How much data do we need for a forecast to make sense?

It depends on the task, but the general rule is simple: the more clean history, the more reliable the forecast. For seasonal demand, ideally you have at least several full cycles — otherwise the model won't see the seasonality, it'll guess. At the data audit we look at your specific case and tell you directly whether there's enough history, or whether you should accumulate more first.

What if our data is dirty or scattered?

That's a normal situation, and cleaning up the data is often the bulk of the work itself. Data lives in 1C, a CRM, several spreadsheets, kept differently across each, duplicated. We consolidate it into one source, clean and reconcile it, and only then build the forecast. Building a model on dirty data is pointless: it will produce a confident but wrong number.

How accurate is the forecast?

The honest answer: accuracy depends on your data, and we don't promise a "magic number." We validate the model against history — checking how accurately it would have "predicted" what already happened — and show the real accuracy and range, not one pretty figure. Where there's not enough history for a reliable forecast, we say so instead of masking the uncertainty.

Does it have to be AI, or can it be simpler?

Often proven statistical methods are enough, and we don't push AI just for the buzzword. Where the task is genuinely complex (many factors, nonlinear relationships), we apply AI models. Where plain statistics is sufficient, we use it, because it's cheaper and more transparent. We pick the tool for the task, not the other way around.

How much does AI analytics cost?

It depends above all on the state of your data: if the database is clean, most of the work is building the model and dashboard; if data needs consolidating and cleaning, that's where the bulk of the work is. So we give a range only after the data audit. Any figure before that would be a guess, and that's not how we work.

When is it too early for analytics?

When there's no accumulated clean data: records aren't kept, or are kept haphazardly, there's little history, sales and inquiries aren't properly logged anywhere. In that case a forecast is guessing with a pretty chart attached. We then recommend setting up record-keeping and data collection first — often through a CRM or automation — and coming back to analytics once the base has accumulated.

Service

AI analytics — forecasts on your own data

Demand forecasting, sales and inquiry analytics powered by AI are useful, but on one condition: you have clean, accumulated data. Straight talk: no database or dirty data means we clean it up first, and only then forecast. No "we'll predict everything." Stack: Node.js + OpenAI / Anthropic.

What we calculate

Three tasks where analytics gives an answer, not a guess

Demand forecasting
From your sales history, we calculate what and how much will sell — by season, day of week, promotions. Less "sold out at the peak" and less money locked up in slow-moving stock.
Good fit: retail and wholesale with years of sales history and stable record-keeping.When it's NOT worth it: no history, or record-keeping was inconsistent — the forecast will just be guessing.
Sales analytics
We pull sales, margin and customer data into a single picture: what's making money, what's dragging it down, what products sell together. Decisions based on numbers, not a hunch that "this seems to be working."
Good fit: businesses whose data is scattered across 1C, a CRM and spreadsheets, with no one seeing the full picture.When it's NOT worth it: records aren't kept at all — there's nothing to collect, build the base first.
Inquiry analytics
We break down the flow of requests and inquiries: what's asked most often, when the load peaks, where customers drop off. You see where to add people and what to automate first.
Good fit: service and support teams with a history of inquiries in a CRM or ticketing system.When it's NOT worth it: inquiries aren't logged anywhere — set up tracking first.

How it works

Data first, then the forecast — and honesty about accuracy

What's included

Data audit as the first step. Before promising anything, we look at what you've accumulated: how much history, how clean the records are, whether there are gaps. If there's not enough data, we say so right away, not after you've paid.
Cleanup, if needed. Data scattered across 1C, a CRM and spreadsheets, duplicated, kept inconsistently — we bring it into order: consolidate, clean, reconcile. Without this, a forecast makes no sense.
A forecast with honest accuracy. We show not one "magic number" but a range and how much you can trust the model on your data. Where there's not enough history for a reliable forecast, we say so.
Understandable dashboards. The result isn't an export for an analyst, but a picture for decisions: what to order, where the peak is, what drags down margin. In your business's terms, not statistical jargon.
Updates on fresh data. The forecast isn't one-off: we connect sources so the model keeps calculating on current sales and inquiries, not frozen data from six months ago.
Stack: Node.js + OpenAI / Anthropic, plus proven statistical models where they fit the task. No hype: where plain statistics is enough, we don't push AI for its own sake.

How we work

We start with data — and say honestly whether there's enough

Data audit
We look at what you have: how much sales or inquiry history, where it lives, how cleanly it was kept. This is the decisive step. Deliverable: an honest assessment — enough data for a forecast, too little and it needs to accumulate, or it needs cleanup first. If analytics is premature, we say so here.
Data cleanup
If the data is scattered or dirty, we consolidate it into one source, remove duplicates, align formats. Sometimes this step ends up being the bulk of the work, and that's normal: without clean data, a forecast is fiction. Deliverable: a single, organized database ready for analytics.
Model and validation against history
We build the model and validate it against past data: how accurately it would have "predicted" what already happened. This is an honest quality test before you act on it. Deliverable: a forecast with clear accuracy + defined limits on where to trust it and where not to.
Dashboard and updates
We deliver the result as a clear dashboard and connect it to fresh data updates. In the first period we compare the forecast against actual results and adjust. Deliverable: a working dashboard + auto-updates + an agreed way to track accuracy going forward.

FAQ

What businesses ask about AI analytics

Ask about AI

It depends on the task, but the general rule is simple: the more clean history, the more reliable the forecast. For seasonal demand, ideally you have at least several full cycles — otherwise the model won't see the seasonality, it'll guess. At the data audit we look at your specific case and tell you directly whether there's enough history, or whether you should accumulate more first.
That's a normal situation, and cleaning up the data is often the bulk of the work itself. Data lives in 1C, a CRM, several spreadsheets, kept differently across each, duplicated. We consolidate it into one source, clean and reconcile it, and only then build the forecast. Building a model on dirty data is pointless: it will produce a confident but wrong number.
The honest answer: accuracy depends on your data, and we don't promise a "magic number." We validate the model against history — checking how accurately it would have "predicted" what already happened — and show the real accuracy and range, not one pretty figure. Where there's not enough history for a reliable forecast, we say so instead of masking the uncertainty.
Often proven statistical methods are enough, and we don't push AI just for the buzzword. Where the task is genuinely complex (many factors, nonlinear relationships), we apply AI models. Where plain statistics is sufficient, we use it, because it's cheaper and more transparent. We pick the tool for the task, not the other way around.
It depends above all on the state of your data: if the database is clean, most of the work is building the model and dashboard; if data needs consolidating and cleaning, that's where the bulk of the work is. So we give a range only after the data audit. Any figure before that would be a guess, and that's not how we work.
When there's no accumulated clean data: records aren't kept, or are kept haphazardly, there's little history, sales and inquiries aren't properly logged anywhere. In that case a forecast is guessing with a pretty chart attached. We then recommend setting up record-keeping and data collection first — often through a CRM or automation — and coming back to analytics once the base has accumulated.

Discuss your project

Show us your data — we'll tell you if it's ready for a forecast

30 minutesOne-on-one online
Flexible formatVideo or phone call
Solution-focusedPractical answers

Learn more

AI analytics: why a forecast starts with data, not a model

AI analytics and forecasting means using models to estimate future demand and analyze sales and inquiries based on historical data. Apricode — a Ukrainian web development studio from Kharkiv, in the web since 2016, working worldwide — builds these solutions on Node.js with AI models from OpenAI and Anthropic, plus proven statistical methods. Let's say it upfront: this is the most demanding direction in our entire AI lineup in terms of readiness, because its result depends entirely on the quality of your data.

A forecast is only as good as the data

Any forecasting model learns from the past. If you have years of clean sales history, it will see seasonality, promotion effects, days of the week, and produce a useful forecast. If your history is only six months old or record-keeping was sloppy, the model will still output a number — but that number will be a guess dressed up as precision. That's why we always start with a data audit, not with promises. The honest first step of "let's see what you've accumulated" saves you from decisions based on a pretty but untrue chart.

Very often the first and main job isn't even the model — it's cleaning up the data. A typical business keeps data in 1C, a CRM and a few Excel sheets, kept differently across each and duplicated. Before forecasting anything, it needs to be consolidated into one source and cleaned. Sometimes at this stage it becomes clear that what's accumulated is still too little for a reliable forecast — and then the honest recommendation is "set up your data collection and come back in six months," not "let's calculate something anyway."AI document processing: building the database →

Where analytics really pays off

When the data exists and is clean, the payoff is concrete. Retailers and wholesalers use demand forecasting to avoid over-ordering and avoid running out of a fast-moving product at peak — that's direct money in working capital and warehouse space. Service companies use inquiry analytics to see when the load peaks and where to add staff. In every case, the goal is the same: decisions based on numbers instead of "it seems to work." And when you need to not just calculate but act on the result, that's where automation kicks in.Business process automation →

And one more honest point: not every task needs AI. Proven statistical methods are often enough, and they're cheaper and more transparent than a heavy model. We pick the tool for the task, not the other way around, and where AI offers no advantage over plain statistics, we say so directly. The easiest place to accumulate clean data is where it's born — in your CRM.CRM for business →

If you want to know whether your data is ready for a forecast, show us what you have, and we'll tell you honestly: build the model now, clean up the data first, or it's still too early for analytics. That's the same level-headed approach we bring to our entire AI-for-business direction.AI for business: an overview →

Pick a convenient time

Request a callback