Most journey maps carry a row for how the customer felt, and it is usually the row nobody can source. The actions came from process documentation, the touchpoints came from a systems list, and the feelings came from whoever was holding the marker. Emotional journey mapping is the work of building that row properly, so its values still hold up when someone questions them six months later.
What emotional journey mapping actually is
Emotional journey mapping is the practice of recording how a customer felt at each step of a journey, based on evidence, alongside what they did and what they were thinking at that step. It is a layer added to an existing journey map rather than a separate artifact. Same persona, same scenario, same steps, with one more row underneath.
The difference between that row and the ones above it is worth being precise about, because everything difficult here follows from it. Actions can be observed: pull the timestamps out of a system, and two people looking at the same evidence write down the same step. A feeling is an internal state that has to be reported by the person who had it, or inferred from something they did, and both routes introduce a gap the action row does not have.
The layer is worth the trouble because it changes which steps get attention. Without it, the map ranks steps by how much work they involve or how often they break, and a step that is fast, functional and entirely on-time can still be the one that ends the relationship. It is also the part of creating a customer journey map that most often gets filled in last, in the final twenty minutes of a workshop, which is why it deserves a method of its own. In practice it takes one of two forms, a line across the steps or a value attached to each step, and that choice matters far less than what sits behind the values.
Where the emotion values come from
Take a broadband provider mapping its move-home journey, the sequence a customer goes through when they move house and want their connection to follow. Eleven steps, from the first search for the cancellation policy to the first bill at the new address, and nine interviews with customers who moved in the previous three months.
The working rule for the whole layer is that every plotted point should be traceable to something specific: a line in a transcript, a survey item, a support ticket, an observed behavior, or a written note saying this one is assumed. A point that traces to none of those is not a finding, and presenting it as one is how the emotion row loses credibility with the people who have to act on it.
What counts as evidence for a feeling
Four sources do most of the work, and each one is good at something different. None of them is sufficient alone, and the choice of which to lean on shapes what the finished curve can be used for.
| Source | What it gives you | Where it skews |
|---|---|---|
| Interviews and diary entries | Intensity and cause, in the customer's own words | Toward the vivid moments people can still narrate |
| Survey items placed at steps | Comparable values across a larger group | Toward the steps you thought to ask about |
| Support tickets and contact reasons | Unprompted evidence that something went wrong | Toward steps that generate contact, away from silent abandonment |
| Behavioral signals | Indirect evidence that a step is costing people something, from drop-off, repeat contacts and time on step | Tells you where, never what the person felt |
These sources disagree in fairly predictable directions. Treat the disagreement as usable rather than as a problem to average away. On the broadband map, the step where customers choose an installation date scored well in interviews and generated more repeat contacts than any other step. Both readings were accurate: people remembered choosing a date as easy, then called back twice to check it had registered. That points at a specific failure of confirmation, which neither source would have surfaced alone.
Volume matters more here than it does for the action row. With fewer than roughly six to eight accounts of the same journey, treat the layer as a hypothesis you intend to test rather than a finding you present, and say so on the map. Six people can tell you a step is a problem. They cannot tell you how it compares to the step before it.
The gap between what people felt and what they remember
Almost all journey research is retrospective, which means the accounts you are working from are reconstructions rather than recordings. People compress long uneventful stretches into a sentence, sharpen the worst moment, and let the way the whole thing finished color everything before it. That does not make the research unusable. It does mean the emotion row records remembered experience, and the map should say so.
Where the journey allows it, capture closer to the event: a short prompt after a specific interaction, or a diary study running while the journey is still in progress. Where you cannot, note on the map that the values are retrospective.
The distortion to watch for is a suspiciously smooth middle. Long journeys get flattened in recall between the opening and the ending, so an interview-built curve often shows detail at both ends and a featureless stretch between them. That stretch is usually under-researched rather than uneventful.
Whose feeling is on the map
A single curve implies one emotional state per step, and nine interviews produce a spread. Averaging participants who felt opposite things at the same step produces a number that describes none of them, and it removes the most interesting thing in the data. When the values at a step disagree, work out what kind of disagreement it is.
The spread tracks who the people are
That is a persona split, and the honest response is to draw the curve twice. Business and residential customers moving the same broadband connection are not one population with high variance.
The spread tracks what happened to them
The map is hiding a branch. On the move-home journey, the installation-appointment step split cleanly by whether the engineer arrived inside the booked window, which is two different paths through one column rather than variation in feeling. Fix the structure before you touch the emotion row.
There is no pattern in the spread at all
Record the range instead of the mean and let its width be visible. A step where the values run from mildly positive to strongly negative is telling you something real about consistency that a midpoint would hide.
Marking what you guessed
Mark every point as evidenced or assumed, visibly, on the map rather than in a separate document nobody opens alongside it. The effect in a review is immediate: the conversation stops being an argument about whether the map is right and becomes a list of which points need evidence, which is also your research plan for the next round.
Steps where you have nothing at all can stay blank, and leaving the gap visible costs you nothing. Filling it in from team consensus converts an evidence gap into something that looks like a finding, and six months later nobody can tell the two apart.
Smaply links each card to the interview, quote or metric it came from, so the emotion row stays defensible.
Building the layer
Anchor to one persona and one scenario first
The emotion row only means anything if it belongs to a specified person doing a specified thing. The same broadband journey plotted for someone moving into a first flat and someone moving a home office produces two curves, and merging them produces a third that belongs to nobody. If the map already has a persona and scenario, inherit them rather than inventing new ones for this row.
Choosing a scale you can apply the same way twice
Scale choice is not a presentation decision. It determines whether two researchers reading the same transcript land on the same value, and whether the map still means anything after the team has changed. A wide numeric range looks more precise and behaves worse, because nobody can consistently distinguish a 7 from an 8 across a hundred transcript lines.
A small number of levels with written anchors works better, and the anchors should be defined by what the customer did or said rather than by an adjective. "Frustrated" means different things to different coders. "Contacted support about this step" does not.
Five levels, each one tied to something the customer actually did, covers most journeys:
Whatever set you write, the anchors have to travel with the map. An undefined scale drifts between sessions and between people, the values stop being comparable across steps, and the curve becomes uninterpretable while still looking finished. Emoji scales read well in a workshop and shed their definitions faster than numbers do, so keep the written anchors attached.
Placing the value on the step
One value per step per persona, attached to the step rather than floating between steps. Feelings that seem to happen in the gaps are usually evidence of a step the map is missing, and waiting is the most common case. An eleven-day gap between ordering a transfer and hearing anything back is a step the customer experienced, whatever the operational view says about nothing happening during it.
Record three things alongside the value so the point stays readable later: where it came from, when it was captured, and one verbatim line where you have one. A number with a quote underneath it survives a challenge. A number on its own gets discounted the first time somebody senior disagrees with it.
Drawing it so it can be read
A connected line reads as a trajectory and invites comparison across steps. Discrete markers read as independent readings and resist the impression of smooth movement. Pick based on coverage: a line if you have a value at every step, markers if you have gaps. Smoothing across missing points is the visual equivalent of inventing data.
The layer gains most of its meaning from being read vertically, against the action, the thinking and the channel in the same column, which is the argument for keeping it inside the journey map rather than maintaining it as a chart of its own.
Reading the curve and deciding what to act on
Experiences are not remembered as the average of their moments. Experimental work by Daniel Kahneman and Barbara Fredrickson established that people judge an experience largely by its most intense point and by how it ended, a pattern known as the peak-end rule. The consequence for the row you have just built is direct: the mean of the curve is not what the customer carries away, or what they tell a colleague who asks whether the provider was any good.
Two eleven-step curves can average out identically and call for completely different responses:
- Neutral across most steps, then a steep fall at step ten
- The whole journey gets described by that drop afterwards
- Response: one step, one named owner, one change
- No distinct high or low anywhere across the eleven steps
- Reads as mild dissatisfaction with everything
- Response: check where the values came from before designing anything
Not every low is worth acting on, and the ones worth acting on are not always the deepest. Three questions separate them:
- How many people reach this step at all? A severe low on a path taken by four percent of customers is a different investment case from a mild low everybody passes through.
- How far does the value drop, relative to the steps either side? The drop isolates what this step did, rather than what the customer brought into it.
- Does the journey survive it? A step people dislike and continue through, a step that produces a support call, and a step people abandon are three different problems.
The flat curve deserves its own diagnosis, because it is the most common disappointing result. A line sitting at mild dissatisfaction across every step, with no distinct high or low anywhere, almost never means the experience is uniformly mediocre. It usually means the values came from team consensus rather than from customers, or the scale had no written anchors so everybody defaulted to the middle, or the steps are too coarse and each column averages several moments together.
On the broadband map, the deepest point was not the step the team expected. The installation appointment, which had absorbed most of the operational attention, scored badly but recovered. What determined how people described the whole journey afterwards was the first bill at the new address, where charges from both properties appeared without explanation. Last step in the sequence, cheap to fix, never on anyone's list.
Acting on a low means what it means anywhere else on the map. Somebody's name goes against it, a specific change gets made, and the same step gets re-read with the next cohort's values to see whether anything moved.
Keeping the layer current
The functional rows get updated when a process changes, because somebody notices and says so. The emotion row does not, since refreshing it takes research rather than a conversation. Two habits keep it honest: date every value so its age is visible, and re-check the affected stretch after a change ships instead of waiting to redo the whole study. Points old enough that nobody would defend them get marked unverified rather than left standing as findings.
On the broadband map, three of the eleven steps still carry nothing on the emotion row. Two are steps almost nobody described without prompting, and one is where the nine accounts contradicted each other so completely that any single value would have been fiction. They are blank on purpose, and the blanks are the shortlist for the next round of interviews.
Emotion lanes, personas and research live in one place, so the curve and its evidence stay together.
Frequently asked questions
What scale should I use for emotions on a journey map?
Use a small number of levels, around five, with written anchor definitions attached to the map. The count matters far less than whether two people applying them to the same evidence land in the same place, and anchors defined by observable behavior travel better than anchors defined by adjectives.
Where do the emotion values come from if I only have a handful of interviews?
From the same interviews, treated as a hypothesis rather than a finding. Below roughly six to eight accounts of one journey, you can identify which steps are worth investigating but not how the values compare. Mark the layer provisional and use the gaps to scope the next round of research.
What do I do when research participants felt opposite things at the same step?
Work out what the disagreement is made of. If it tracks who the participants are, split the persona and draw two curves. If it tracks what happened to them, the map is hiding a branch and the structure needs fixing first. If there is no pattern, record the range rather than the average.
Should the emotion row be a curve, a score, or quotes?
Both a value and a quote, wherever you can manage it. The value gives you comparability across steps and the ability to see movement after a change, and the verbatim gives you the cause. A layer with values and no quotes gets argued with, and one with quotes and no values cannot show whether anything improved.
How is emotional journey mapping different from an empathy map?
An empathy map captures what one person thinks, feels, says and does around a situation, without a time axis. Emotional journey mapping records a value per step across a sequence, so it shows movement and lets you compare one moment against the one before it.
How often should the emotion layer be refreshed?
Trigger it by events rather than by calendar: after a change ships to the steps in a stretch, after a channel is added or removed, and when support contact at a step shifts noticeably.
.png)
