7. AI and Operations

"Operations" in this chapter has a wide meaning: everything your company does to create what it sells — a physical product, a service, a project, a case file — and to deliver it. It applies to a factory, to a professional firm, to a service company, to an online retailer with its own brand.

AI intervenes here in two very different ways, and a CEO must keep them apart. The first concerns what you produce: designing and creating new products and services, for new markets or more competitive than your rivals'. The second concerns how you produce: running processes with automation and with decisions taken at the speed of data instead of the speed of meetings. By the end you will have the outline of your first decision playbook.

Designing and creating new products

The cycle from idea to market gets shorter

The traditional cycle — idea, analysis, design, prototype, test, launch — takes months because every step depends on people collecting information, processing it and presenting it. With an AI Team each step compresses: an agent reads the market (reviews, customer requests, competitors' moves, trends) every day instead of every quarter; an agent generates and compares design variants; an agent simulates costs and margins; an agent prepares launch material. People decide, choose, judge.

The result is not only speed: it is the number of attempts. A company that could afford two new products a year can test ten, measure the response and invest in the two that work. It is the same rhythm principle from chapter 4, applied to the product.

More competitive products

A product's competitiveness rests on three things AI makes accessible to an SME as well. Customer knowledge: agents read everything customers say and do — reviews, support requests, returns, abandonments — and extract the recurring reasons for satisfaction and dissatisfaction; the next product is born there. Personalization: a product or service configurable for each customer, with a quote or proposal in hours instead of days, is more competitive than a fixed catalog. Continuous quality: agents that monitor production data, complaints and defects find the correlations nobody has time to look for.

New markets with the same structure

Agents lower the cost of entering a new market: a new language, a new channel, a new segment. A company that sold only domestically can handle requests, documentation and support in five languages without five people; one that served only large accounts can profitably serve small ones, because the cost of serving them has collapsed. The product can stay the same: what changes is who you can afford to serve.

Example. A packaging-machinery maker with forty employees received technical requests from abroad and handled them with a single export manager: one-week response times, many opportunities lost. An AI Team — one agent qualifying and translating requests, one preparing the technical proposal from catalogs and previous configurations, one following the correspondence — brought response times to one day and let the export manager cover three markets instead of one. In twelve months export sales went from 18% to 27% of revenue. (Case anonymized.)

Services are products too

If your company sells services — consulting, support, design, management — creating new products matters even more. A service that used to require hours of a professional can become a fixed-price service delivered largely by an AI Team under the professional's supervision: more customers, higher margin, a freer professional. Chapter 13 shows how this plays out by sector.

Running production processes

Decisions at the speed of data

In most companies operational decisions — what to produce, how much to order, when to do maintenance, how to allocate people — are taken from memory, in meetings, with data from weeks ago. Agents change the rhythm: they read data continuously, apply the rules you decided, execute what falls within thresholds and propose what exceeds them.

The model is called a decision playbook: a set of rules of the form "if stock falls below X and reorder lead time is Y, then order Z," or "if the machine shows these signals, schedule maintenance within N days," with numeric thresholds that you and your managers have decided. Inside the thresholds the agent acts; outside, it proposes and waits for a person. It is how a company takes a thousand decisions a day without a thousand meetings, and without losing control.

The typical areas

Planning and inventory. The agent reads sales, orders, stock and supplier lead times, and proposes or executes reorders and production plans. The typical result is less capital tied up and fewer stock-outs at the same time — two things that usually move in opposite directions. AI inventory optimization has been shown to cut overstock write-offs by 14% and stock-outs by 11%.

Maintenance. In companies with machinery, predictive maintenance — intervening before the failure, on the basis of signals — is the use case with the best-documented results: unplanned downtime down 30-50% and equipment life up 20-40%, with payback in 6-18 months. Yet only 27% of manufacturers use it actively. For an industrial SME it is almost always one of the first three agents.

Quality. Agents monitor process data, defects, complaints and returns, and find the recurring causes. They do not replace quality control: they give it eyes on everything instead of a sample.

Purchasing and suppliers. Agents compare offers, track deliveries, flag delays and anomalies, prepare orders. The purchasing manager negotiates; the agent does the rest. In a documented case, a $400 million manufacturer cut supply-chain exception resolution from 3.8 to 2.1 days and staff time on it by 44%.

Production administration. Documents, confirmations, delivery notes, reconciliations: the "gray" work that absorbs hours and that, across the EU, is the most common use of AI among companies that use it.

Data point. AI-based predictive maintenance cuts unplanned downtime by 30-50% and extends equipment life by 20-40%; reported returns are 300-500% with payback in 6-18 months. Unilever's plant in Brazil reported $2.3 million in annual savings and a 45% cut in maintenance cost, recovering a $1.2 million investment in under seven months. Still, only 27% of manufacturers use it actively. Source: IIoT World, AI Predictive Maintenance 2026, citing McKinsey and MaintainX.

What stays with people

In operations more than anywhere, the 70-80/20-30 rule must be applied strictly: agents take the repetitive decisions inside the thresholds; people take those that touch safety, key customers, investments, strategic suppliers. Every playbook has absolute prohibitions (safety, compliance, spending above cap) that no agent can override, and every agent has a stopping point beyond which it asks.

CoS for operations

What to take away

AI intervenes in operations in two ways: on what you produce (shorter cycle, more attempts, more competitive products, new markets with the same structure) and on how you produce (decisions at the speed of data, with a playbook and thresholds you decide). Predictive maintenance, inventory planning and administration are almost always the first agents. And the rule holds: inside the thresholds the agent acts, outside a person decides.

Your to-do list.

1. Ask your heads of production and purchasing to list the ten decisions they take most often and the rules — including the unwritten ones — they use to take them. That sheet is the draft of your first decision playbook.
2. Write the absolute prohibitions no agent may override: safety, compliance, spending above cap.
3. Pick one of maintenance, inventory or administration as the first operations agent, and its CoS from the table above.