Antonym: The Last 5% Edition

How to use a 95% reduction in cost.

Not a real advert. But a real saving has been made.

Dear Reader

Goldman Sachs revealed something startling last week: AI has cut the time needed to prepare complex stock market documents by 95%. But the real story isn't about the cost savings - it's about what happens next.

When a technology promises to slash costs by 95%, the obvious move is to pocket the difference. But that's exactly the trap that could separate tomorrow's accelerated winners from its laggardly losers.

Goldman Sachs just revealed something startling: AI has cut the time needed to prepare complex stock market documents by 95%. But the real story isn't about the cost savings - it's about what happens next.

When a technology promises to slash costs by 95%, the obvious move is to pocket the difference. But that's exactly the trap that could separate tomorrow's winners from its losers.

The Goldman Sachs

Let's look at what Goldman Sachs is actually automating. When companies go public, they need something called an S1 filing - essentially a corporate autobiography mixed with a detailed business plan. It's the kind of document that can make or break a multi-billion-dollar IPO.

When a company goes public on the New York Stock Exchange, its bankers must prepare an S1 filing - a detailed registration statement that includes crucial information about the company's business model, financials, risks, management team, and planned use of the funds raised from the public offering (IPO).

Think of it as a company's detailed autobiography and business plan combined. It has to be extremely thorough and accurate, as it helps investors make informed decisions while protecting the company from legal issues around disclosure requirements.

The S1 for Stripe, the payments company that Goldman Sachs helped raise $6.5bn for in 2023 is 445 densely formatted pages, including pages like:

The numbers

Consider the numbers: A typical S1 filing might cost around $1 million to prepare (lawyers, accountants, compliance experts and an army of interns) and would be billed at $2-3 million to clients. With AI acceleration, that cost drops to just $50,000. For perspective, that's like turning a month-long project into a day and a half.

If we call the costs of creating the S1 $1 million pre-AI, then it is now $50,000. It's likely the S1 is part of a bigger service from Goldman, the cost absorbed as part of a commission on an IPO or raise.

What to do with a 95% budget refund

This isn't just about investment banking. At Brilliant Noise, we're seeing similar patterns across knowledge work. While few tasks achieve Goldman's 95% efficiency gain, we regularly see 20-50% improvements in everything from proposal writing to data analysis. The question facing every business is the same: what do you do with these sudden efficiency gains?

Here's a simplified example:

We completed two comparable projects for the same client, two years apart. Each had a £10,000 budget to analyze customer behavior data and transform it into engaging content. Content production consistently took 10% of the budget. The remaining work fell into two categories: data analysis (finding and selecting potential stories) and creative development (ideation, decision-making, refinement, briefing, and design).

• In the 2022 project, 70% of the budget went to data analysis, which produced three viable stories. Creative development—including ideation and refining the chosen story with the client—took 20%. The final 10% covered production costs. (Media was not included in this project's scope.)

• Late 2024: In a very similar project we used about 20% of the budget on data analysis. The tools available through AI models made the process faster and—though not quantifiable—more thorough. The creative process expanded to 50% of the total work. Production costs remained stable at about 15%. The ideas were much stronger and likely contributed to the overall proportional success of the campaign (hard to prove but studies suggest that between 45 and 80% (quite a range, I know) of a marketing campaign's effectiveness is driven by the quality of the creative idea.

If you have experience of working in a professional creative process, it seems strange that more time to develop ideas might result in a better idea. The assumption often is that an idea pops into existence and is good or bad. The reality is that the best creative ideas come from having lots of ideas to choose from, connecting them to a strategy as part of the design and development process makes them better.

Three options

The temptation to simply reduce costs and increase profits is strong. But our experience, and Goldman's hints about their approach, suggest three more strategic options:

1. Make it better: Raise quality, delight the customer with better service.

2. Remix: Simulating different audiences or customers can show how you can produce different versions to choose from, or tailored versions for people with different needs.

3. Split the difference: Reduce the cost of your service to stand out from competitors who are less advanced with their AI-acceleration programmes, perhaps while increasing your businesses margin for this work.

The reality check: implementation challenges

Goldman Sachs is in corporate finance, we are in transformation consultancy and marketing services. The path forward isn't without obstacles:

1. System resistance: Professional services by and for large organisations are governed by procurement systems that will be based on time and materials. They will always want more for less and will need to be convinced that AI-supported services deliver more.

2. Half-way houses: Productising elements of professional services will be easier. We have done it with our smallest product – AI-B-C™ Power Hours – and want to extend the learnings from that into the rest of the things we sell.

3. Radically disruptive new business models: The few we have seen already are early, exploratory plays. Brandtech's Pencil approach seems the most potential in the marketing creative services niche. But business innovation in other sectors is still at a very early stage.

The bottom line

The real opportunity isn't in cost-cutting - it's in reimagining what's possible when routine work takes a fraction of the time. The winners won't be those who save the most money, but those who invest those savings most wisely in quality, creativity, and customer value.

Agents not ready for prime time

What: AI agents and agentic AI are apps or custom bots that can go and complete a task for you making decisions and using other tools as they go.

So what?: Three AI companies launched agent tools this week. ChatGPT’s Operator, Google Gemini and Anthropic’s Claude.

What next?: This is a “Next” thing for a lot of non-tech companies. Advanced AI users and teams should be experimenting with this but it is not something ready for use by most people. The frustrations described by early testers of the OpenAI Operator agent report a lot of difficulties getting it work. This could even put people off the tech if they are introduced to it when they and the tech are not ready for each other.

Interesting AI reads

Google Gemini has introduced Gemini 2.0, featuring multimodal capabilities and chained actions. These updates enhance the versatility of AI assistants, making them more useful for both consumers and businesses. Expect further integration into Google's ecosystem and new applications in the near future. Read more.

Anthropic CEO Dario Amodei Believes A.I. Could Double Human Lifespans in 5 Years: Anthropic's CEO has made a bold claim that AI could double human lifespans within five years. This highlights the potential for AI to revolutionize healthcare and longevity research, attracting significant investment and partnerships. Continued advancements in AI-driven healthcare solutions are anticipated. Read more.

DeepSeek's new open-source AI model can outperform o1 for a fraction of the cost: DeepSeek has released the R1 reasoning model as an open-source alternative to OpenAI's o1. This model provides a cost-effective, high-performance AI solution, challenging established players and democratizing access to advanced AI. Watch for increased adoption and further developments in open-source AI models. Read more.

Copilot gate-crashes users’ work: Writing for Unherd, Sam Leith asks “How do I kill my Microsoft Copilot?”. Microsoft has made Copilot a “standard” part of every Office365 subscription, although it is being described by many as an example of “dark pattern design”, as it isn’t obvious to users that they don’t have to pay extra.

If the only way you can sell something is by hoping people won’t notice or won’t bother to object or won’t take the trouble to find out how to get rid of it, what you have isn’t a product: it’s a parasite. “How to kill my Microsoft Copilot,” by Sam Leith

That’s all this week…

See you next week. If you enjoyed it, please click the 🤍 to make it ❤️. Thank you!

Antony