A large Nordic telecom operator recently reduced its inbound customer support volume by 30%. The result came from an AI knowledge agent we built for them, drawing on their own product documentation as its knowledge base. Their support team stopped fielding the same questions repeatedly and started focusing on the genuinely complex cases, the interactions that actually require a human. The agent did not replace anyone. It gave the team back their time and their capacity to do the work that matters.

By William Hermansson, Sr. Digital Strategist & Project Manager

In conversations with marketing and commercial leaders across industries, we see a consistent pattern. The intent is there, the budgets are being allocated, and the announcements are being made. What tends to be missing is a clear-eyed understanding of what it actually takes to get from an AI pilot to concrete business results.

Why projects fail

Our own experience reflects a consistent pattern. Organisations start building before they understand where they are. They deploy tools into teams that don't have the competence to use them properly, so the tools never deliver what they are actually capable of. Processes rarely get the same attention. Either the new AI application is built without real regard for how the team actually works, or the team's ways of working are never updated to make room for it, and the two end up pulling in different directions. They try to solve five problems simultaneously, which means none of them gets solved well. And they underestimate, almost universally, how much the quality of the underlying data determines the quality of everything that follows. An AI solution, whether that is an agent, an internal tool, or something else entirely, can only perform as well as the data and thinking behind it. When that foundation is weak, the result is something that produces confident-sounding outputs that are vague, inconsistent, or wrong.

The numbers back this up. MIT found that 95% of generative AI pilots failed to deliver any measurable impact on the business. No discernible financial savings. No uplift in profits. BCG found that only 26% of companies had developed the capabilities to move a project beyond proof of concept, and only four percent of those who made it past proof of concept consistently generated significant business value from AI.

The picture these studies paint is of an enormous amount of effort and investment producing very little. And the reasons, when you look at them carefully, have little to do with the technology itself.

The MIT research found that executives tended to attribute failure to the AI models themselves, assuming the technology was not capable enough. The actual findings pointed elsewhere. The problems were organisational. Poor data foundations, teams that were not equipped to work with AI, use cases that were too broad to solve properly, and a fundamental mismatch between what got built and how the business actually operates.

"The data question is where most projects quietly fall apart. The organisations that succeed spend serious time on their data before they touch the build. The ones that skip that step end up rebuilding, or abandoning the project entirely."

- William Hermansson, Senior Digital Strategist at Pointseven

How to succeed

The organisations that make it into the small percentage that generate real business value from AI tend to do three things, and they tend to do them in order.

They start by understanding their actual position, not their aspirational one. They make sure they have the human capability and processes in place. And they build something specific before they build something ambitious.

This is the three phase process that we advise our clients to follow:

Phase 1: Assessment and action plan

Before any investment in building, it is crucial to understand the starting point to be able to direct the efforts efficiently. Through structured stakeholder interviews across leadership and commercial teams, and a detailed audit of existing data, tools, and ways of working, we build a clear picture of where an organisation stands on AI readiness.

The output is a concrete action plan. Not a strategic document with observations and recommendations. A prioritised set of initiatives, with clear owners and a realistic timeline, designed to go into a leadership meeting and drive decisions.

Phase 2: Academy

Building an AI capability into an organisation where the people are not ready for it is one of the most reliable ways to waste a build budget. Tools that land without context tend to be used inconsistently, then distrusted, then quietly dropped.

Academy is the phase where teams are trained to use the tools properly, and where business processes are put in place to make the new way of working actually stick. We typically advise a six-week programme for commercial and marketing teams, built around the organisation's own tools, workflows, and context. Sessions are weekly, hands-on, and focused on specific areas of AI in practice. Between sessions, participants work through individual assignments that put the learning into use immediately rather than in theory.

At the end, AI readiness is re-measured and compared against the baseline established in the Assessment phase, giving leadership a concrete picture of where the organisation has moved and what is now possible.

Phase 3: Create

The build starts narrow, by design. Scoping defines the use case, the target audience, and the starting point where the best data already exists. That last part matters more than most people expect. The single most important decision in any AI build, whether that results in an agent, an automated workflow, or another type of application, is where you start.

Data preparation follows, and its duration is genuinely variable. If the documentation is already structured and comprehensive, the process is straightforward. If content needs to be created, verified, or reorganised from scratch, it takes significantly longer, and requires meaningful involvement from subject matter experts on the client side. Any honest proposal in this space names that dependency clearly.

Build and test runs in parallel with the later stages of data preparation. Testing is where most of the real work happens. Outputs get evaluated against real use cases, gaps are identified and filled, and the solution improves iteratively. The subject matter expert involvement that shaped the data is just as important during testing.

After launch, the work isn't finished, but the fine tuning starts and the real value accrues.

Real usage and user feedback are monitored closely. Updates to the underlying data and technology improve the quality of the solution over time, and along the way, the organisation can gain new insights about its customers. As confidence in the foundation grows, the scope can expand to additional audiences and use cases without rebuilding from scratch. That way the value of the AI investment can be multiplied.

Getting started

Understanding where you stand, building the right competence and processes in your team, and then building the right thing, in that order, is what makes AI initiatives actually work. Get in touch, we'd love to help you figure out what that looks like for your organisation.