There is a pattern to failed AI projects. Someone sees a demo, a team is told to “do something with AI”, and six months later there is a chatbot on the homepage that answers questions nobody asked. The technology was chosen before anyone looked at the service.
Service design offers a better starting point. A journey map shows what the user is trying to do and where it goes wrong. A service blueprint shows what staff and systems do behind each step. Put the two together and the places where AI could help become obvious, and so do the places where it would do harm.
Start with the as-is
Map the journey as it really happens, with evidence: interviews, observation, analytics and the call logs that show failure demand. For each step note three things. What is the user trying to do? What does the organisation do in response, front stage and back stage? Where does it break?
Breaks tend to fall into a small number of types. Waiting for someone to look at something. Repeating information already given. Not knowing what to do next. Being asked a question in language that does not match the user’s. Each type points to a different kind of help.
Sort judgement from process
Now go through the back stage. Some steps are process: classify this, route that, extract these fields, draft that standard letter. Others are judgement: is this person eligible, is this evidence credible, what should we do about an unusual case?
Process steps are candidates for automation with human oversight. Judgement steps are candidates for decision support, where AI assembles and summarises but a person decides. Treating a judgement step as process is the most common and most damaging mistake we see.
Design the touchpoint, not the feature
For each candidate, design the touchpoint as part of the journey. What does the user see? Do they know AI is involved? What does the AI need to know and where does it get it? What happens when it is unsure, and how does a person take over? How will you know, a month after launch, whether it is working?
Written down, a touchpoint looks like a row on the blueprint: user action, AI action, human action, data, failure path, measure. If you cannot fill in the failure path, it is not ready.
Score and sequence
You will end up with more opportunities than you can test. Score them on user value, feasibility given the data you actually have, and risk. Pick the strongest one and build a proof of concept in front of real users. The blueprint tells the rest of the organisation what you are doing and why.
A worked example
On a procurement service we designed for the health sector, the as-is journey showed clinicians spending hours finding and comparing evidence about devices. The blueprint showed that evidence lived in a dozen places and that the real judgement, which device for this patient group, had to stay with the clinician. The right touchpoint was retrieval and summarisation with sources shown, feeding a human decision, not a recommendation engine. That distinction came straight from the map.
If you would like help finding where AI belongs in your service, our AI opportunity discovery does exactly this.
