Your AI Strategy Have a 1/9/90 Problem.

May 1, 2025 · no comments

One individual stands apart from a large crowd, illustrating the 1/9/90 participation divide.

Why most companies report they don’t feel they get as much value from AI as they would like and what to do about it.

There are few discussions as consequential for companies these days as the those about AI.

Every organization is trying to understand how to utilize AI in their organizations, almost everyone ends up with mixed to bad results. A recent survey by Boston Consulting Group reveals that only 26% have developed the necessary capabilities to move beyond proofs of concept and generate tangible value, with 74% struggling to show tangible value from their AI use.

Why? Because deploying AI is not the same as productizing AI. The gulf between a clever prototype and a system that thousands of employees can trust every day is wide—and cultural as much as technical.

While the potential of AI in any organization is obvious, what is less obvious is how you actually implement it in a way that make sense. Very often it’s decided from the top that a companys employees will have to use AI in the form of co-pilots or other chat based agents, but while these can be helpful for the some of the more technically savy employees, most don’t really get get to see them as unnecessary convolutions of business processes, i.e. yet another thing to learn.

Having spent more than 3 years actively building AI agents for various businesses at Faktory, here are some realistic pieces of advice for how most organizations can avoid the same fate as those from the BCG report and actually benefit from using AI in their organization regardless of industry, size or employees technical proficiency.

The 1/9/90 Problem

There is an unspoken rule stemming from online communities which suggests that:

1% of users create.
9% interact with it (comment, share, etc.).
90% passively consume it.

This rule can also be applied to most other scenarios where creativity is part of the value creation, and that’s a problem, let me explain why.

Most people are not creative, neither do they strive to be. To them AI isn’t something exciting, it’s just yet another technology they are forced to learn how to deal with in their work. They might be impressed by it or they might even fear it but they don’t see AI and think about all the great things they can do with it.

Yet many AI roll‑outs hinge on that very assumption. Organizations deploy generic co‑pilots or chatbots and wait for a productivity boom that never materializes. The result is a familiar paradox: the people who could benefit the most (the 90 %) abandon the tool, while the power users become even more productive further widening the gap.

While there is absolutely nothing wrong with co-pilots, they just quickly run into a series of challenges that AI itself can’t solve easily. This might be the need for access of data outside the perimeters of the platform or the need to deliver consistent outputs for batch processing or to automate more complicated workflows. While a technical person could overcome those issues, most people can’t and thus you end up with AI that can only help you with things you can already do yourself, while unhelpful for things that would actually save you time.

In other words. Most AI strategies fail to add value not because AI is useless in the organization but because it’s assumed that everyone needs the same type of AI to be productive and know how to get the most out of it.

So how should organizations apply AI?

I want to be as concrete as possible so let me list up my 5 rules for implementing AI in your organization.

Build Task‑Focused Agents

Start with a single, clearly‑defined job that moves the needle—e.g., drafting audit‑ready SOC‑2 evidence, triaging support tickets, or compiling personalized sales briefs. Give the agent:

The minimum data it needs.
A deterministic policy for when to defer to a human.
A tight integration with the system of record where the work belongs.

Focus on getting one agent right rather than just letting it loose in the entire organization. I.e. integrate AI with conviction instead of spray and pray. The more concrete the implementation is the easier it is for the 90% to see how it can make their jobs easier.

Prioritize Reliability Over Awe

One of the things I have learned working intensively with AI and AI agent’s for more than 3 years now is that it’s much harder to get AI to do something useful consistently than it is to get it to do something impressive sporadically.

In most businesses, consistency is key and so most companies would benefit from prioritizing using AI as an optimizer rather than a foundation.

Instead of trying to get an AI to behave like an application use it as an optimizer of workflows. As an example. While an LLM can do search and replace in a text it can’t do it consistently and not if we are talking about very large amounts of texts. Instead develop classic batch solutions like search and replace that the agent can the utilize. That way you get the creativity of AI with the reliability of classic deterministic programming environments. Believe me it will save you a lot of frustration once you understand this.

Design for the 90 %

Assume zero prompt‑engineering skill. Surface AI through the interfaces employees already inhabit: a button in the ERP, an auto‑suggestion in the CRM, their email, slack, SMS or a background cron job that simply “does the thing” and notifies stakeholders. Even better treat AI as part of the backend and just do the things that the AI can do before it even get to the user.

This will ensure that more people use it more frequently and that it can actually add value.

Build Across Silos.

Besides the obvious uses like copywriting, research and ideation, AI Agents provide most value when they allow you to work across silos that were previously impossible to connect. That could for instance be getting som specific news from Google News, then create a news segment utilizing a language model and then narrate that segment using ex. Elevenlabs and viola you have a pretty cool workflow for creating audio podcasts.

If you are more serious you can use an agent to monitor your inventory for various events (low inventory, stale inventory etc) and then either send an email to you with a prefilled purchase order or just have it added directly to your vendors order mangement system.

Because of the power of language models to just use plain english, they make working between several unrelated domains very doable allowing agents to communicate via language and translate where needed.

Make your agents biased.

While most LLMs strive to be as neutral as possible, it’s actually when you give them bias they become useful. Not bias in the way we normally thing about it but rather with regards to how it goes about its task. A brand manager agent should be biased towards how copy is written or what images are “on brand”, an investment agent should be biased towards the trading strategies of the specific company, a research agent should follow the principles and unique methods of it’s user. You can’t run a business being neutral. You have to be biased towards your own success and your own way of doing things.

Building bias and even personality into your agents is not just important it can be used to make the blend more naturally into a company culture. Whether in slack or email, as a copywriter or assistant. Bias is key making agents be able to make decisions inline with your organizations goals and methods.

The fastest way is slowly but with conviction.

There is an almost infinite amount of ways AI can be used to improve your business and and even more ways it can be misused and wasted on absolutely nothing. Business will waste a lot of money and times committing to platforms, hiring expensive consultants and forcing everyone to become AI natives. The best way to to do what most businesses always have done, identify problems and opportunities and use AI to help resolve them.