Data Storytelling
Data storytelling delivers an analytical finding as a narrative, to people who have to decide and who did not run the analysis. It brings together three things: the data that carries the finding, a visual that makes it readable at a glance and a narrative that supplies the business context, ties the dips and the peaks to the events that explain them and states the decision at stake. The IIBA guide places it at the end of the analytics cycle, in the interpretation and reporting of results.
Purpose
Data storytelling carries an analytical finding through to the decision it is meant to serve. A validated result changes something on the day when someone who neither wrote the query nor saw the data draws a conclusion from it and decides. The IIBA guide puts it plainly: figures, curves and charts mean little without their business context.
The deliverable is a narrative with its visuals, short enough to be read or presented in a few minutes, which states the main finding, explains it through events the audience recognises and ends on a costed option. That deliverable has a named recipient: the same finding taken to the executive committee and to the product team gives two narratives.
The word storytelling also names an elicitation technique, in which stakeholders are asked to recount their experience of the work so that the culture and the tacit knowledge of an organisation surface. The two techniques move information in opposite directions: elicitation storytelling gathers, data storytelling gives back.
Usage
When to use it
- A validated finding aimed at a decision: a committee has to decide and has not followed the analysis.
- A counter-intuitive figure: a fall or a spike that the dashboard displays without explaining.
- An audience far from the data: management, the board of directors, external partners, business functions that do not analyse.
- An unwelcome conclusion: laying the reasoning out step by step makes the finding arguable on its merits.
- A recurring output nobody reads any more: a monthly dashboard published out of habit, which answers the questions of whoever already knows what to ask it.
When not to use it
- An audience of analysts who followed the work: the analysis notebook and the dashboard tell them more.
- Regular operational monitoring: metrics and KPIs with alert thresholds hold up better over time.
Description
The three components
Brent Dykes, in Effective Data Storytelling, describes the technique as the overlap of three things: data, a narrative and visuals. He labels each two-way overlap by what the pair makes possible; each pair can just as well be named by the artefact obtained when the third component is missing. Data and visuals without a narrative give a dashboard, accurate and consultable, that nobody knows what to do with. Data and a narrative without visuals give a written report the audience skims. A narrative and visuals without data give an illustrated opinion, all the more convincing for being well made. Data storytelling occupies the area the three have in common.
The element list of the IIBA guide follows the same split and adds the audience, which governs the other three.
The audience decides the rest
The guide makes audience analysis the first question of the work: who receives the finding, which questions these people want answered, which questions they might ask and what they need to know in order to decide. The level of detail is set by that answer, from a single figure to a table of assumptions. The format fixes the degree of formality, the tone and how much technical vocabulary is acceptable. The relationship decides what the analyst is entitled to say: the guide observes that a recognised expert prescribes the next steps where a beginner goes no further than suggesting possible directions. Identifying and characterising the recipients belongs to Stakeholder List, Map, or Personas.
Cole Nussbaumer Knaflic, in Storytelling with Data, reduces those choices to two questions asked before anything is drawn: who is being addressed and what one wants those people to know or do. A vague answer to the second gives a presentation that shows everything and asks for nothing.
The narrative carries what the visual does not show
A concise dashboard leaves out most of the analysis. The narrative picks up part of that discarded material: the explanation of the dips and the peaks, obtained by tying them to dated incidents or events, and the business context that says why the movement matters. A curve shows that repeat purchase fell in March. The narrative says that the sign-up procedure changed in March, which no system data holds, because the information sits in the minutes of a meeting.
The guide describes its shape: an opening that sets the context and introduces what is at stake, a middle built around the successes and the difficulties, an ending that delivers the lessons. The guide ties its effectiveness to the emotional part of a decision, which validated results on their own leave aside, and credits it with activating more areas of the brain than a factual report. It backs that second claim with no study.
The visual has to deliver the main finding fast. The guide subordinates four decisions to that objective. The type of representation follows the data at hand; the colour scheme works through contrast, an accent colour on the single series that carries the message and grey for the others. The size of the text and of the elements is set by their importance, and position leads the reader's eye. The detail of these choices belongs to technical visualizations; what data storytelling adds is the rule of subordination, one visual for one message.
Running the work
Write the main finding in one sentence, with its number, before any formatting. A sentence that does not fit on one line signals a finding that is still vague or two findings mixed together.
Name the recipient and the decision expected. The recipient is a specific role, the executive committee or the head of the loyalty programme. The decision expected is written as a dated action.
Sort the findings, because the analysis produces far more of them than the narrative will carry. The guide separates primary findings, which directly answer the research questions asked at the outset, from secondary findings, which supply context or justification. The first go into the dashboard and into the narrative. The second may stay in the narrative; they do not go into the visual. The rest is dropped, applying the concision principle the guide states. This sorting comes before the visuals are built: the reverse order leads to keeping a chart because it cost two hours.
| Finding | Destination | Reason |
|---|---|---|
| 90-day repeat purchase of members signed up at the checkout: 46% before March, 31% since | Dashboard and narrative | Answers the research question that was asked |
| 62% of checkout sign-ups since March carry no e-mail address | Dashboard and narrative | Answers the second question asked, the one about the cause |
| The fall is the same across all fourteen stores | Narrative only | Rules out a local cause, adds no curve |
| The age structure of the membership is unchanged | Dropped | Checked, no effect on the decision |
Choose one visual per primary finding and title it with the message rather than with the names of the axes. "Repeat purchase among members signed up at the checkout has lost fifteen points since March" tells the reader more than "90-day repeat purchase by sign-up channel".
Write the business context that explains each movement kept. This step goes outside the data: procedure changes, campaigns, incidents and decisions come from asking the people who were there.
Order, then cut. The guide notes that a disjointed order loses the audience and that a sentence adding nothing is struck. The last step is a re-reading whose only purpose is to cut.
The traps
The selective narrative
The guide places it among its limitations: when a bias or a personal interest drives the narrative and only the data that supports it is shown, the audience is misled. The trap belongs to the technique: the devices that make a true finding memorable make a distorted finding just as memorable, and the intended audience is by construction the one with no access to the data. The distortion can also be unintentional, a narrative lending credibility to a finding that is still fragile and should have been firmed up before it was taken to the meeting. Two checks cost nothing. Does the narrative state what contradicts its conclusion, where such data exists? Has someone who took part in the analysis re-read the version being presented?
Correlation told as a cause
The narrative ties a dip to a dated event, and the audience hears a cause. The curve falls in March; the procedure changed in March; the conclusion forms by itself in the room with nobody having stated it. The remedy lies in the vocabulary and in what goes with it: say that the timing coincides, name the other explanations ruled out and the means by which they were, then announce an A/B test when the question deserves to be settled.
A narrative longer than the attention available
A rambling narrative buries the message, and the long version is the one that gets written first. The constraint is set in advance: a number of minutes, a number of screens, one main finding.
Quality that depends on the author
The guide counts among its limitations the fact that narrative writing is a craft and that the effect of the narrative depends on the abilities of whoever writes it. A narrative-plan template and a peer review by another analyst close most of the gap.
AI considerations
The first useful application is drafting a first version of the narrative from findings already sorted: the figures are supplied, the structure is imposed and the model produces a text to correct rather than a blank page. Next comes rewriting for a given audience: length, vocabulary and level of detail change together, mechanical work the model does fast. The model also proposes several wordings for the title of each visual, where one usually stops at the first that comes to mind.
The last application is a check rather than help with production. Submitting the finished narrative to a model and asking it what decision it would take, then what objections it would raise, tests whether the message reaches a reader who does not know the file. An answer wide of the mark signals a narrative to rework before the meeting.
Two limits remain. The business context is not in the data. Asked for the cause of a dip, a model supplies a plausible explanation where the technique asks for an established one. The second limit is that the selective narrative becomes cheaper: a model asked to defend a conclusion will defend it, with polished prose and the data selection that goes with it. The peer-review check applies first of all to texts produced that way.
Examples
A retail chain in French-speaking Switzerland with fourteen stores tracks repeat purchase among the members of its loyalty programme. The analysis compared members signed up at the checkout with those signed up in the mobile app. The narrative plan below is what the analyst takes to the executive committee.
| Moment | Message, in one sentence | Visual | What the narrative adds |
|---|---|---|---|
| Opening | 90-day repeat purchase among new members signed up at the checkout has gone from 46% to 31% since March. | Monthly curve with two series, checkout and app, only the checkout series in the accent colour | The loyalty programme accounts for 38% of the chain's turnover |
| Middle | Members signed up in the app have stayed at 48%, so the fall is down to the sign-up channel. | The same curve, the app series brought to the front in its turn | The fall is the same across all fourteen stores, which rules out a local cause |
| Middle | Since March, 62% of checkout sign-ups no longer carry an e-mail address, up from 4% before. | Stacked bars, share of sign-ups with and without an address, before and after March | The procedure was shortened in March to cut the wait at the checkout and the address became optional. With no address, the welcome voucher is not sent |
| Ending | Making the address mandatory or sending the voucher by SMS puts CHF 335'000 of annual turnover back in play. | A single figure in large type, the calculation assumption written underneath | 12'000 checkout sign-ups since March, fifteen points of repeat purchase lost, average basket CHF 62 and three visits a year. The assumption: a member who does not buy again within 90 days is lost for the year |
The sorting that precedes this plan dropped more material than it keeps. The stability of the age structure was left out of the case put to the committee, checked and without effect on the decision. The uniformity of the fall across the fourteen stores stayed in the narrative without becoming a curve, because it serves to rule out a local cause and nobody will ask to see it. Four messages survive from an analysis that had produced some twenty findings.
The dashboard had been showing the first two messages since March and nobody had moved. The committee ruled in twenty minutes: the e-mail address becomes mandatory again at checkout sign-up, with the welcome voucher sent by SMS to members enrolled since March. It sent one question to an A/B test in three stores, whether the fall is down to the missing voucher or to a different customer profile. The cause, the costing and the option to settle were in no data.
Visualisations
Two things call for a graphic form. The overlap of the three components is drawn, since the point is what each intersection produces. The working grids are written as tables: the narrative plan carries one moment per row, the sorting of findings one destination per finding.
The visuals of the narrative itself belong to technical visualizations. Storyboarding gives the paper draft of the narrative plan before the tool is opened.
Cost
| Phase | Level | Justification |
|---|---|---|
| Preparation | Medium | Sorting the findings takes an hour once the analysis is finished. Recovering the business events that explain the movements means going and questioning several people, and no tool does that search for you. |
| Execution | Medium | The IIBA guide counts production time among the technique's limitations. Writing the narrative, building the message visuals and cutting take of the order of a day for a twenty-minute delivery. |
| Documentation | Low | The narrative is itself the deliverable. What is kept behind it, the query, the dataset and the assumptions, belongs to the analysis and already exists. |
Tools
Paper or a whiteboard remains the first tool: the narrative plan is laid down in four lines before anything is opened, and a plan laid down inside a visualisation tool takes the shape of the charts that tool can produce.
Visualisation platforms carry the narrative natively: Tableau chains views into what it calls a story, Power BI combines bookmarks and narration to freeze a commented sequence of states, Qlik Sense offers the same mechanism under the name data storytelling. They have the advantage of keeping the link to the data source, hence of surviving a refresh of the figures, and the drawback of pushing towards a chain of charts where the business context is text.
Presentation software paired with the spreadsheet remains where the delivery is most often built, at a price: the figures are frozen at the moment of copy-and-paste and the version presented can no longer be traced back to the query that produced it.
Notebooks (Jupyter, Quarto, R Markdown) suit a technical audience or a delivery that has to be replayed every period: the text, the code and the charts live in one versioned file. Narrative generators, such as the Smart narrative visual in Power BI, write the descriptive commentary of a chart. They describe what the curve does, never why it does it.
Sources
- IIBA, Guide to Business Data Analytics, §3.7 Data Storytelling: the definition, the six principles, the four elements (audience, narrative, visualisation, storytelling), the distinction between primary and secondary findings, the strengths and the limitations.
- Brent Dykes, Effective Data Storytelling: How to Drive Change with Data, Narrative and Visuals, Wiley, 2019: the three-component model, data, narrative and visuals.
- Cole Nussbaumer Knaflic, Storytelling with Data: A Data Visualization Guide for Business Professionals, Wiley, 2015: designing the visuals from the audience, decluttering and the rule of one message per visual.