AI transformation leadership in industrial B2B.

By , Global Head of Digital Marketing, ABB.

An executive field guide to moving industrial B2B enterprises from AI pilots to measurable operating impact through practical models, governance and ROI.

The executive agenda behind the technology

AI transformation is a leadership job before it is a technology project. Three responsibilities belong in the same mandate.

Transformation leadership

Set the mandate, assign decision rights and build an operating model where governance and adoption live inside the work.

AI value and execution

Fund what has a baseline, an owner, a full cost and a clear reason to continue or stop.

Industrial commercial transformation

Connect customer operations, CRM, data and agentic workflows to how industrial businesses actually serve customers.

A public case: AI-enabled customer ordering at Holcim

In a McKinsey interview published on 6 February 2025, I described the ordering pilot in Spain: its adoption, where product suggestions went wrong and the option to reach a person. It shows why customer habits, workflow design and ownership matter as much as the model.

Evidence and scope

Spanish pilot with construction customers

As reported in the interview, for that pilot only. Not an audit or a promise for other contexts.

  • Digital ordering adoption reported moving from 25% to 93% in the pilot.
  • Around 66% of initial order proposals generated by the assistant were accepted.

Turn the argument into an executive mandate

A checklist for the AI mandate and the first 90 days: scope, decision rights, portfolio, baselines and adoption.

Compare notes

I welcome conversations with leaders working on the same problems, and invitations to speak, write or be interviewed.

Customer ordering

Can customers place and correct an order through a familiar channel, with a clear route to a person?

Bring customer experience, commercial ownership, integration and adoption under one plan. Judge expansion on what the pilot actually reported.

The published Holcim case covers the Spanish pilot and my role in it.

Quotation and proposal preparation

Does faster preparation shorten the complete customer response, or simply move the queue to an approval?

Agree who owns the finished quote, which commitments need approval and where the freed time goes. Measure quality, elapsed time and rework together.

A hypothetical example, worked through in the essay.

Account research and CRM decisions

Does the system help sellers pick better accounts and bring better evidence to the next conversation?

Set rules for source quality, freshness and permissions, and keep sellers accountable. Count qualified conversations and buyer responses, not just activity.

Questions to test locally; no pipeline or revenue gain is claimed.

Activity is easy to fund. Captured value is harder to own.

The portfolio fills with pilots because starting work is politically easier than naming the P&L line, baseline, owner, and legacy process that will change.

Can one business owner state the baseline, target, decision date, and what will be stopped if the system works?

Central standards and local accountability have to coexist.

A central AI team becomes a queue. A fully federated model creates duplicate vendors, incompatible controls, and no portfolio view.

Is there one named workflow owner with authority over adoption, process redesign and the outcome, while shared teams run the common platforms?

A policy is not proof that a workflow was controlled.

Teams classify risk in a spreadsheet, then lose the connection between sources, model behavior, human judgment, released output, and monitoring.

Can the organization reconstruct the evidence, approval, version, permissions, and incident path for one consequential AI-assisted decision?

Process and outcome

Define trigger, endpoint, customer, owner and performance gap.

Constraint and evidence

Investigate waiting, rework, exceptions and competing explanations.

Intervention alternatives

Compare process repair, integration, rules, predictive models and generative AI.

Delivery conditions

Check data access, evaluation, action authority, review, integration and adoption.

Economics and decision

Estimate net benefits and total costs, then choose the next experiment.

Sources

Questions executives ask

What does industrial B2B AI transformation actually mean?
It means changing the decisions and workflows that matter, then using AI where more intelligence moves a measurable result. That can cover commercial growth, CRM, customer experience, operations, data and agentic systems. The unit of change is the workflow, not the model.
Where do industrial AI initiatives usually stall?
When a pilot has no outcome owner, when process and data are fragmented, or when governance and adoption wait until after the technology choice. A good demo cannot fix any of those.
What operating model should an industrial enterprise use?
There is no universal org chart. A practical start: a central team owns shared standards, data services, risk controls and infrastructure, while workflow owners in the business own the outcome. Regulation, business-unit autonomy and scarce technical talent set the balance.
Is agentic AI ready for industrial B2B enterprises?
In narrow workflows with clear permissions, visible state and a defined exception path, yes. Account research, CRM hygiene, proposal support, operational briefings and anomaly triage are good candidates. Broad autonomy without those controls is a much weaker claim.
How should an industrial company measure AI ROI?
Start from a baseline the function owns. Then track adoption, cycle time, decision speed, rework and the financial or operating result that follows, alongside the full cost of implementation. A productivity estimate becomes value only when time, cost, capacity or revenue actually changes.
How does the EU AI Act change the work?
It depends on your organisation's role, the system, its intended use and its risk class. Keep an inventory of use cases, classify early, keep the evidence and check the Commission's current guidance and the legal text. Get qualified advice before consequential deployments.

Canonical URL: https://juanbeltran.ch/ai-for-complex-b2b-sales