Diagnose Before You Automate
AI can make the wrong process run faster, at greater scale and behind a more convincing interface. This six-part series re-centres AI discovery on root causes. It starts with a retailer that asked for a chatbot when the real requirement was to stop the calls, then covers fixing the information gap before adding retrieval, process mining before agent design, when a chatbot genuinely is the right answer, measuring avoided demand instead of conversations, and turning discovery evidence into an architecture decision record. It follows Bontebok Furniture Group, a fictional South African furniture retailer with stores and home delivery; its figures are illustrative. Research claims were checked on 5 November 2026.
6 articles
- Part 1They asked for a chatbot. The real requirement was to stop the calls.A furniture retailer asks for a chatbot because customers keep phoning about deliveries. Part 1 of Diagnose Before You Automate follows the discovery that found an information problem instead, and sets out a nine-step path with a written gate that says when not to automate yet.10 min read
- Part 2Fix the information gap before adding RAGRetrieval-augmented generation can only find information that exists, is current and has an owner. Part 2 of Diagnose Before You Automate sets out five kinds of information gap, shows why operational state is not a retrieval problem, and gives a short audit to run before anyone builds an index.8 min read
- Part 3Process mining before agent designBefore an agent is designed for a process, the process should be measured from its own event data. Part 3 of Diagnose Before You Automate shows how to build an event log from operational systems, what discovery and conformance reveal, and how each mined variant points to a rule, a workflow, an upstream fix or, occasionally, an agent.9 min read
- Part 4When a chatbot is the right answerA chatbot is a poor fix for missing information and a good interface for varied questions about information that is already right. Part 4 of Diagnose Before You Automate gives a six-question fit test, and shows a retailer's assistant that starts from the customer's reply to a delivery message.7 min read
- Part 5Measure avoided demand, not conversationsA chatbot dashboard that counts conversations can show success while customers contact you more than ever. Part 5 of Diagnose Before You Automate defines a demand measure that counts every contact in every channel, separates failure demand from value demand, and uses a comparison group to show what was actually avoided.8 min read
- Part 6From discovery evidence to architecture decisionDiscovery that ends in a slide deck is forgotten by the next budget cycle. Part 6 of Diagnose Before You Automate shows how to carry evidence into decision records that name the options rejected, the AI decision taken and deferred, the measure that proves it, and the conditions that would reopen it.8 min read
Working through any of this in your own organisation?
These articles come out of delivery work rather than theory. If the problems in this series look familiar, a short scoping conversation is usually the quickest way to work out what is worth doing first.
Talk to CloudNala