Last updated: 30 October 2026
Yes, for a first draft. Claude can produce a usable customer journey map in about half a minute. How useful it is depends on what you give it: a one-line request returns a generic map, while real facts about your leads let it name a likely drop-off. Treat that as a hypothesis to check against your own data.
We tested this on 30 October 2026. We ran three prompts in fresh Claude sessions, each with no earlier conversation and no access to the web or to our files, and then compared the answers with what we know about our own lead process at QD Academy.
What did we test and how?
| Level | What we gave Claude | What we looked for |
|---|---|---|
| 1. Generic | One line: build a customer journey map for a Singapore training provider | Specificity, and any invented detail |
| 2. Structured | A persona, a goal, five stages and the five table columns we wanted | Fit to the template and realism of the pain points |
| 3. With facts | Plain facts about our lead sources and follow-up, without saying what the problem was | Whether it found the real drop-off and said what it could not know |
For level 3 we described the process: paid ads and event forms, follow-up by hand when staff are free, and low conversion from ads. We did not say that slow replies were the cause. We already knew the answer from our own results, so we could judge whether Claude would reach it. Our phrase “follows up by hand” is a strong hint, so read the result as encouraging. It does not prove that Claude will find your drop-off.
What did Claude get right with a one-line request?
Quite a lot. From one line, Claude built two personas (an individual learner and a corporate buyer), seven stages and a table covering goals, touchpoints, thoughts, pain points and opportunities. It added a list of gaps to investigate and a set of quick wins. It picked up Singapore details without being asked, such as SkillsFuture Credit, confusion over subsidy eligibility and the manager approval that sponsored courses need. It also ended with a reminder to recheck funding rules, because they change.
Where did the one-line map fall short?
Every detail was an assumption. The personas and their pain points were invented to fit the type of business, and none of it came from our customers. Suggestions such as a two-hour WhatsApp reply standard were reasonable, but nothing in the answer showed they would work for us.
It was also long. Two personas and seven stages are more than most small teams can act on, so you would need to cut it down before using it.
What happened when we gave Claude a persona and stages?
The answer came back in the exact table we asked for. The pain points were more specific to the persona, a marketing manager enrolling two colleagues: separate forms for each person, an unclear net fee per participant, worry that an admin slip could cost her the funding, and concern about putting company information into AI tools. Claude raised that last point on its own and suggested a short guide to safe AI use under the PDPA.
It also proposed a group booking flow and a 48-hour seat hold while approval is pending. These are ideas worth testing. They are not findings.
Did Claude find the real drop-off when we gave it facts?
Mostly yes. Claude placed the biggest drop-off in the wait between a lead submitting a form and understanding the fee, which matches what we found. Its reasoning was that every lead passes through that point and depends on a person being free to reply, and that ad leads are the coldest because they have not met anyone from the provider. It named the wait for manager approval as the second most likely drop-off. We have not measured that one.
Claude also said plainly that this was a hypothesis, because our facts said conversion was low but not where leads left. It then listed the data to check first: time from form submission to first contact, conversion by response-time band, the share of leads never contacted, and duplicate or missing records. It advised measuring each month’s leads over a fixed window, such as 60 days, so that a fresh month is not compared with an old one.
Here is how it played out for us. Leads from our ads were converting poorly, and the cause was manual follow-up: leads waited a long time for a reply and dropped off. After we set up an automated answer workflow in GoHighLevel, the response rate went up by 50% and the enrolment rate rose as well. You can see how we set it up in our guide to automating the customer journey by stage.
What did Claude not know or get wrong?
It cannot know your numbers. Without your data, any drop-off it names is a guess, and it said so itself. It also assumed that conference leads convert better than ad leads because they have met the team. That sounds plausible, but we gave it no data and we did not check it.
Funding rules were the other weak spot. Claude hedged correctly and told us to confirm current details with SkillsFuture Singapore, which is the right habit for anyone using an AI-made map in Singapore.
How do you get a better map from Claude?
- Say who the customer is and what they are trying to do.
- Give your real stages and the table columns you want.
- Paste anonymised facts about how your leads behave. Remove names, phone numbers and email addresses first.
- Ask for the biggest drop-off and for the data needed to confirm it.
- Ask what Claude is assuming.
A prompt you can copy: “Here is how our leads move through our business: [paste anonymised facts]. Build a customer journey map with these stages: [your stages]. Name the stage where you think most people drop off and why, list what you are assuming, and tell me what data I would need to check before changing anything.”
If you need a layout to start from, our journey map examples and templates show one you can copy.
How should you check an AI-made map before acting on it?
- Compare each pain point with real customer messages or call notes.
- Check every number against your own records, and delete any you cannot source.
- Ask a few customers whether the map matches their experience.
- Test the biggest fix on one stage before changing the rest.
- Review the map every quarter.
When is Claude not enough?
A first draft is a starting point. The harder work is gathering the evidence, deciding which fix comes first and turning it into a process your team follows. Our step-by-step guide to building an AI-powered customer journey map covers the full process.
Frequently asked questions
Can Claude build a customer journey map?
Yes, for a first draft. In our test each answer came back in about half a minute. The result is only as reliable as the facts you give it, so treat it as a hypothesis to check.
Is it safe to paste customer data into Claude?
Remove names, phone numbers, email addresses and other identifiers first. Check your obligations under the PDPA and the terms of the plan you use before you paste anything that could identify a person.
Is Claude better than ChatGPT for journey mapping?
We only tested Claude, so we cannot compare. You can run the same three-level test on any AI tool and judge the results against what you know about your own customers.
Can I use the free version?
We did not test a specific plan. Usage limits and available models change over time, so check the current plan details before you rely on one.
Do I need to know prompt engineering?
No. Plain sentences that state your customer, your stages and your facts worked well in our test. Asking Claude to list its assumptions is the most useful habit to add.
Learn to map and automate customer journeys with AI
QD Academy’s 3-day course, Developing an Effective Customer Journey Map using AI and Automation, teaches marketers, sales and service teams to build persona-based journey maps from their own data and to set up AI follow-up workflows. Eligible Singapore learners can use SkillsFuture funding, and our team handles the grant paperwork. See grants and subsidies for eligibility, and check the course page for dates and fees.