AI Analytics
Ask a question in plain words, get the answer from your own numbers.
What is AI analytics?
AI analytics connects your business systems so that questions asked in ordinary language are answered from your own data, with the calculation shown alongside the answer so it can be checked rather than trusted.
Most business data sits unused because getting an answer means asking somebody to build a report and then waiting several days for it.
By the time the report arrives, the decision has usually been made on instinct, and the report confirms or contradicts it without changing anything.
We connect your systems so questions can be asked directly. Which products lost money last quarter. Which customers have stopped ordering. Which branch has the slowest deliveries. The answer comes with the figures behind it, so you can check the working.
Why the calculation stays on screen
Duara builds analytics access from Nairobi for businesses whose data is spread across an accounting package, a spreadsheet and whatever the operations team uses.
We always show the calculation alongside the answer, and we resist requests to hide it for a cleaner interface. An unverifiable number presented confidently is worse than no number, because it gets acted on.
Who this is for
- Reports take days to produce
- Decisions get made on instinct
- Your data sits in systems nobody queries
- Different departments quote different figures
Industries we serve
- Retail chains
- Distribution and wholesale
- Manufacturing
- Financial services
- Hospitality groups
- Logistics operators
Problems this solves
These are the conditions where question-based analytics changes behaviour rather than adding another dashboard nobody opens.
- Reports arrive after the decision. A question takes three days to answer, so the decision is made on instinct and the report merely files itself.
- Only one person can query the data. Everything goes through whoever knows the reporting tool, which makes them a bottleneck and a single point of failure.
- Numbers that do not reconcile. Sales and finance quote different revenue figures because each pulls from a different system with a different definition.
- Data nobody has ever looked at. Years of transactions sit in the accounting system, unexamined, because examining them requires effort nobody has spare.
What we deliver
Your systems connected into one place, questions answerable in ordinary language, every answer accompanied by its calculation, and recurring questions saved as scheduled reports.
Your systems connected into a single place to query
Plain language questions answered with figures
The calculation shown so every answer can be checked
Recurring questions saved as scheduled reports
Agreed definitions so departments reconcile
Access rules so people see only what they should
Features
- Questions in ordinary language. No query syntax to learn, which is what determines whether anybody beyond the analyst actually uses it.
- Workings shown by default. The calculation appears alongside the answer. We decline to hide it, because an unverifiable number gets acted on.
- Agreed definitions. What counts as a sale, when revenue is recognised, which customers are active. Settled once so figures reconcile.
- Saved recurring questions. Anything asked repeatedly becomes a scheduled report rather than being retyped weekly.
- Row level access control. Branch managers see their branch, directors see everything, and the rules are enforced rather than assumed.
Benefits
- Answers arrive during the decision. A question answered in seconds changes the decision. One answered in three days documents it.
- The analyst stops being a bottleneck. Routine questions are self-served, which frees them for work that genuinely needs analysis.
- Meetings stop arguing about numbers. Agreed definitions mean the discussion moves to what to do rather than whose figure is right.
- Dormant data starts earning. Years of transaction history become answerable, and the first month of questions usually surfaces something unexpected.
Technologies we use
- System connection. Direct links to your accounting, sales and operational systems so figures come from source rather than from exports.
- Semantic layer. Agreed definitions encoded once, so every question uses the same meaning of revenue, customer and order.
- Query generation. Ordinary language translated into database queries, with the generated query visible for checking.
- Access control. Row and column level rules so people see only the data their role permits.
- Scheduled reporting. Saved questions run automatically and deliver to whoever needs them.
Definition agreement takes longer than the technical connection and matters more. Most reconciliation failures are definitional rather than technical.
How the process works, step by step
Analytics access usually takes four to six weeks, with definition agreement occupying more of that than the connection work.
-
Step 1
Free question review
We collect the questions people actually ask, since those determine what needs connecting.
-
Step 2
Definition workshop
We settle what a sale, a customer and an active account mean, which is where most reconciliation problems originate.
-
Step 3
System connection
We connect the sources and validate that totals match what your finance team already reports.
-
Step 4
Supervised querying
A small group asks real questions and checks the workings, which finds definitional errors quickly.
-
Step 5
Wider rollout
Access opens up with permissions applied and the common questions already saved.
Quality assurance and how we work
The commitments attached to a system people will make decisions from.
- The calculation is always visible. We do not hide the workings for a cleaner interface. A number nobody can verify still gets acted on, which is the danger.
- Totals are validated against your accounts. Before anybody uses it, headline figures are reconciled against what your finance team already reports.
- Definitions are agreed, not assumed. We will not guess what counts as an active customer. That gets settled with your team and written down.
- Access rules are enforced at the data. Permissions are applied where the data lives rather than in the interface, which is the only version that holds.
- We say when the data cannot answer. Where a question needs information you do not collect, the answer is that, not an estimate dressed as a finding.
Frequently asked questions about AI analytics
Do we need a data team?
No. That is much of the point. Setup requires technical work, but using it requires only knowing what you want to ask.
What if the answer looks wrong?
The calculation is shown, so you can see how it was reached. A wrong-looking figure usually means the underlying data is wrong, which is worth knowing.
Can you hide the workings for a cleaner look?
We decline. An unverifiable number presented confidently still gets acted on, and that is more dangerous than a slightly busier interface.
Why does definition agreement take so long?
Because most reconciliation failures are definitional rather than technical. Sales counts an order when placed, finance when invoiced, and settling that is the work.
Which systems can it connect to?
The common accounting, CRM, ERP and point of sale systems, plus spreadsheets where that is genuinely where data lives.
Will figures match our accounts?
They are validated against what your finance team already reports before anybody uses the system. If they do not match, we find out why first.
Can we restrict what people see?
Yes, at row and column level, enforced where the data lives rather than in the interface.
How long before we can start asking questions?
Four to six weeks, with definition agreement taking more of that time than the technical connection work.
What sort of questions can it answer?
Anything your data supports. Which products lost money, which customers stopped ordering, which branch is slowest.
What if we do not collect the data needed?
Then the answer is that the question cannot be answered, rather than an estimate presented as a finding.
Can questions be saved?
Yes. Anything asked repeatedly becomes a scheduled report that runs and delivers automatically.
Does it replace our dashboards?
It complements them. Dashboards answer the questions you knew you would have. This answers the ones you did not.
Can it explain why something changed?
It can show where in the data a change originated. Whether that is the cause is a judgement, and it will say when the data cannot establish it.
Does our analytics data leave our systems?
Queries run against your own systems. Data is not exported to train public models and stays within your control.
Can it analyse Swahili text data?
Yes, and questions can be asked in either language, which matters for staff who are more comfortable in Swahili.
How current are the answers?
As current as the connection allows, which for most systems means near real time rather than yesterday.
What if two departments disagree with a figure?
That is usually a definition problem, and the semantic layer is where it gets settled permanently rather than argued each month.
Can it handle several years of history?
Yes, and historical questions are frequently where the most useful findings appear in the first month.
Who should attend the definition workshop?
Finance, sales and operations together. Any absent department will disagree with the result later.
What does analytics access cost to run?
Query volume and data size determine it. We project the monthly figure from your own usage expectations before anything is agreed.
Glossary of terms
Analytics vocabulary in plain terms.
- Semantic layer
- The agreed definitions of business terms, applied consistently so every query means the same thing.
- Reconciliation
- Checking that figures from different systems agree, and understanding why when they do not.
- Row level security
- Restricting which records a person can see, so a branch manager sees only their branch.
- Drill down
- Moving from a summary figure to the individual records behind it.
- Scheduled report
- A saved question that runs automatically and delivers on a set rhythm.
- Data source
- A system supplying figures, such as your accounting package or point of sale.
Related services in AI Implementation
Talk to us about AI analytics
Write down the five questions you most often want answered and cannot get quickly. That list tells us within an hour whether this is feasible.
Start your project