How to choose the first workflow to automate: the bottleneck economics test

Companies that fail with AI usually fail at the selection step. They pick a workflow because it is visible, or because a vendor demoed it, or because an executive wants a story. Then they discover the workflow was cheap to begin with, or too irregular to systematize, or so entangled with judgment that the system just adds a step.

The workflow you automate first should be chosen the way you would choose any investment: by cost today, by what changes, and by how sure you can be. Here is the test we run inside the Bottleneck Diagnostic.

1. Name the wound, not the technology

Start from a sentence a COO would say without prompting:

• “Three people spend most of Monday assembling the pipeline report.”
• “Quotes take four days because pricing approval lives in Slack.”
• “Every escalation starts with someone reconstructing the account’s history.”
• “When Maria is out, that process stops.”

If you cannot describe the workflow as a specific, recurring pain with people attached, it is not ready. Notice that none of those sentences mention AI.

2. Estimate the annual cost of the workaround

Be conservative and label your numbers. Roughly:

• Labor. People × hours per week × loaded hourly cost × 48.
• Delay. What a slow cycle costs: deals that cool, invoices that slip, customers who churn while waiting.
• Errors and rework. How often the manual process gets it wrong and what a wrong one costs.
• Key-person risk. What happens when the one person who understands it is unavailable.

You are not looking for precision. You are looking for the order of magnitude that tells you whether a system is even worth discussing. A workflow that costs a company a few thousand dollars a year does not justify a system. One that consumes a meaningful fraction of a senior person’s year does.

3. Check the structure

Cost alone is not enough. The workflow has to have a shape a system can hold.

• Routine majority. Most cases follow the same path; a minority need judgment. If every case is bespoke, you have an expertise problem, not a workflow problem.
• Named sources. The context the work needs lives in identifiable systems and documents, not only in someone’s head.
• Clear boundaries. You can say where the workflow starts, where it ends, and which decisions inside it are consequential.
• Existing tools. The actions at the end land in systems you already run.

A workflow that passes on cost but fails on structure is a candidate for redesign first, then automation.

4. Define success before you build

Agree the metric in advance, and agree how you will measure the baseline. Hours per week. Cycle time from request to resolution. Number of handoffs. Error rate. Time to reconstruct context for an escalation. Pick the one or two that map to the wound in step one.

If nobody can say what number should move, stop. A system without a success metric is a demo with a budget.

5. Decide honestly among six outcomes

The purpose of the exercise is not to force AI into the workflow. It is to decide which of these is true:

1. Automate. Routine, structured, well-sourced: build the system.
2. Integrate. The pain is data moving between tools; connect them and add rules, with little or no model judgment.
3. Redesign. The workflow is the problem; fix the process, then automate what remains.
4. Augment. Keep the person in the flow, but give them assembled context and prepared decisions.
5. Build. The real fix is a piece of software that does not exist yet.
6. Leave alone. Not expensive enough, or not structured enough, to justify a system this year.

Saying “leave alone” out loud is part of the value. It is why our diagnostic is framed as one economic decision rather than a sales call.

6. Prove one before you plan ten

Once a workflow passes, build enough to prove it on the real workflow with real permissions and the agreed metric. Measure against the baseline. Then, and only then, choose the second one. A single system with evidence is worth more than a roadmap of twenty without it, and it teaches you where the authority boundaries really belong.

That sequence, diagnostic, pilot, production, expansion, is how every Blocpod engagement runs. If you have a workflow in mind, bring us the bottleneck.