The pattern
In almost every engagement we take on, the client arrives with the same framing: "We want to automate X." They have identified a tool. Sometimes they have already paid for a subscription. What they have not done is map the process they are trying to automate. This is where most AI projects fail. Not during implementation, and not because the technology is limited. They fail in the first conversation - when everyone nods along to a tool name without asking the prior question: what does this process actually look like today, step by step, exception by exception?
Process mapping is not a preamble. It is the work.
When we sit down with a client for a discovery session, we spend the first 40 minutes asking one question in many forms: "What actually happens?" Not what should happen. Not what the SOP says. What actually happens - including the exceptions, the workarounds, the steps only one person knows how to do. This is where you find the real complexity. A straightforward invoicing automation project reveals that 20% of invoices require a manual override because one client negotiated a custom billing structure four years ago. An inventory reorder project reveals that one supplier has a 10-week lead time and must be treated differently from all others. None of this is in the SOP. All of it will break your automation if you do not account for it.
"Rule of thumb: if you cannot draw the current process on a whiteboard with the person who does it every day, including every exception they can name, you are not ready to automate it."
The data readiness problem
You cannot automate a process that relies on data that does not exist in structured form, or that exists in three different places with three different formats. Data readiness is a prerequisite for automation, not a detail to sort out later. When we take on an engagement, we assess data quality in week one. If the data is not ready, the first deliverable is a data normalisation plan.
The constraint-first principle
Identify the constraint in the workflow - the step that limits the throughput of everything downstream. Automate at the constraint first. Automating anywhere else is optimisation. Automating the constraint is transformation.
The question worth asking
Before engaging any consultant, vendor, or tool, sit down with the people who actually do the work and ask: "If we could not change this process for the next six months, what would break first?" The answer is almost always the highest-priority automation candidate.