Case studies

What we built, with the numbers.

We always show the starting point and what changed. Clients appear by sector until they approve using their name.

Industrial manufacturer

An AI agent that drafts the technical support team’s replies

The technical support team answered service requests from around the world using memory and old emails. The goal was to answer every request within 24 hours.

We organised the request history as searchable cases. For each new request, the agent suggests a reply and shows the cases behind it. The team reviews and sends. Twice a day, a status update on open requests lands in Teams.

AI agentsMicrosoft TeamsPower AutomateGitSQL
Before the pilot 62%
Pilot target 100%
3,483past requests organised as searchable cases
62% → 100%requests answered within 24 h: the baseline and the pilot’s target

Industrial manufacturer

Spare-part delivery dates

The parts team received around 30 emails a day from sales asking when a part would arrive. Many were repeats.

We built a web platform connected directly to the ERP: sales searches for a part or order and sees which purchase order covers it and when it should arrive. It also shows where parts can be reallocated between orders.

ERP integrationWeb appPython
Emails a day asking for dates ~30
With the portal: self-service 0 emails
~30/day“when will it arrive?” emails the portal now answers
1,273supplier receipts behind the lead-time estimate (median)
Dailyautomatic, read-only refresh from the ERP
BUSINESS SYSTEMERPalready in placeSales ordersopen customer linesPurchasesto suppliersReceiptsactual lead timesdirect linkWEB PLATFORMDelivery portalwhat it does, every day1Collectsdaily, on its own2Matchesline by line3Estimatesthe arrival dateSales teamchecks, no emailsReallocationsbetween ordersParts teamfewer questions
The platform in use: open lines, coverage by purchase order and expected date (customers and parts blurred)

Services company

Finance back office on Microsoft Fabric and Power BI

We use at NAVRA what we propose to clients. We connected the company bank account directly to Fabric to get categorised costs and income without exporting statements.

A Fabric notebook collects transactions daily, with credentials kept in Azure Key Vault. Data flows through a layered lakehouse (bronze, silver, gold) into a Power BI report in Direct Lake.

Microsoft FabricPower BIAzure Key VaultVAT forecasting
0 statementsexported by hand: transactions arrive on their own every day
3 layersbronze, silver and gold in the Fabric lakehouse
0 secretsin code: credentials live in Azure Key Vault
Banktransactions every dayRulescompany categoriesTax calendarVAT deadlinesLAKEHOUSEOrganised datathe bank, with no statement exportsBronze· collectedas they come from the bankSilver· cleaneddeduplicated, categorisedGold· readyfor reports and forecastsCostscategorised automaticallyDaily trackingbalance and costs todayVAT forecastwhat to pay this quarter

NAVRA internal project

Vasco, the assistant that works with us every day

We are two partners, and the company’s context was scattered across emails, meetings, Notion and SharePoint. After every meeting, one of us spent around 30 minutes writing down what was decided and preparing the next steps.

We built Vasco on Claude Code, with a knowledge base of plain text files in a Git repository. Each morning it tells each partner what matters that day. After every meeting it reads the transcript, saves the summary, does what it was asked to do and posts a status update in the Teams chat. It answers questions with the source cited and drafts emails in each partner’s style. It never emails clients: technical rules block it.

AI agentsAgent skillsMicrosoft TeamsGitKnowledge base
~30 mina partner used to spend after every meeting recording decisions and preparing next steps
15 minevery 15 minutes on working days, Vasco looks for meetings that have ended and handles them
0 emailssent to clients by the assistant: it only creates drafts, which the partner reviews and sends

E-commerce

Sales and returns dashboards for management

Returns processing relied on manual reports and several sources that did not reconcile.

A single data model over ERP, CRM, Excel and API sources, row-level security by department, and dashboards for sales performance and returns analysis.

Power BIData modellingERP, CRM, Excel & API data
Processing time before 100
After 44
−56%in processing time
4 sourcesERP, CRM, Excel and APIs in one model
NAVRA Power BI report

Next step

Start with a 30-minute call.

Tell us where time goes on data and reporting. On the call we will say plainly whether and how we can help. If it makes sense, the next step is a diagnostic with a prioritised plan, effort and budget.

You talk directly to José or Gonçalo.
Book a 30-minute call →