Give Your AI a Job Description

Last year your company tried AI. Someone set up a chatbot, bought seats for a tool or ran a pilot with a consultant. There was a demo and a few excited messages. Six months later nobody can say what it changed, and the subscription renews quietly every month.

In 2025, MIT's NANDA initiative reported that about 95% of the enterprise generative AI pilots it studied delivered no measurable profit-and-loss impact. The causes the researchers and later surveys point to are mostly outside the model. The tools didn't fit the daily workflow, didn't learn from corrections, and the data they needed was scattered across systems. Put plainly, the AI was hired without a job.

Write these down before anything gets built. If you can't, that's the first problem to solve.

1. The task. One repeated piece of work, described the way the person doing it would describe it. "Read supplier invoices and enter them in Xero" is a task. "Use AI in finance" is a wish.
2. The owner. The person whose week gets easier, who checks the output and says when it's wrong.
3. The number. Hours per week, response time, error rate or money recovered, measured before you start.
4. The boundary. What the AI may do alone, what needs a person's approval, and what it must never touch.

Picture a sales team that spends ten hours a week copying enquiry details from email into the CRM. A workflow does it in seconds and flags the enquiries it's unsure about, and the owner reviews only those. After a month you compare hours. That's the whole pilot, and it either paid for itself or it didn't.

Once one job works, the next is easier, because the data connections and the habit of reviewing output already exist.

For a travel company heading to WTM London 2026, I built a private outreach agent with exactly this shape. Its job is to find and score prospects, draft emails in the owner's voice into Zoho Mail and handle follow-ups, and the owner approves what goes out. My Campaigner engine researches businesses and sends reviewed email batches in the same way.

• Fractional CTO Session at $150 for a 90-minute call and a written plan. We write the job description together and price it.
• n8n AI Automation from $150 for one workflow with an AI step, tested end to end. Up to three linked workflows with error alerts and a run log for $420.
• AI Knowledge Chatbot from $220, when the job is answering questions from your documents.

Bring one task you'd like off someone's plate. Book a free call and we'll write its job description before you spend anything on building.