A connected machine should show its owner every value it sends home

I believe machine-tool makers have been asking the wrong question about the data our machines produce in customers' plants. We ask how much we can collect and what it might be worth to us. The better question is what a customer's production engineer needs to see before allowing one of our machines to send anything beyond the factory fence. My position, after a decade of building remote service at Isaragi, is that every value a connected machine sends should be visible to its owner on the machine itself; that part programmes and production records should stay in the plant unless the owner decides otherwise; and that a maker should earn its money from the downtime it prevents, not from the data.

What a production engineer is afraid of

When we launched remote service in April 2015, only 11% of the customers buying new machining centres agreed to connect them. Our service division had planned for 40%. I spent much of the following winter visiting customers who had declined, and their reasons were more specific than the general nervousness about security we had assumed.

A contract machining shop in Gunma supplied three competing automotive groups. Its owner worried that cycle times and utilisation, passing through our servers, might reach one of those groups through our own sales relationships with them. An aerospace parts supplier was forbidden by its customer from allowing any outside connection to machines that made flight hardware. A medical-device plant in Germany wanted to know whether operator identifiers counted as personal data. And several production engineers told me, with some embarrassment, that they simply did not know what the machine would send, and that agreeing to something they could not describe was not a decision they could defend to their own managers.

None of these concerns was about the quality of our encryption. They were about knowledge and control. The customer could not see what we collected, could not limit it and could not be sure how we would use it. We had designed the technology carefully and the relationship carelessly.

The list belongs on the machine's own screen

In 2016 we changed four things. Every value a machine sends home is now listed on its operator screen in plain language, with how often it is sent and why. Owners can switch categories off, and the machine shows which are off. Part programmes, workpiece measurements and production counts stay in the plant unless the customer chooses to share them. And we do not use utilisation data to plan sales calls without written consent.

The last rule was the hardest inside the company. Utilisation is the best signal we have of when a customer will need another machine, and the general manager of our service division argued, reasonably, that using it could help customers plan capacity before they were overloaded. I did not dispute the benefit. I disputed taking it without asking. The shop owner who feared his data would leak to his own customers would have been right to fear it reaching our salespeople.

The connection rate rose to 34% in the first full year under the new rules and to 57% by 2018. In the first half of this financial year it has been above 70% of new machines. Some customers have switched categories off; fewer than we feared, and their reasons have shown us which values we did not need. We removed two categories from the default list entirely, because almost every customer turned them off and our service engineers found they seldom used them.

For customers who could never connect outward, we built an on-premises gateway that keeps everything inside their network and lets a service engineer view it only during a visit or an agreed remote session. We offered it in 2018, three years later than we should have.

Charge for the visit that did not happen

The commercial argument against my position is that data is an asset, and a maker that declines to exploit it leaves money for others to take. I think that argument misreads what is valuable here. Machine-health data is worth most to a maker when it helps fix a machine faster and design the next one better, and both are worth a great deal. Since our connection rate rose, first-visit fix rates have climbed from 61% to 84%, because engineers arrive with the right parts, and average downtime per service call has fallen from 3.9 days to 1.6. Field data has changed the cooling circuit of a ball screw and the lubrication of our tool changers.

We charge for that value directly. Remote service is included for three years with a new machine, and afterwards customers can buy a contract priced on the response and diagnosis it provides. Paid contracts now cover more than 6,500 machines. Finance would have preferred to charge from the first day, and we gave up about two years of subscription revenue to build trust first. I would make the same choice again.

There are honest limits to this argument. A company selling scheduling and production-planning software across many brands of machine is selling the analysis of production data, and its terms will rightly differ from ours. Expectations vary by country, and our contracts have to follow each market's law on data, which our legal team now reviews before a launch rather than after a customer asks. And a published data list is only as good as its accuracy. We audit the values each machine actually sends against the list on its screen at every software release, and once found a mismatch our customers had not noticed. We told them before we fixed it.

I would ask a machine maker's board three things about its connected service. Can a customer see, on the machine, everything it sends? Can the customer stop any of it? And is the company earning from the service it provides, or from what it learns about the customer's business? If the honest answer to the third is the latter, the business is building on a trust it has never asked for, and its customers will eventually notice.