How to Build an FMCG Sales Intelligence Dashboard With AI on Creatr

FMCG sales teams drown in spreadsheets - territory files, SKU lists, weekly sellout reports - and still make gut calls because no one has time to cross-reference it all. This tutorial walks you through building a sales intelligence dashboard on Creatr where the AI is not a widget tacked onto a table; it is the analysis layer that makes the data useful to a field rep who has three minutes between store visits.

The build covers data import, target management, an explore layer with charts, a plain-English "ask" interface grounded in your actual numbers, automated alerts, and role-based views for reps versus managers. No code is written. Every step is a prompt you type into Creatr.

Before you start

1. Gather your data files. You need at minimum: a territory master (territory ID, region, assigned rep), a SKU master (SKU code, category, brand, price tier), and at least one period of historical sellout data (date, territory, SKU, units sold, value). CSV exports from your ERP, distributor portal, or even a well-structured Excel sheet all work.
2. Decide on your target grain. Targets in FMCG are usually set at territory x SKU category x month. Know this before you start - it shapes how the target table is designed.
3. Map your roles. At minimum you will have field sales reps and sales managers. If you have a national sales manager or a category head who needs a different view, note that now.
4. Have a Creatr account. Sign up at getcreatr.com. The AI build service ships your app; you describe what you want.

Step 1: Describe the dashboard and its users

Start a new project in Creatr. The first prompt is your product brief. Be specific about who uses it and what decisions it has to support.

Build a sales intelligence dashboard for an FMCG company. Two user roles: field sales reps and sales managers. Reps use it on mobile while traveling - they need their territory performance, their targets, and a way to ask questions about their numbers. Managers see all territories, can compare performance across regions, and set targets. The dashboard should show sellout volume and value, performance against target, trend over time, and top and bottom performing SKUs per territory. The AI layer should let any user type a question in plain English - like "which SKUs are declining in the North region?" or "show me territories below 70% of target this month" - and get a direct answer with the supporting data. No decorative charts. Every number should trace back to the uploaded data.

Creatr will confirm the structure it intends to build. Read it carefully. If the role breakdown or the described views do not match what you need, correct them before proceeding.

Step 2: Import sales and territory data

Once the app shell is confirmed, tell Creatr how data enters the system.

Add a data import module. It should accept CSV uploads for three entity types: Territory Master (columns: territory_id, territory_name, region, assigned_rep_email), SKU Master (columns: sku_code, sku_name, category, brand, price_tier), and Sellout Data (columns: date, territory_id, sku_code, units_sold, value_inr). On import, validate that territory_id and sku_code exist in the master tables before accepting sellout rows. Show a summary after each upload: rows accepted, rows rejected, reason for rejection. Store all imports with a timestamp and the name of the user who uploaded. Managers can upload any file type. Reps can only upload sellout data for their own territory.

After this step, upload your actual files through the app interface. The validation step matters - garbage data at import is the most common reason AI answers go wrong downstream. Fix rejections before moving on.

Step 3: Set up target management

Targets need to be editable without re-importing the whole dataset. Build a dedicated target management view.

Add a target management section visible only to managers. Targets are set at the level of territory x SKU category x calendar month. The table should show: territory name, SKU category, month, target units, target value, and who last edited the row and when. Managers can edit targets inline. Add a bulk import option so managers can paste or upload a CSV of targets rather than editing row by row. When a target is saved, flag any territory-category-month combination that has no corresponding sellout data yet so the manager knows it is forward-looking. Show a "targets coverage" summary: what percentage of active territory-category combinations have a target set for the current month.

Test this by entering a handful of targets manually. The coverage summary will tell you how complete your setup is before you move to the analytics steps.

Step 4: Build the explore layer - filters, charts, and trend views

This is the visual analytics layer. It needs to be fast to navigate and honest about what it is showing.

Build an explore section. Default view shows the current calendar month. Filters: date range (single month or custom range), region (multi-select), territory (multi-select, cascades from region), SKU category (multi-select), brand (multi-select). Charts to include: a bar chart of sellout value by territory for the selected period, a line chart of weekly sellout trend for the selected territory or region, a heatmap or ranked table showing SKU category performance by territory (units sold vs. target, color-coded green/amber/red by percentage of target achieved), and a top-10 / bottom-10 SKU table for the selected scope. All charts should show the exact numbers on hover. No chart should show data outside the user's permitted scope - reps only see their own territory, managers see what their filter is set to. Add a "vs previous period" toggle that adds a comparison line or column to the trend chart without replacing the primary data.

Once this is live in the preview, spend time stress-testing the filters. The most common problem at this stage is that a cascading filter (region then territory) breaks when you select multiple regions. Note any gaps and feed corrections back as follow-up prompts.

Step 5: Wire up the AI ask layer

This is where Creatr AI does the heavy work. The ask layer lets any user type a question and get a direct answer drawn from the data already in the system. The critical constraint is grounding - the AI must answer from the numbers, not from general knowledge about FMCG.

Add an "Ask" tab to the dashboard. It should contain a text input where any user can type a question in plain English. When the user submits a question, the AI should query the sellout data, targets, and master tables to produce a direct answer. The answer must cite the specific numbers it is using - for example, if the answer is "Biscuits in the East territory are down 18% versus last month," it should show the underlying figures (this month: 4,200 units; last month: 5,120 units) so the user can verify. The AI should respect role-based data access: a rep asking "what is my worst-performing SKU?" only sees their own territory data. A manager can ask across all territories. If the question cannot be answered from the available data, the AI should say so clearly rather than guess. Show the last 10 questions and answers in a history panel for the session.

Use Creatr AI Knowledge to make the ask layer aware of your domain context - for example, which SKU categories are seasonal, what your company's threshold for "underperformance" is, or how territories are grouped into regions. Feed this as a follow-up prompt:

Connect the Ask layer to a knowledge base. Add a knowledge settings page (managers only) where I can type or paste context: business rules like "target achievement below 70% is flagged as underperforming," seasonal notes like "ice cream category targets are 40% higher in Q2," and territory grouping notes. The AI should use this context when answering questions. Show which knowledge entries were used in generating a given answer.

Step 6: Set up under-performance alerts

Passive dashboards do not change behavior. Alerts push the signal to the right person before the end of the month.

Add an alerts system. Alerts should fire on two conditions: (1) a territory-category combination drops below a configurable threshold of target achievement (default 70%, editable by managers), and (2) a SKU that was in the top 10 by volume last month drops more than 20% in the current month versus the same period last month (week-over-week comparison). Each alert should show: the territory, the category or SKU, the current figure, the expected figure, and the gap. Alerts are visible in a dedicated Alerts tab. The assigned rep for a territory receives their own territory's alerts. Managers see alerts for all territories in their region. Add a "dismiss" action with a required note field - the rep or manager must explain why they are dismissing before the alert clears. Dismissed alerts stay in a log visible to managers.

The dismiss-with-note requirement is deliberate. It creates a paper trail and discourages reps from clicking alerts away without acting on them.

Step 7: Build role-based views

The same underlying data should surface differently for a rep checking their numbers at a store versus a manager reviewing the week in a Monday meeting.

Build two distinct home screens based on user role. Rep home screen: shows their territory name at the top, current month achievement as a large percentage figure with color coding (green above 90%, amber 70-90%, red below 70%), their top three under-performing SKU categories this month, and any open alerts for their territory. There should be a single prominent "Ask a question" button that opens the Ask tab. Keep this screen usable on a mobile browser - large tap targets, no horizontal scroll, no dense tables. Manager home screen: shows a regional summary table - one row per territory - with columns for territory name, rep name, current month value target, current month value actual, percentage achievement, and open alert count. Clicking a territory row opens a detailed view for that territory. Managers can also see a "Team alerts" panel showing all open alerts across their territories sorted by severity.

After this step, log in with a rep-level test account and a manager-level test account and verify that each sees only what they should. Data leakage across roles is a hard defect - catch it now.

Step 8: Test the AI answers for accuracy and grounding

Before you invite the team in, the AI ask layer needs structured testing. This is not optional - field reps will lose trust in the tool the first time an answer is wrong, and they will not come back.

Build a test set of at least 15 questions with known correct answers you can verify manually from your import files:

I want to run a structured test of the Ask layer. For each of the following questions, show me the answer the AI gives and the exact data rows it used to compute it. I will verify manually: (1) What is the total sellout value for the North region in [current month]? (2) Which territory has the lowest target achievement this month? (3) Which three SKU categories have declined most versus last month? (4) List all territories below 70% of target. (5) What is the best-performing brand by units sold this month?

Work through the list. Where an answer is wrong, identify whether the error is in the query logic (the AI pulled the wrong rows) or in the data (the import had a problem). Feed corrections as follow-up prompts specifying exactly what went wrong. Repeat until every answer in your test set matches the manual calculation.

Also test the grounding behavior:

Ask the AI: "What will our sellout be next quarter?" and "Which new product launch should we prioritize?" Confirm that for both questions it responds by saying it cannot answer from the available data rather than speculating.

If it speculates, that is a prompt correction to make before shipping.

Step 9: Ship and iterate

When the test set passes and the role separation is verified, deploy.

Deploy the app to production. Set up user accounts for the following roles: [list your actual reps and managers with their email addresses and role assignments]. Send each user an invite email with instructions to set their password. After first login, each user should land on their role-appropriate home screen.

On day one, do a 20-minute walkthrough with two or three reps and one manager. Watch where they hesitate. The most common friction points are:

• The Ask layer not understanding company-specific terms (fix by adding them to the knowledge base via the knowledge settings page)
• Alert thresholds set too low, generating noise (adjust the threshold via the alerts configuration)
• The trend chart defaulting to a time range that does not match how the team reviews performance (fix with a follow-up prompt to change the default date range)

Each of these is a follow-up prompt, not a rebuild. Creatr's iterative build model means you describe the correction and the change ships without touching code.

For ongoing data updates, the import module handles weekly or monthly file uploads. If your ERP can export a CSV on a schedule, the uploads can be done by whoever manages the data feed - it does not require a developer.

Recap

This build has nine steps and produces an application that: ingests and validates multi-table FMCG data, manages targets at territory-category-month grain, surfaces trend analytics with filter controls, answers plain-English questions grounded in the actual numbers (not generic FMCG knowledge), fires structured alerts before month-end, and separates the rep and manager experience at the view level.

The AI layer is load-bearing here, not decorative. Without it, this is a reporting dashboard that requires someone to know what question to ask before they open a chart. With it, a rep with three minutes between stops can type "am I going to hit target this month?" and get a direct answer with the numbers behind it.

The accuracy testing step in Step 8 is the one most builders skip. Do not skip it. An AI that is occasionally wrong is worse than no AI at all for a team that needs to trust their tools.

For related builds, see Build an admin dashboard if you need a more general data management interface, and Add an AI assistant to your product if you want to go deeper on the ask layer patterns used in Step 5.