The tools are not the bottleneck
Access to AI is no longer the hard part. Capable models are available to every organization, built into the software you already pay for, and priced within reach of a small team. Yet many businesses that have rolled out AI tools are struggling to point to a result that shows up in the numbers.
The reason is almost never the technology. AI creates value when it changes how work gets done, and most organizations have added AI to their work without changing the work itself. A faster way to draft an email that nobody needed to send is not a transformation.
Unlocking AI is a business design problem. Here is how we approach it.
Start from decisions and workflows, not models
The most common starting question is "What can we do with AI?" It produces long lists of ideas and very few results. A better question is "Where does our business lose time, money or quality, and why?"
Look for work with three characteristics:
- It happens often. Hundreds or thousands of times a month, so small improvements compound.
- It involves reading, writing or judging information. This is where modern AI is strongest.
- The cost of delay or error is visible. Slow quotes, missed renewals, inconsistent answers, backlogs.
Map that work as it actually happens: who does what, with which information, in which systems, and where it waits. The opportunities usually reveal themselves.
Four places AI pays off reliably
Across sectors, the use cases that deliver tend to fall into four families.
1. Reading and routing
Emails, forms, documents, tickets and claims arrive in volume and need to be understood, classified and sent to the right place with the right details extracted. AI does this quickly and consistently, and people handle the exceptions.
2. Drafting
First drafts of proposals, reports, product descriptions, replies and summaries, written in your style and grounded in your own information, for a person to check and finish. The gain is not that people stop writing. It is that nobody starts from a blank page.
3. Finding
Most organizations know far more than any one employee can find. AI search over your own documents, policies and past work turns that knowledge into answers, with links back to the source so people can verify them.
4. Spotting and forecasting
Patterns in operational data, such as unusual transactions, early signs of customer churn, demand shifts or equipment wear, surfaced in time to act on them.
None of these are glamorous. All of them are measurable, and together they cover a large share of the work in most organizations.
Choose with a simple scorecard
When you have a list of candidates, score each one on three questions:
| Question | What good looks like |
|---|---|
| How much value if it works? | Clear impact on time, cost, revenue or quality, at real volume |
| How feasible is it now? | The data exists, is accessible and is good enough |
| How much risk if it goes wrong? | Errors are easy to catch and cheap to correct |
Start with the opportunities that score well on all three. Leave the high-risk, high-ambition ideas until you have built the skills and the safeguards on simpler ones.
Prove it small, then build it in
Do not commit a large budget to an untested idea. Build a working prototype on your own data in a few weeks, put it in front of the people who do the work, and measure the difference. Our discovery process, The Idea Forge, exists precisely to separate the ideas worth building from the ones that only sound good.
When the prototype proves itself, the real work begins: building it into the workflow. That means deciding where the AI's output appears, who reviews it, what happens when it is wrong, and how it connects to the systems of record. A brilliant model that lives in a separate tab will be ignored within a month. A modest one built into the tool people already use every day will change how they work.
Measure outcomes, not usage
"Number of employees using AI" is an easy metric and a misleading one. Measure what the use case was meant to change:
- Turnaround time from request to result.
- Cost per case, per order or per customer served.
- Error and rework rates.
- Revenue or retention where the use case touches customers.
Measure a baseline before you start, so the improvement is a number rather than an impression.
Bring your people with you
AI changes jobs, and people know it. The organizations that succeed are open about what is changing and why, involve the people who do the work in designing the new way of working, and invest in training that is specific to their roles rather than generic. They also make it clear who is accountable for AI-assisted decisions: always a person, never the tool.
Trust is built the same way it is with a new colleague: start with low-stakes work, show the reasoning, and expand responsibility as the track record grows.
Get the foundations right
Three foundations determine how far you can go:
- Data. AI can only be as good as the information it can reach. Clean, accessible, well-governed data is the single biggest accelerator.
- Governance. A clear policy on which tools may be used, with which data, and with what human oversight. In Europe, the AI Act makes this a legal question as well as a sensible one.
- Privacy and security. Know where your data goes when an AI system processes it. For sensitive work, models that run inside your own infrastructure are increasingly a practical option rather than a compromise.
AI is a capability, not a project
The organizations getting the most from AI are not the ones with the biggest budgets or the most pilots. They are the ones that treat AI as an ongoing capability: a steady pipeline of well-chosen use cases, each one proven small, built into the work and measured honestly.
That is how our AI Transformation team works with clients in every sector. Start with one workflow, prove the value in weeks, and build from there.
