AI Dashboard Development
Built around a decision, not around whatever the data happened to allow.
What is AI dashboards?
An AI dashboard presents your operating figures alongside plain language explanation of what changed and what is unusual, built around the specific decisions the people using it have to make.
Most dashboards are built from the data outward. Somebody lists what can be measured, puts it all on a screen, and the result is looked at twice and then ignored.
A dashboard that gets used is built from a decision backward. Which decision does this person make weekly, what would change their mind, and what is the smallest set of figures that answers it.
The AI part is narrow and useful: describing in plain language what moved and what looks unusual, so somebody who does not read charts fluently still gets the point.
The question we ask before designing anything
Duara builds dashboards from Nairobi for management teams who have reporting but do not use it, which is more common than having no reporting at all.
We start by asking which decision the dashboard serves, and where the answer is that it would be nice to see everything, we say that is how unused dashboards get built. A screen with forty charts is a data dump with a colour scheme.
Who this is for
- You have dashboards nobody opens
- Charts are read differently by different people
- Numbers are questioned rather than acted on
- Reporting shows everything and answers nothing
Industries we serve
- Retail and distribution
- Manufacturing
- Financial services
- Healthcare groups
- Logistics
- NGOs and donor programmes
Problems this solves
These describe reporting that exists and is not used, which is a design failure rather than a data one.
- Dashboards nobody opens. A screen was built showing everything available, and it answers no specific question anybody has.
- Charts read differently. Two managers draw opposite conclusions from the same chart because nothing states what it means.
- Figures questioned, not acted on. A number is challenged and cannot be traced, so the meeting discusses the data rather than the decision.
- No signal about what changed. Everything is presented at equal weight, so the one measure that moved is not obvious.
What we deliver
Dashboards designed around named decisions, with plain language explanation of what changed, anomalies surfaced rather than buried, and every figure traceable to source.
Dashboards designed around named decisions
Plain language explanation of what changed
Unusual movements surfaced rather than buried
Every figure traceable back to source
Views by role rather than one screen for everybody
Delivery to phone and email, not only a login
Features
- Designed backward from a decision. We start with what somebody has to decide, because starting from available data is how unused dashboards get built.
- Explanation in plain language. What moved and what is unusual, written out, so somebody who does not read charts fluently still gets the point.
- Anomalies given prominence. The measure that changed is surfaced rather than presented at the same weight as everything else.
- Traceable figures. Any number can be followed to source, which keeps meetings on the decision rather than the data.
- Delivered where people are. A summary to phone or email, since a dashboard requiring a login gets opened far less than one that arrives.
Benefits
- Reporting gets used. A dashboard answering a real decision is opened. One showing everything is not.
- Meetings move faster. Traceable figures and stated explanation remove the argument about whether the number is right.
- Changes get noticed. Surfacing what moved means a shift is seen when it happens rather than at the next review.
- Non specialists participate. Plain language explanation widens who can use the reporting beyond the people fluent in charts.
Technologies we use
- Decision led design. Building each view around a named decision and the person who makes it.
- Data connection. Reading from your source systems rather than from manual exports.
- Anomaly detection. Identifying genuine departures from your normal ranges rather than ordinary variation.
- Narrative generation. Describing what changed in plain language, stating movement rather than asserting cause.
- Drill through. Following any figure back to the underlying records.
The explanation layer must be conservative. A dashboard that confidently states a cause it cannot know teaches people to distrust everything else on the screen.
How the process works, step by step
A dashboard usually takes three to six weeks, and the first conversation is about decisions rather than data.
-
Step 1
Free decision review
We establish which decisions the dashboard should serve and who makes them.
-
Step 2
Measures agreed
The smallest set of figures that answers those decisions is agreed, which usually means removing most of the original list.
-
Step 3
First version narrow
A deliberately small dashboard goes live, since additions based on use are better than removals after complaint.
-
Step 4
Explanation tuned
The narrative layer is checked against what your team would actually have said about the same movement.
-
Step 5
Delivery and access
Views by role and delivery to phone or email are set up so it reaches people rather than waiting for them.
Quality assurance and how we work
The commitments attached to reporting people will make decisions from.
- We ask which decision it serves. Where the answer is that it would be nice to see everything, we say that is how unused dashboards get built.
- The explanation states movement, not cause. A dashboard that confidently asserts a cause it cannot know teaches people to distrust everything else on the screen.
- Every figure is traceable. A number that cannot be followed back gets challenged in a meeting and cannot be defended.
- We start smaller than you ask for. Adding views based on real use works. Removing them after nobody looks is a conversation nobody has.
- We report when the data cannot support the view. Where a requested measure is not reliably available, we say so rather than displaying something approximate.
Frequently asked questions about AI dashboards
Why do our dashboards go unused?
Usually because they were built from available data outward rather than from a decision backward. A screen with forty charts is a data dump with a colour scheme.
What do you need from us to start?
Which decisions the dashboard should serve and who makes them. That conversation comes before any discussion of data.
What does the AI actually do here?
It describes what changed and what looks unusual in plain language. That is a narrow job and it is the useful one.
Will it tell us why something changed?
No. It states movement rather than asserting cause, because a dashboard that confidently guesses a cause destroys trust in everything else on it.
Can we see where a number came from?
Yes, any figure can be followed to source, which is what keeps a meeting on the decision rather than the data.
Can we have everything on one screen?
You can, and we will advise against it. We start smaller than clients ask, because adding on use is easier than removing after nobody looks.
How long before a dashboard is live?
Three to six weeks, with the first version deliberately narrow.
Does it work on a phone?
Yes, and summaries can be delivered by message or email, since a dashboard requiring a login gets opened far less often.
Which data sources can it use?
Your ERP, ledger, CRM or operational systems, read directly rather than from manual exports.
What if two systems disagree?
We surface the disagreement rather than picking one, and resolving it is usually a definition conversation.
Can different roles see different things?
Yes, and they should. One screen for everybody is a common reason dashboards stop being relevant to anyone.
Will it alert us to problems?
It surfaces unusual movements. Where you want active alerting, that is a monitoring job and we would scope it separately.
Can it show donor or grant reporting?
Yes, and NGO clients frequently need programme views alongside the financial ones.
What if a measure we want is unreliable?
We say so rather than displaying something approximate. An unreliable figure on a dashboard gets treated as fact.
Do we need to drop our current BI platform?
Not necessarily. Where you have one that works, we build on it rather than adding another platform to maintain.
Can it compare branches or sites?
Yes, and comparison views are often where the most actionable findings appear.
Can the explanation be written in Swahili?
The narrative layer can be produced in Swahili where that suits your users.
How often does it refresh?
As often as your sources allow and the decision requires. Daily is common; real time is rarely necessary and costs more.
Who can see which figures?
Permissions follow your existing access rules, so nobody sees data they could not already reach.
What does a dashboard project cost?
View count and data source complexity determine it, quoted after the decision review.
Glossary of terms
Dashboard terms in plain language.
- Decision led design
- Building a view around a specific decision rather than around available data.
- Anomaly
- A movement outside the normal range for that measure, as opposed to ordinary variation.
- Narrative layer
- Plain language description of what changed, sitting alongside the charts.
- Drill through
- Following a figure on screen back to the records behind it.
- Role based view
- A version of the dashboard showing what one role needs rather than everything.
- Data dump
- A screen showing everything measurable, which is what most unused dashboards are.
Related services in AI Development
Talk to us about AI dashboards
Check when your current dashboard was last opened, and by whom. That answer usually decides whether this is worth doing.
Start your project