Your AI calls UC when a question involves culture.
In development
This offering is in development. We are building it now and will open it first to a small number of companies. If your company would like to be among them, write to John Magee at john.magee@understand-culture.com.
Your AI isn’t enough
For the first time, an AI can help people understand cultural differences while they work, at the moment they need it. But it cannot do this on its own.
Left to its training, your AI repeats the etiquette tips and the stereotypes it has seen most often. It also agrees with whatever view the person brings to it. The answer sounds fluent and confident, but it is shallow.
That is the first problem, and it is easy to see. The second problem is harder to see. Your AI can only answer the cultural questions people think to ask, and most are never asked.
When something goes wrong between colleagues, people usually think the other person is either incapable or unwilling, sometimes both. They do not suspect culture. So your AI looks reliable, because it is judged on the questions it receives, not on the many it never receives.
How it works
An engineer in the United States writes to a colleague in Germany. Something between them is not working. He asks your AI assistant for help.
Your assistant sees that the question involves culture. It identifies the country and the topic, and then asks UC.
UC sends back the research. It says what the pattern is and how it appears in daily work, and it includes a link to the page it comes from.
Your AI assistant writes the answer and includes the link. The engineer reads it in the tool he was already using. If he wants to see the analysis and the evidence, he clicks.
Each call carries one country and one topic. A typical question leads to two or more calls.
Why it’s better
An AI that answers from its training alone has one basic problem. It is guessing. Give it researched material to read, and the guessing stops.
It cannot check facts. An AI has no way to verify what it tells people. With UC it does not need to. Human beings research the material, write it, and check it before your AI ever sees it.
It cannot judge sources. Left to its training, an AI reads advice articles, blog posts, expat forums and marketing material. The claim that appears most often wins, not the claim that is most accurate.
UC works differently. The research is human-led. We choose and weigh the sources before your AI sees them.
On the topics researched to evidence level, every statement carries its source directly underneath it, on the page. It may be a law, a study, a court decision, a public statement or a news report. Anyone can read it and check it.
Germany and the United States have the most of this, and other countries are being brought to the same standard, topic by topic.
It reads one culture through another. This is the most important point. Left to its training, an AI describes a low-documented culture using the assumptions of a high-documented one. Our research is done the other way. Each culture is described on its own terms.
Its knowledge is fixed. A language model stops learning on a certain date. Our research does not. We add to it and improve it continuously.
It invents. When an AI has no information, it fills the gap, often making things up. With our research in front of it, there is little left to invent.
It can show where the answer came from. A correct answer and a wrong answer can look the same when the source is hidden. Here the source is named. Your colleagues see that the answer comes from UC’s research, and the link takes them to the analysis.
What you get
The intelligence. The intelligence is research on how people in a culture think and work. It explains the reasons behind the behavior, not etiquette. It is written in plain language and organized so that an AI can read it and use it directly.
The recognition rule. The research only helps if your AI knows when to use it. We supply a draft recognition rule to start from, and we improve it as we see how it performs across clients. Your teams adapt it to the way they work.
Access for your people. Anyone with an email address at your company can register on UC and read all of the research, country by country, topic by topic. There is no charge, no seats to buy, and no list to maintain. It is part of the agreement.
The site also has a culture advisor for talking a situation through. That is a separate personal subscription, and your people are free to take it or not. Reading the research needs nothing beyond their work email.
What you give
Each time your AI asks UC, it sends two pieces of information, one country and one topic, and nothing else.
It does not send the question, the message, the situation or the person. What colleagues write stays inside your system.
Your own team writes the connection to UC, so you do not have to trust us on this. Your people can read their own code and see exactly what it sends.
UC has no access to your system, no account in your environment, and no software running on your side. If UC does not answer, your AI writes its reply as it would have done anyway.
What it costs
You pay one fee for each call. There is no charge per user, there are no tiers, and the fee is not negotiated. Under the fee per call there is a fixed annual minimum. If your calls cost less than the minimum, you pay the minimum. If they cost more, you pay for the calls.
Access to the research for everyone at your company is included.
We invoice every three months, after the fact, based on UC’s record of the calls. You can see that record. It shows how many calls were made, which countries and topics they covered, and how the number changes during the year.
That record is useful on its own. It shows you where cultural difficulties actually arise in your organization, measured rather than guessed.
The fee per call does not rise as your use grows. You are not penalized for using UC more, and there is no renegotiation when the agreement renews.
Where it stands
The research exists and is used every day through our own advisor. What is new is this way of reaching it. No company is using it as of yet, and you would be the first.
Nothing here is difficult to undo. Your own team writes the connection. If you stop, you remove the rule and the calls stop. There is no system to shut down and no data of yours to recover.