In your operation
Sales are recorded. The next purchase is still a guess.
The history lives in the ERP, the POS and spreadsheets. How much to buy depends on someone’s memory and a workbook nobody else understands.
Predictive commercial intelligence
Anticipate your sales. Optimize your decisions.
Predictily turns the sales your company already records into demand and revenue forecasts, inventory alerts and purchase suggestions you can verify.
- No forced plans
- No ERP replacement
- You approve every decision
Gray cement 50 kg · Downtown branch
- Expected demand
- 1,432 u
- 80% range
- 1,342 – 1,522 u
- Historical error
- 18 %
- Recorded sales
- Forecast
- Uncertainty range
It connects to what your company already uses
- ERP
- POS
- Excel · CSV
- SQL
- API
Built with technology from
- AWSCloud and infrastructure
- CloudflareDNS, network and email
- MicrosoftSQL Server connectivity
- AnthropicAI agents with Claude and MCP
AWS, Cloudflare, Microsoft and Anthropic are trademarks of their respective owners. Predictily uses their technology; it is not a partner of these companies and is not endorsed by them.
In 25 seconds
See Predictily at work.
Forecast, inventory and decisions, the way your team sees them.
Illustrative view with fictitious data.
The starting point
The problem is not a lack of data. It is deciding without foresight.
Your company already records every sale. What is missing is turning that history into a clear signal of what comes next.
In your team
You find the stockout after the customer has left.
Products that were selling run out, while capital sits in products that do not move. Your team reacts to the problem instead of getting ahead of it.
Underneath it all
A forecast you cannot verify is an opinion.
A figure with no range, no error history and no visible assumptions is not enough to commit budget. You need to know how far to trust each number.
The platform
A forecasting layer on top of the systems you already use.
Three layers, one commitment: every figure traces back to your data and every recommendation is approved by your team.
- 01
Connect
On top of your system, not instead of it.
It takes in your sales, inventory and supplier lead times from Excel or CSV files, through an API, or by reading your database in read-only mode. A wizard suggests the column mapping and reports data quality. Re-importing never duplicates.
- 02
Predict
Several models compete on your own history.
Statistical and machine learning models are tested against your past sales and against a simple reference method. The one with the lowest error is used, with uncertainty ranges and the Ecuadorian calendar: holidays, paydays and statutory bonuses.
- 03
Decide
Prioritized recommendations your team approves.
Reorder point, suggested purchase quantity, customers likely to buy again and opportunities ranked by economic impact. Predictily suggests; people review, approve and execute, with a record of who did what.
What your team sees on screen
Demand forecast
Expected units by product, category, branch and company, by day, week or month.
Revenue forecast
Units times an explicit price assumption. Units and money are never mixed.
Inventory and purchase plan
Stockout risk, days of cover, reorder point and a plan that respects budget, capacity and suppliers.
Customers and repurchase
How likely each customer is to buy again, when, how much and which products.
Revenue radar
Opportunities by customer and product, with a follow-up agenda for your sales team.
Decision center
Recommendations with economic priority and an audited review, approve and execute flow.
Accuracy history
Yesterday’s forecast compared with today’s actual sales, product by product.
Scenario simulator
Change price, cost or budget with explicit assumptions and see the effect on revenue and margin.
Alerts
Warnings for stockouts, accuracy decline and data quality problems.
Why trust it
We know why you distrust forecasts.
You have seen tools that promise “95% accuracy” without saying how they measured it. You have seen dashboards that describe last month very well and say nothing about the next one.
That is why Predictily shows each model’s error on your own history before you approve it, and keeps measuring it against actual sales afterwards.
How every figure is calculated
- Every model is always compared against a simple reference method.
- The test uses only what was known on each date: no peeking at future data.
- 80 and 95% ranges calibrated with the actual errors from the test.
- It separates sales from demand when stock ran out.
- No generative AI model calculates the figures.
- Every forecast keeps its version, its date and its assumptions.
Evidence
Accuracy measured, not promised.
- Lower error than the reference method *
- Actual hit rate of the 80 and 95% ranges *
- Series evaluated *
- Period held out for the test *
* Result on our fictitious demonstration data (a simulated hardware store). It is not a customer result. With your data, the error is measured before the model is approved.
How we work
Five steps. You decide at each one.
There is no fixed plan to accept. There is a route with clear deliverables, and you approve the model by looking at its error, not at a brochure.
- 1
No commitment
First conversation
We learn how you operate: branches, products, systems and the decisions that cost you the most today.
- 2
Tailored
A proposal for your operation
Scope and terms based on what your company needs. No bundles that make you pay for what you will not use.
- 3
Your data
Connection and data quality
We connect the source in read-only mode and give you a quality report before any modeling.
- 4
Evidence
A test on your history
Models are evaluated against your past sales. You see each one’s error and approve the one that will be used.
- 5
Ongoing
Operation and measurement
Scheduled forecasts, alerts and recommendations. Accuracy is compared with actual sales, month after month.
What it is and is not
Six differences you can check for yourself.
We would rather tell you what Predictily does not do before you find out on your own.
It is: An independent forecasting layer.
It is not: Another ERP, POS or accounting system to roll out.
It is: Tested statistical and machine learning models.
It is not: Generative AI making up figures.
It is: Suggestions your team reviews and approves.
It is not: A system that places orders or writes to your ERP.
It is: Visible error history, product by product.
It is not: A brochure accuracy percentage.
It is: An uncertainty range on every figure.
It is not: A single number with no context.
It is: A proposal based on your operation.
It is not: A fixed plan you have to accept.
Who it is for
Built for companies that buy, stock and sell.
The engine is the same. The forecast is trained on your products, your branches and your calendar.
Distributors
Wholesale and retail with many products and repeat customers: demand by product and repurchase by customer.
Hardware stores
Wide catalogs with occasionally sold products: models designed for intermittent demand.
Auto and farm parts
Thousands of slow-moving items, where running out costs the sale and overstock ties up capital.
Multi-branch retail
Forecasts by branch and by category, with totals that always match the detail.
Importers
Long supplier lead times: reorder point and suggested quantity calculated with your delivery times.
Recurring consumption
Customers who come back: who is about to buy, who has stopped and which product to offer.
Your sector is not on the list?
If your company records sales by product and date, let’s talk. We will tell you frankly whether Predictily is useful for you today.
Where it runs
You choose where it runs and where your data lives.
The same product in three deployment options. The right one depends on your IT team and your data policies.
Self-hosted
On your own infrastructure
Predictily is installed with Docker on a server your company owns or in your own cloud. Your data never leaves your network and your team controls access, backups and updates.
VPS
On a server dedicated to your company
An instance just for you on a virtual private server, with automatic TLS certificates and backups. Recommended: 4 vCPU and 8 to 16 GB of RAM; no GPU required.
Shared server
On the multi-company platform
Your company on a shared instance, with your data isolated from everyone else’s at every layer. The fastest way to start, with no servers to manage.
In all three options the engine, the models and the security controls are the same.
Security and control
Your data belongs to your company. This is how we treat it.
Concrete controls you can review with your IT team before signing.
Isolation per company
Each company’s data is separated at every layer, with automated tests that verify it.
Roles and second factor
Role-based permissions (owner, admin, analyst, viewer) and two-step verification with an authenticator app.
Read-only connection
Predictily reads your sales; it does not modify your ERP, place orders or message your customers.
Encryption
Connection credentials and your customers’ names are encrypted with AES-256. Backups with rotation and restore tests.
Audit trail
A record of access and sensitive actions: who approved what, and when.
No lock-in on exit
Export all your data whenever you want. If you request deletion, it is erased after a 30-day grace period.
We design against OWASP ASVS level 2 and Ecuador’s Personal Data Protection Law (LOPDP) as references. We do not advertise certifications we do not hold: in the conversation we share the detail of each control.
Frequently asked questions
What a company asks before signing.
A predictive commercial intelligence platform. It takes the sales, inventory and supplier lead times your company already records and turns them into demand and revenue forecasts, inventory alerts and purchase suggestions you can verify.
No. Predictily is an independent layer: it reads your data and does not modify your systems. Your ERP, POS and accounting keep working as they do now.
Sales history by product and date is enough to forecast. Purchase recommendations also need inventory and supplier lead times. You can start with about 8 weeks of sales; more history enables longer horizons and more reliable forecasts.
There are three routes: Excel or CSV files, an API to send sales, or reading your database in read-only mode (with a local agent if the database must not be exposed to the internet). In the first conversation we review which one fits your system.
It depends on your data, which is why we do not promise a percentage. Before a model is used, we show you its error on your own history, compared with a simple reference method. After that, every forecast keeps being measured against actual sales.
Not to calculate. The figures come from statistical and machine learning models evaluated on your data. No language model generates the numbers.
No. It suggests quantities and priorities; people on your team review, approve and execute. Predictily does not place orders, write to your ERP or message your customers.
There are three options: self-hosted on your company’s infrastructure (with Docker), on a virtual private server (VPS) dedicated to your company, or on the shared multi-company platform with your data isolated. The product is the same in all three.
Yes. Predictily includes MCP servers so an agent (for example Claude) can load data and query the analyses with each person’s permissions. The agent queries and explains; the figures are still calculated by the statistical and machine learning models.
There are no forced plans. The proposal depends on your operation: branches, number of products and data sources. We start with a conversation and send you a tailored proposal.
Today the platform interface is in Spanish, in US dollars (USD), with Ecuador’s commercial calendar. The architecture is ready for other Latin American countries.
You can export all your data at any time. If you request deletion of your company, the data is permanently erased after a 30-day grace period.
Contact
Let’s talk.
No forced plans: we start with a conversation. Tell us how your company operates. A person from the team replies, not a bot, and we tell you frankly whether Predictily is useful for you today.
What happens next
- 1We read your message and reply to schedule a call.
- 2On the call we go over your systems, your data and the decisions you want to improve.
- 3If it makes sense for both sides, we send a proposal tailored to your operation.