MID-DECADE, 5 GROWING TRENDS.
Yearly digital trend study of innovation specialist Space and Lemon Innovations. 10th edition.
A study for the year ahead. This year’s trend study highlights 5 growing trends, obvious and non- obvious, trends to help find growth in a tough year. Trends to grow with. In 2025, technology and digital help find both growth and efficiency. Drawn by a formidable AI race and new forms of business models and habits growing. A mid-decade view, always optimistic.
Space and Lemon Innovations has been publishing a yearly digital trend study since its foundation in 2016. This is the 2025 edition.
2025, in search of growth in a tough year.
Let’s hold breath. We’re already mid-decade, and many geopolitical and economic challenges are ahead. This year’s trend study highlights 5 growing trends, obvious and non-obvious, trends to grow with. In a year where everyone in Europe talks about growth and questions business models. Here, a look at promising developments every student, researcher, professional and company should look at.
LLMs get better and better, downsides to the minimum.
Large Language Models (LLMs) have been getting better for months. They are very good at generating text, images, videos. And they can “read” mountains of words. Very large models, expensive to use. This has changed. Models of any size to fit usage, less costly, more flexible. Their training has brought better answers — bluntly said, less hallucinations. Vast amounts of training, at a point where almost all data in the world has been used.
A growing trend: the reliability, flexibility and affordability of the LLM. Exactly against the downsides criticised by many. Now, it opens doors for many and for diverse use cases — generating text and code, reading and dissecting documents. Even voice generation and task completion, real-time — this was a challenge. The makers of LLM have been successively working on new chunks. The next frontier is reasoning. Generating a logical chain of thoughts. It generates a simple thought after the other, like generating a word after the other.
Every maker of AI works on reducing the downside of LLMs and praising the upsides. A race of a few GenAI native actors like OpenAI, Anthropic, Mistral, Cohere. A race of tech giants partnering and launching better, new models (Google, Meta; xAI), even beyond their own squadrons building larger autonomy (Microsoft beyond OpenAI, Amazon beyond Anthropic). Results are fewer downsides and better LLMs. Humans, companies can imagine any use case, there will be an LLM solution.
Talk to a machine. And decorate the conversation.
From a human perspective, from a professional perspective, a LLM is about ChatGPT, about conversing with a machine. In written or spoken form. This will grow to a strong habit along with massive growth of LLM chatbots. Chatbots that are in no way comparable to the earlier NLP-based ones that have shown so many limitations.
Conversations to search, general search or search company documents for ex. Conversations to generate a text, an image, a website. Conversation to generate a code. Conversation in a call centre — 2025 will be the year of LLM in customer support. One can “talk to a machine” to even build one of those.
Reactions are mixed and it is important to control what the AI does. Above all when building a search chatbot for a company that retrieves information from a database. Coding an LLM chatbot in 2025 implies safeguarding what happens in the conversation. Controlling input (prompt) and output (answer) is key to gain efficiency, credibility and adoption in a professional context. Focus will be on how the bot is built and controlled. Safeguarding them. This will be solved in 2025, the prompts can be “decorated”, meaning forbid any unrelated topics, block credit card numbers, interpret prompts to spare the AI time and energy consuming interpretation. A bot is safe if we want to make it safe.
This automatically applies to AI-agents, a system where a bot can take over tasks when prompted. AI-agents have been the “word” of 2024 and will manifest itself in 2025. Exactly because the control of the conversation is advanced and done.
Ephemeral is the mood of retail and social networks.
It’s difficult to bring retail and social networks together. However, we observe a similar development. Ephemeral things are happening. They are dominated by short, ephemeral formats like stories, posts, promotions, events. Replacing traditional persistent narratives like brand building.
In social networks, content lifecycles are compressed into ever shorter windows. A shift from sustained brand campaigns to ephemeral experiences. Stories, ephemeral posts and ephemeral interactions become the primary forms of engagement. A consistent brand profile loses its importance. Do one ephemeral after the other!
In retail a fragmentation of traditional retail events and promotions happens. Amazon’s Prime Day is an example of this shift — what began as a single annual event has evolved into a complex tapestry of temporal touchpoints: Prime Week, Prime Day Big Deal, pre-Christmas events, Black Week and holiday promotions.
Promotions, driven by new retail from China (like Temu) are more ephemeral than ever. A discount, a game like the “wheel of fortune”.
An advocate is better than any paid promotion.
The end of a sales funnel, the advocate. They promote brands and products freely. The solution in a tough year. It has been fifteen years since marketing has focused on the marketing funnel at the beginning of it. Recruiting leads, converting them, having them to repeat… One can increase the performance by 0,1 percent. But it doesn’t bring enough. Today, sales funnels start at the end. With regular clients, friends, the advocates. The sales funnel upside down.
What is an advocate? It is difficult to put trust in an overloaded social media network. See the logic of the TikTok algorithm. An advocate is sincere, a friend of a brand, or a client of a product, who had a good experience, an exceptional experience. And tells about it. No one can buy an advocate. An influencer is not an advocate.
The upside of focusing on an advocate in a tough, busy, and ephemeral world. Advocates, driven by genuine enthusiasm rather than financial incentives, generate word-of-mouth recommendations, re-postings, clicks that eclipse any paid promotion. Their sincere endorsements, amplified through social networks and media channels, create brand moments that reverberate across the digital landscape.
The concept of adaptogens. An adaptation and an alternative.
Originally, adaptogens are substances that adapt to a specific body need, a large trend we observed on our tech travels in the US. Adaptogens help the body resist stress, they can be a molecule, a vitamin, a mushroom. All of it being propagated and targeted on social networks. They represent an alternative to consumer goods based on indulgence.
This has become a broader concept — products or services designed to adapt to changing needs and unpredictable markets. An adaptogen is an alternative to a product or a product category. Flexible, functional, and positioned against traditional offerings. It adapts to the consumer. Solutions have to be personalised and responsive.
An example? Revolut. An adaptogen. It is an alternative to traditional offers and is conceived to adapt to the market. Quickly. A flexible alternative, designed for digital, mobile-first users. Adapting to changing financial habits with personalised services. And growing. 50m customers in November 2024.
In 2025, markets will become even more dynamic, sometimes unpredictable, Adaptability will be critical for staying ahead.
Technology and digital help find both growth and efficiency. Drawn by a formidable AI race and new forms of business models and habits growing. A mid-decade view, always optimistic.
Space and Lemon Innovations. The trend + AI specialist and optimist. In Hamburg, Berlin and everywhere where tech happens. Permanent scouting close-to-business.
Hamburg/Berlin, 16.01.2025
Authors:
Joshua Barthel (LinkedIn)
Stefan Barmscheidt (LinkedIn)
Laurent Burdin (LinkedIn)
