.
• SGA PR, "The Shimmering Story of D2C: From Attention to Acquisition" (2026): on how trust, community, and Tier 2/3 discovery are reshaping India's next wave of D2C brands.
