Mauzo ya tawi la CBD wiki iliyopita yalikuwa kiasi gani?
Talk to
your systems. Upesi AI — conversational business intelligence
upesi (Swahili) — quick, fast
Upesi AI turns your business database into a colleague you can simply ask — in English or Swahili. Answers, forecasts and early warnings in seconds, while your data stays read-only, masked and audited.
- Desktop & phone
- Kiswahili & English
- Read-only by design
What will the chain sell over the next 8 weeks?
Is anything unusual happening at our branches?
Which business customers owe us the most?
Why did Kasarani's sales drop in March?
Who are our best customers — and who stopped buying?
~8 stypical answer to an everyday business question
~30 sfor a full “why did this happen?” investigation
100%of everyday test questions answered correctly (21/21)
0write access — the AI can read, never change, your data
The product
One conversation. Every answer your business needs.
A desktop app and a phone app share the same clean interface: ask, watch it work, and open the data behind every answer. Shown here with our demo supermarket chain.
Real screenshots from the app running on demo data (a fictional Nairobi supermarket chain).
How it works
From question to decision in three steps
-
01
Connect
We connect Upesi AI to your system with a read-only account and describe your data in business words — including Swahili terms your team uses.
-
02
Ask
Type a question on your desktop, your phone, or inside an AI assistant. Watch live as it looks up tables, runs checks and builds the answer.
-
03
Act
Get a clear answer with the numbers behind it, charts on the dashboard, and one-click Excel, PDF or Word reports to share.
Features
Everything from quick lookups to deep investigations
Start with plain questions. Grow into forecasting, anomaly detection and customer intelligence — all in the same conversation.
Ask in plain language
No SQL, no report builders. “How much did CBD sell last week?” is enough. Follow-up questions keep the context.
The data behind every answer
Open the exact tables an answer came from, with the query shown. Numbers you can check, not just trust.
Excel, PDF & Word exports
Turn any answer into a report in one click — up to 100,000 rows in Excel, formatted tables in PDF and Word.
Live dashboard
Forecast, unusual activity, debtors and customer segments on one screen, per branch or for the whole business.
Live progress
See each step as it happens — “Looking at sales… Forecasting… Checking stock” — and stop any time.
Desktop, phone & AI assistants
A desktop app for the office, a phone app for the road, and a connection for the AI assistants your team already uses.
Advanced analysis
Built-in analytics that a data team would charge for
Computed by our own models on your data — not guessed by the AI. Each result says how reliable it is.
Sales forecasting
Up to 12 weeks ahead with a likely range, knowing weekdays, payday (25th–3rd), Kenyan public holidays and Eid. Always tested against a simple baseline and says plainly when it doesn't win.
Demo chain: 3.1% error vs 4.8% baseline
Anomaly detection
Spots failed M-Pesa payment spikes, unusual voids, stock-outs and revenue drops over time — and compares cashiers with colleagues and branches with the chain, flagging only what isn't chance.
Found all 7 planted signals in two years of demo data
Customer segments
Recency, frequency and spend scored into Champions, Loyal, New, At risk, Fading and Lapsed — loyalty and business customers apart, with names masked.
Warns when a top buyer isn't paying
Debtors ageing
Who owes what: not due, 1–30, 31–60, 61–90 and 90+ days, with accounts near their limit, nothing paid in 60+ days, habitual late payers and write-offs.
Chase the right account first
“Why?” investigations
Ask why sales fell and Upesi AI works through a checklist — stock, staff, payments, prices, orders — before it concludes. It never names a person as a suspect without comparing them with colleagues and other branches first.
Why Upesi AI
Less time digging. More time deciding.
Upesi means “quickly”. Questions that used to wait for a spreadsheet, an analyst or the end of the month are answered while you're still thinking about them.
Save time
Everyday questions answered in about 8 seconds; full “why did this happen?” investigations in about 30 — work that typically takes hours of pulling and joining reports.
Save cost
No extra analyst headcount and no BI licences per seat — the whole management team gets answers from day one.
Save effort
No queries, pivot tables or report requests. Anyone who can type a question — in English or Swahili — can use it on day one.
Catch problems sooner
Payment outages, unusual voids, stock-outs, falling sales and overdue debtors surface on their own — before they show up in the month-end numbers.
Estimate your savings
Move the sliders to match your team today.
Hours saved per month—
Staff cost saved per month—
Estimate: assumes about 1 minute per question with Upesi AI (asking and reading) and 4.33 weeks per month. Your results will vary.
Languages & interoperability
Speaks your language. Plugs into your systems.
Ask in English or Kiswahili — Upesi AI answers in the language you used, understands the Swahili terms your team uses for products and branches, and labels exports to match.
Ni wateja gani wa biashara wanaodaiwa zaidi?
Hoteli moja kuu inadaiwa KES 756,000 zilizopitwa na muda — imefika kikomo chake na haijalipa tangu tarehe 16 Mei. Ifuatilie kwanza.
Works with what you already run
- Any SQL databaseread-only connection
- Desktop appfor the office
- Phoneinstallable web app
- AI assistantsvia the open MCP standard
- REST APIwith live streaming
- Excel · PDF · Wordreports & data
- M-Pesa datapayments analysed
- Kenyan calendarholidays, payday, Eid
Running an ERP or a custom system? Talk to us — the guarded tool layer is built to be extended.
Security
Secure at every stage — from your screen to your database
The AI never touches your database. It can only call a few guarded tools, and every one of them is checked, masked and logged.
-
1 · Your device
- Saved sign-in encrypted by the operating system
- Access tokens kept in memory only
- Encrypted connections (HTTPS) to any server
-
2 · Sign-in
- Industry-standard password hashing
- Short-lived, automatically rotated sessions
- Protection against password guessing, plus role-based access
-
3 · Query guard
- Read-only queries on approved data only
- Every query checked and rebuilt before it runs
- Size and time limits on every request
-
4 · Database
- Read-only access, verified before the service starts
- A second read-only safeguard on every connection
- App data kept apart from your business data
-
5 · Privacy masking
- Names & phone numbers become tokens like
Person#4F2A - The AI model never sees who your customers are
- Exports pass the same guard and masking
- Names & phone numbers become tokens like
-
6 · Audit trail
- Every tool call, export and dashboard view logged
- Who, what, when — and the exact query
- Records outlive deleted conversations
FAQ
Questions managers ask us
Can Upesi AI change or delete my data?
No. It connects with a read-only database account and refuses to start if that account can write. Every query is checked to be read-only and limited to approved data before it runs.
Does the AI see my customers' names and phone numbers?
No. Personal details are replaced with stable tokens such as Person#4F2A before anything reaches the AI model, so it can still count and compare customers without knowing who they are.
Can I ask questions in Swahili?
Ndiyo! Ask in English or Kiswahili and Upesi AI answers in the language you asked in.
Which databases does it work with?
Any SQL-based database. Upesi AI connects with a read-only account and works on the data you already have — no migration needed.
Where can I use it?
On the desktop app, on your phone as an installable web app, and inside the AI assistants your team already uses.
Are the forecasts reliable?
Every forecast comes with a likely range and an accuracy check against a simple baseline over recent weeks — and it tells you plainly when it doesn't beat the baseline, or when a branch is too new to forecast well.
Get in touch
Ready to talk to your systems?
Book a short demo on your own kind of data. We'll show you what your team could ask on day one.
Based inNairobi, Kenya