GBCORP BLOG · AI
Intelligent Automation: Where AI Meets the Workflow
Intelligent automation is a workflow engine with a model at the steps that need judgement. Rules handle the routine, the model handles the ambiguous, and a person handles the rest. Getting the split right is the whole job.
Intelligent automation joins two things GBCORP builds into its software: a workflow engine that moves work through steps with rules, and model steps that read, classify, extract, draft or decide where a rule cannot. The result is a process that runs itself for the routine cases, asks a person for the hard ones, and reports on both.
Rules, models and people
| Case | Handled by | Example |
|---|---|---|
| Routine, unambiguous | Rule | Order over a limit needs approval; invoice matches purchase order |
| Ambiguous, high volume | Model with review at first | Classify a message; extract fields from a scan |
| Judgement, consequence | Person, with the context prepared | Approve a refund; sign off a contract |
The workflow engine
Steps, conditions, assignments, deadlines and escalations, defined in the console rather than in code. A step can be a rule, a model call, a form for a person, or an integration with a system. Every item shows where it is, who has it and how long it has waited. The engine is part of GBCORP’s core, so a new process is configuration.
Review steps become measurements
Each model step starts with a person confirming its output. The confirmations are the measurement: accuracy per field per week. Where errors are harmless, the review is removed; where they are not, it stays and the person still saves most of the time because confirming is faster than doing. Cost per item is tracked beside accuracy.
Processes that pay first
Inbound documents into records. Enquiries classified and drafted. Orders checked against rules. Approvals routed and chased. Reports summarised. Onboarding checked for completeness. Each has volume, repetition and a queue a person works through every day, which is the signature of a process worth automating.
Checklist before automating a process
- Map the process as it runs today, with the exceptions.
- Measure it: items per week, minutes per item, error rate.
- Collect fifty to two hundred real examples with correct outcomes.
- Decide which steps are rules, which are models, which stay with people.
- Name who reviews at first and who owns the process afterwards.
- Pick the number that proves it: minutes per item, backlog, errors.
The numbers to expect
Scoping in ten days. First process running with review in four to eight weeks at a fixed price. Review removed on safe steps within the first quarter. A support plan covering the engine, the evaluation runs and cost reporting. Customised or more complex functionality is quoted separately before any work starts.
A worked example
An insurance broker receives claims documents by e-mail, message and post, and three people spend their days reading, checking and re-typing them. The automated process: documents arrive into a queue; a model step reads each into a draft claim with fields highlighted; a rule checks completeness and policy validity; a person confirms; complete claims post to the system and incomplete ones trigger a request to the client, drafted by the model. After a quarter, the confirm step is removed for standard claim types, the three people handle exceptions and clients, and the backlog that used to build up every Monday is gone.
Owning the process afterwards
The process definitions live in the console and belong to the business. The evaluation set, the most valuable thing the project produces, is the business’s. The cloud project is in its name. A support plan keeps the engine current, runs the evaluation on every change and reports cost and accuracy weekly. A new process is configuration by the business’s own operations lead, with GBCORP adding a model step where judgement is needed.
Pricing that matches the plan
The first process is priced as a fixed release with the review period attached, so the business knows what it pays and what it gets before day one. The support plan afterwards is a known monthly figure covering the engine, evaluation runs, model updates and cost reporting; each further process is a new fixed release. Model usage is passed through at cost and shown on the weekly report beside accuracy.
What the first quarter shows
Week one, items flow through the queue with a person confirming. Month one, accuracy per step settles and the first review steps are removed. Month two, minutes per item and backlog show the saving in the business’s own units. Month three, the second process is scoped from the queue the team now has time to see.
Review steps are not a weakness
A supplier who promises full automation on day one is selling the demo. Every model step in GBCORP’s automation starts with a person confirming, because the confirmations are how accuracy becomes a number. Where the consequence of a mistake is trivial, the review is removed after the numbers hold. Where it is not, the review stays and the person still saves most of the time, because confirming a draft is faster than producing one. The weekly report shows both the accuracy and the minutes, so the business decides on evidence.
Questions we hear first
Is this the same as robotic process automation?
Older automation clicks through screens; intelligent automation works through interfaces and adds model steps where judgement is needed, which is more robust and easier to change.
Do we need to replace our systems?
No. The workflow engine joins them through their interfaces.
What if the model is wrong?
The review step catches it at first, the case joins the evaluation set, and the weekly report shows the rate.
Who owns the automation?
You do: the process definitions, the evaluation set and the cloud project.
Where to start
The GBCORP AI Development and Automation product is the priced starting point; the GBCORP AI development hub sets out the method. For the ninety-day plan, read what an AI automation company should do in ninety days.
GBCORP operates across 17 countries on 4 continents through six pillars: AI & IT, Innovation, Sustainability, Infrastructure, Business Lounge, and EV's & Solutions.
