Ask your business data anything, in plain language. Power BI shows the KPIs you planned for. The data analytics agent answers the questions that come after, with the figures behind every answer.
What's our conversion rate by client group this month?
readfct_applications
readdim_client
definitionconversion = signed loans / applications · owner Sales · v2
select c.risk_group,
date_trunc('month', a.applied_at) as month,
count(*) as applications,
count(*) filter (where a.status = 'signed') as signed_loans
from fct_applications a
join dim_client c on c.client_key = a.client_key
where a.applied_at >= date '2026-09-01'
group by 1, 2;
24% of applications became signed loans this month, down from 26% in September.
Low and medium risk held steady, at 41% and 26%. The drop is in high risk: 9%, down from 14%.
Why is conversion dropping for high-risk clients?
readfct_applications
readfct_offers
readfct_rate_changes
readdim_client
definitionrisk group = scoring band at application · owner Risk · v4
select date_trunc('week', a.applied_at) as week,
count(*) as applications,
count(*) filter (where a.status = 'signed') as signed_loans,
avg(o.offered_rate) as offered_rate,
avg(a.hours_to_decision) / 24 as days_to_decision
from fct_applications a
join dim_client c on c.client_key = a.client_key
left join fct_offers o on o.application_key = a.application_key
where c.risk_group = 'high'
and a.applied_at >= date '2026-09-07'
group by 1
order by 1;
It started on 5 October, when the rate for high-risk clients went up 3 points. Since then about 8 in 100 of them sign, against 14 in September.
Demand held: high-risk applications are up 18%. The process held too: approval still takes about a day.
So they apply, see the new rate and walk away. It is the price, not the process.
Product demonstration with illustrative sample data.
Power BI is good at the reports and KPIs you defined up front. Then a number moves and someone asks why. Answering that used to take an analyst days or weeks, and junior analysts rarely got there. The agent answers from one question in plain language, with no scope written first.
Based on a question from a client demo. The figures are illustrative.
Speed is only part of it. Your data is connected and modelled, so the agent can follow a question across systems. It finds links no report was built to show.
Also found · same chat
| Risk group | New loans | Interest income | Converts |
|---|---|---|---|
| Low | 54% | 31% | 41% |
| Medium | 31% | 33% | 26% |
| High | 15% | 36% | 9% |
High-risk clients are 15% of new loans but 36% of interest income. Each point of their conversion is worth about five times a point from low-risk clients.
Core lending system, anonymised · October
The next questions they asked
It does better work than most senior analysts, and it showed me things I would not have thought to look for.
From a client demo on its own data. The client's name is withheld, and the figures are illustrative.
Finance, marketing, operations and leadership each ask their own questions, from the same modelled data and within their own access. Questions that come back every month become a Power BI page.
Accounting, payment and ERP data in one model. Ask why a margin moved or which invoices to chase, then open the source records behind the answer.
Example question: “Which cost categories explain the margin drop compared with last month?”
Ad spend next to orders and CRM data. Google, Meta and TikTok ads come in through standard connectors, so the cost per customer uses your own sales.
Example question: “Which channels bring customers with the highest repeat-purchase value?”
Spend and first-order revenueFirst-order revenueAd spend
First-order revenue by channel
Campaign performanceSeptember 2026
| Campaign | Channel | Spend | New customers | CAC | ROAS |
|---|---|---|---|---|---|
| Brand search | Google Search | €4,120 | 162 | €25 | 21.4 |
| Generic, wholesale | Google Search | €10,470 | 222 | €47 | 11.8 |
| Shopping, all products | Google Shopping | €9,800 | 241 | €41 | 9.8 |
| Prospecting, Baltics | Meta Ads | €8,960 | 98 | €91 | 3.2 |
| Retargeting | Meta Ads | €2,840 | 54 | €53 | 6.8 |
| Win-back email | Email and CRM | €1,200 | 118 | €10 | 36.7 |
Definitionsapproved by Marketing
Platform-reported conversions are not compared directly. Leads that ordered within 30 days: Search 6.8%, Meta 2.1%.
Illustrative sample data.
Operations and finance data in one place. Spot projects running over budget, stock that is not moving and how many deals the team closes each day.
Example question: “Which projects are forecast to exceed budget, and what has changed?”
Cost variance by projectactual against budget, € thousands
Under budgetOver budget
Tallinn warehouse expansionwhat changed
Overrun+€52k
Sources: ERP project ledger, purchase orders, timesheet spreadsheet
Slow-moving stockno movement for 90+ days
| SKU | Product | Warehouse | Days | Value |
|---|---|---|---|---|
| LV-RK-1102 | Racking beam, 2,700 mm | Riga | 163 | €24.9k |
| LT-CR-2210 | Cold-room shelving kit | Vilnius | 214 | €18.4k |
| EE-PK-0418 | Pallet wrap, 23 µm | Tallinn | 187 | €11.2k |
| LV-TL-0731 | Tail-lift battery | Riga | 141 | €7.8k |
| LT-FL-0905 | Floor marking tape | Vilnius | 122 | €3.1k |
Illustrative sample data.
A monthly brief across finance, sales and operations, written by the agent from the same definitions as every report. Each line names its source.
Example question: “What changed this month, and what should I pay attention to?”
Illustrative sample data.
An agent is only as good as the data behind it. Before anyone asks a question, we connect your sources and clean and test the data. Then we agree each metric's definition with you. Power BI and the agent read the same model, so a figure is the same in both.
Built on
PostgreSQL
dbt
Claude
Power BIEU hosting
Codnity WarehouseEU-hosted PostgreSQL
AccessRead-only
PermissionsMinimum required
Sources agreed6
Personal dataAnonymised
fct_invoice_linesCleaned, joined, tested
| Invoice | Customer | Net € |
|---|---|---|
| NV-LT-0912 | Nordwind Logistics | 18,240.00 |
| NV-LV-1204 | Riga Metalworks | 12,115.00 |
| NV-LT-0913 | Baltic Foods Group | 9,870.50 |
| NV-EE-0388 | Vento Nordic | 6,402.75 |
Business definitionsDocumented, with owners
Gross marginFinance · v3
RevenueFinance · v2
OverdueAccounting · v1
Revenue · model vs ledgerSeptember 2026 · € thousands
| Entity | Model | Ledger | Diff. |
|---|---|---|---|
| LV | 1,982.4 | 1,982.4 | 0.00 |
| LT | 1,104.7 | 1,104.7 | 0.00 |
| EE | 612.9 | 612.9 | 0.00 |
Reconciled · difference €0.00
Before every refreshTests and access
142 tests passed
Definitions reviewedFinance
Role permissions4 roles
Governed business modelNorvi Group · September
Tables41
DefinitionsWith owners
Refreshed06:12
Power BI reports
Ask the agent
Inspect the evidence
The data analytics agent needs each department's data in one place. Google, Meta and TikTok ads use standard connectors. Your other tools connect as custom integrations. Open a folder to see the main systems and how each one connects.
Horizon (Visma)custom integration
Jumiscustom integration
Dynamics 365 Business Centralcustom integration
Moneolive connector
Shopifycustom integration
Magentocustom integration
Salesforcecustom integration
HubSpotcustom integration
Google Adsstandard connector
Meta Adsstandard connector
TikTok Adsstandard connector
Google Analyticscustom integration
Excel and SharePointcustom integration
Google Sheetscustom integration
Jira and ClickUpcustom integrationNot on the list? If a system has an API or a file export, we can usually connect it. We confirm each connector during scoping.
For the data analytics agent, we agree the access rules with you, role by role. The data layer checks them before any result reaches the AI model. Salaries and personal data need explicit permission, and every question is logged.
Separate environments: your data is kept apart from other clients' data.
Access by role: finance, marketing and HR each see only their own data.
Read-only access: we only read from your systems, where the connector supports it.
Anonymised warehouse: names, ID numbers and contact details can be replaced before the data reaches the warehouse.
Traceable answers: each answer shows its definition, sources and query.
Agreed in writing: which sources we connect, who sees what, how long data is kept and which AI provider handles questions.
Show salary costs by person for September
Querying Codnity WarehouseCodnity Warehouse: request denied by the data layer · payroll is outside the Marketing role's scope
I can't show salary costs by person. Payroll is outside the Marketing role's permissions, and the data layer refused the query before any payroll figures were returned to me.
I can help with what your role covers, such as September ad spend by channel, campaign results or cost per new customer. Which would be most useful?
Role accessby data domain
| Role | Finance | Marketing | Sales | Payroll | Operations |
|---|---|---|---|---|---|
| Finance | access | no access | access | no access | access |
| Marketing | no access | access | access | no access | no access |
| Sales | no access | access | access | no access | access |
| HR | no access | no access | no access | access | no access |
| Leadership | access | access | access | access | access |
Product demonstration with illustrative sample data.
Codnity Data builds the Power BI reports and the data analytics agent on one model. A figure reads the same in both, and this is one of those models.
Confidential client · Latvia SMB
Ten report pages run off one model and refresh together, so the numbers agree.
Power BI, data engineering and the data analytics agent, for finance and operations.
A data analytics agent is an AI assistant connected to your company's modelled data. You ask a business question in plain language, and it queries the data and answers with the figures, the sources and the calculation behind them. Codnity connects and models the data first, so the answers match your reports.
Power BI shows the reports and KPIs you defined in advance, and a data analytics agent answers the questions that come after them. When a KPI moves, you ask the agent why, without waiting for a new report. Both read the same data model as our Power BI development work, so the numbers agree.
No, you don't define the question or the scope first. You ask in plain language, as you would ask an analyst, and follow up in the same chat. The scoping work happens once, when we connect and model your data, so a new question needs no new project.
A data analytics agent does the slow part of an analyst's work: finding the data, joining it and running the first analysis. Answers that took days or weeks arrive in minutes, so questions that never got asked now get answered. People still decide what to do with an answer, and the hardest questions still benefit from an expert's review.
The data analytics agent answers any question your connected data can answer. For example, why margin fell, which channel brings customers who stay, which projects run over budget or which client groups are priced wrong. Finance, marketing, operations and leadership each ask within their own access.
No, the agent can't answer every question correctly. Answers depend on the data connected, the agreed definitions and how the question is asked. Each answer shows its sources, definition and calculation, so a person can check it before acting on it.
No, a data analytics agent is more than a chatbot on a database. Before it answers anything, we connect your sources, clean and test the data and agree each metric's definition with you. The agent and your Power BI reports read the same model, so a figure reads the same in both.
The data analytics agent connects to accounting, ERP, CRM, ecommerce, ad platforms and spreadsheets. Google, Meta and TikTok ads come in through standard connectors, and Moneo has a live connector. Other systems with an API or a file export are connected as custom integrations, confirmed during scoping.
The data is refreshed daily as standard, and more often if the source system allows it. Every answer and every report shows when its data was last updated, so you always know how fresh a figure is.
Each person sees only the data their role allows. We agree the access rules with you role by role, and the data layer checks them before any result reaches the AI model. Salaries and personal data need explicit permission, the warehouse can hold anonymised data, and every question is logged.
Your data sits in a dedicated PostgreSQL environment in the EU, and it stays yours to export. Backups, retention and deletion are set out in the proposal. Where an AI provider processes a question depends on the provider and plan, and we document it.
It depends on the AI provider and plan, and on the setups we use it is not. Anthropic's commercial terms for Claude Team, Enterprise and the API say inputs and outputs are not used for training by default. Before go-live, we document what is sent to the provider and how long it is kept.
Yes, the data analytics agent runs in Claude, connected to your data warehouse. Your team asks questions in an app they may already use, and every answer shows its evidence. People new to Claude get a short setup and example questions for their role.
You start with a walkthrough: tell us where your numbers live and one question you want answered. We connect one department's sources first, so you ask your own questions on your own data early. The cost depends on the sources and the modelling, and we quote it after the walkthrough.
Tell us where your numbers live and one question you want answered. We'll show you what the data analytics agent would need to answer it.
The agent answers from your own data, so who can ask what matters as much as the answer.
What we learn building AI systems, reports, secure software and stores, written down.
All insightsCody
Lending · Monthly KPIs
