Describe your business before you automate it
Automation and AI tend to fail in the same way: neither knows what your business is made of. An afternoon spent describing it, in your own words, fixes that, and gives every tool you add afterwards something solid to work from.
In short
- Automations break and AI guesses when every system has its own idea of a customer or a machine.
- A model is a plain description of your business: the things you deal with, what you know about each, how they connect, the numbers you watch and what you are aiming for.
- Write it in your team’s own words, starting from the decisions you make every week.
- Once it exists, your screens, connections, monitoring and AI all work from the same description, and change when it does.
Why automation projects stall
Most automation starts with a tool: when a form comes in, add a row; when an order ships, send an email. Each one works, until the business changes or two systems disagree. The warehouse system calls it a SKU, the shop calls it a product, the accounts package calls it an item, and every automation in between has to translate, with nobody quite sure which version is right.
AI has the same problem in a new form. Ask a general-purpose assistant which machines need attention this week and it has to guess what counts as a machine, which readings belong to it and what “attention” means to you. It answers fluently either way, which makes the guesses harder to spot.
The fix is the same in both cases, and it isn’t another tool. Write down what your business is made of, once, and let everything else work from that.
What a model is, in plain English
A model (the technical word is ontology) is a description of your business as data. It answers five questions:
- What do you deal with? Customers, orders and invoices; or sites, lines, machines and work orders.
- What do you know about each? An order has a number, a date, a status and a total.
- How do they connect? An order is placed by a customer. A sensor is fitted to a machine, for a while.
- Which numbers do you watch? Revenue, minutes stopped, work orders still open.
- What are you aiming for? In your own words, including what matters more than what.
That’s all. No diagrams and no database design. If you can explain your business to a new starter, you can describe it.
A worked example
Here is part of the model for a site that moves goods along conveyor lines, written the way its manager would say it.
In the model, in plain English
A site has lines. A line is made of assets: drive motors, gearboxes, belts and sorters.
A sensor is fitted to an asset for a while. Its readings stay with the asset it was on, even after it moves.
A sensor reports vibration in mm/s and temperature in °C. Raise an alert if it goes quiet for 15 minutes, or vibration reads above 7.1.
An asset records its running hours since its last service. Hours to the next service is its service interval minus those.
A work order is raised for an asset. A stoppage is logged against a line, with how many minutes it lasted.
Minutes stopped is the total length of the period’s stoppages.
Six sentences. From them, Evalu8 Pro builds a list and a page for every line, asset and sensor, forms to add and edit them, live charts, alerts when a sensor goes quiet or reads out of range, the numbers on the dashboard, and an API for your other systems. Add a sentence and it appears everywhere at once.
How to write yours
Start from decisions, not data
List the handful of decisions your team makes every week: which job goes first, which machine to service, which customer to chase. For each, write down what you look at to make it. Those are your objects and their facts. Data that nobody decides anything with can wait.
Use the words your team already says
If the floor calls them drives, call them drives. The names in the model become the labels on every screen and the words the AI uses when it answers, so they should be the ones people say out loud.
Choose what each thing is known by
Every kind of thing needs one identifier: an order number, an asset tag, a device ID. It is how records from different systems turn out to be the same machine. If your systems don’t share one yet, agreeing it is the most valuable hour you will spend.
Write each number so anyone can check it
“Minutes stopped” should come with its working: which records, which of them count, and what to add up. If two people would work it out differently, say which way you mean. In Evalu8 Pro every number on the dashboard is a few plain steps that anyone can open and read.
Say what you are aiming for, and what mistakes cost
Aims are the part most teams skip, and the part AI leans on most. “Keep the outbound lines moving: no stoppage longer than 30 minutes, and every drive serviced before it fails. A missed fault costs us a morning; a false alarm costs an engineer an hour.” That last sentence tells anyone, person or machine, how to weigh a borderline call.
Expect to change it
You will get some of it wrong, and the business will change anyway. A good model is cheap to change: add a field and every screen, form and export picks it up. Evalu8 Pro checks every change before it is saved.
What changes once it exists
Connections stop being translation projects. Each system’s fields are mapped onto your words once, and records from different places arrive as the same customer or machine. One machine, five systems walks through it.
Monitoring knows what matters. You said what each sensor measures, how long it may go quiet and where its limits are, so an alert means something. From alert to approved fix covers the rest.
AI stops guessing. Evalu8 Pro’s assistant, Ask, gets its tools from your model: your objects, readings and numbers are what it can look up, and your aims are in its instructions. It answers in your words, names the records it used, and proposes actions for a person to approve rather than taking them.
Where to start
- Pick one area where a decision is made every week with poor information.
- Describe the things involved, and the one number that matters most.
- Bring in what you already have. A spreadsheet export is enough to begin.
- Write down what you are aiming for.
- Connect the live systems once the description holds up.
Evalu8 Pro has ready-made models for a trading business and a materials handling site, which you can adjust with simple forms. Or start with nothing and let each file you upload add the objects and fields it brings.