THE CHALLENGE
Retail sales data arrived from four vendors large FMCG vendors in different Excel formats. Inconsistent worksheet structures and gaps in key fields made it difficult to bring the files together reliably and compare performance over time.
The work was framed around three practical questions:
What we built
A standardised retail data model. A common Excel template was defined to align the four vendor datasets and support consistent downstream reporting.
Our system automatically mapped and sorted vendor data into the required fields using predefined logic and percentage-based confidence scoring. Where the system was confident in a match, the data was assigned automatically. Where confidence fell below the required threshold, the system flagged the fields for staff review, allowing uncertain matches to be validated rather than silently accepted.
This approach reduced manual data mapping while retaining human oversight and improving the consistency of the data prepared for downstream reporting.
Configurable reporting
Six key reports was developed to organise the information into distinct views:
The Sales Performance, Brand Performance and Priority Report pages are connected to live data sources and use actual, validated figures. The visuals for Market Share, Promotion Effect and Stock in Stores have been developed, but their figures remain placeholders or preliminary values pending receipt of final, validated source data.
DataTalk: ask questions in natural language
DataTalk adds a conversational way to explore business data. Instead of needing to know SQL or navigate complex tables, users can ask a question in plain English. The system translates the question into a database query and returns an answer from the available data, helping non-technical users investigate performance and follow up on what they find.
For Primeline, this complements Power BI: dashboards provide a consistent view of key metrics, while natural-language querying gives users a flexible way to ask specific questions of the underlying information. It shifts the proposition beyond reporting alone towards making the data easier for business teams to interrogate and use.
The exact production scope, connected data sources and validation status of DataTalk should be confirmed before publication. This avoids implying that every capability is live or fully deployed where that status has not been verified.
Foundations underneath
Cloud hosting and resilience
AWS provided resizable compute capacity without requiring physical server purchases, allowing capacity to be adjusted as demand changed. The source material describes AWS service-level commitments of up to 99.99% for applicable services and notes that monthly availability depends on the deployment design, including whether instances are deployed across multiple Availability Zones.
The hosting configuration used the Dublin availability zone to support European data residency requirements. The source material also identifies AWS alignment with recognised regulatory frameworks and independent compliance audits.
How the work was delivered
The work was organised in three phases: data cleaning, data automation and reporting. The initial phase established the target schema and identified differences between vendor files. The reporting phase delivered the report structure and connected the first three pages to live, validated data.
DataTalk used logic to automatically map and structure fragmented vendor data, assigning confidence scores to potential matches and flagging uncertain results for staff review. Once the data was structured, users could query it in natural language, making it easier to retrieve insights without having to manually navigate datasets or construct database queries.
Why It Matters
The combined approach is intended to make retail data both more consistent and more accessible. Standardisation supports reliable comparison across vendor files; Power BI provides repeatable views of performance; and DataTalk allows business users to ask targeted questions in everyday language. Together, these capabilities can reduce the friction between having data and getting an answer from it, while creating a foundation for deeper analysis of market share, promotional uplift, stock cover, out-of-stock incidents and distribution gaps once the required data is validated.
TECHNOLOGY USED
Flutter
Figma
Node.js
AWS
