Insights
Getting real value from AI: start with the work, not the model
Most AI projects stall between pilot and production. Here's a practical way to choose use cases that actually pay off.
Many organisations have run an AI pilot. Far fewer have an AI system running in production, doing useful work every day. The gap is rarely the technology. It is usually the choice of problem, and what happens around the model.
Start with the work
The best AI use cases are found by looking at how work is done today, not by browsing model features. Look for work that is:
- High-volume and repetitive, such as reading, checking or re-keying documents
- Rules-heavy but judgement-light, where an expert’s time is spent on the routine 80%
- Slow because of information, where people wait on someone to find, summarise or compare
Contracts, invoices, remittances and compliance checks often fit all three.
Pilot
- Works once, on clean sample data
- No owner in the business
- Unclear how success is measured
Production
- Runs daily, on real data and volumes
- Business owner and people in the loop
- Measured by volume processed and capacity built
Design for production from day one
A demo only has to work once. A production system has to work every day, at scale, with real data. Before building, agree:
- What “good” looks like. Which measurable outcome will improve, and by how much?
- Where people stay in the loop. Which decisions does AI suggest, and which does a human approve?
- How it fits existing systems. Where does the data come from, and where do results go?
- How risk is managed. Covering data protection, auditability and what happens when the model is wrong.
- 01Systems of recordERP, contracts, inboxes
- 02AI readsExtracts structured data
- 03Rules decideDocumented, auditable policy
- 04People approveExceptions and judgement
- System / automated
- Control point
- People decision
Keep it simple
The most successful AI deployments we’ve seen are narrow, well-understood and measured. Start small, prove the value, then scale what works.
How to add your own article: create a new .md or .mdx file in src/content/articles/, copy the block between the --- lines at the top of this file, change the details, and write your article below it. See the README for how to add diagrams.
Frequently asked questions
Why do so many AI pilots fail to reach production?
Usually because of the problem chosen and what surrounds the model: no business owner, no agreed measure of success, and no plan for integration, controls or exceptions.
What does "AI reads, rules decide" mean?
AI extracts structured data from documents such as contracts, invoices and remittances, and transparent business rules plus human approvals make the decisions, so outcomes are consistent and auditable.
Published 5 November 2026 · Simplabs Limited