6 min read

Friday Fun : The Escapement Department

An engineer needed three hours to put an invoice number on a screen. The company turned it into a programme. By the time the programme had funding, fixing the problem would have undermined the business case for fixing it.
Friday Fun : The Escapement Department

A very small interruption

A mechanical watch spends most of its life holding something back. The mainspring, wound tight inside its barrel, exerts torque through the gear train. At the far end of that train sits the escape wheel, whose teeth are shaped to meet two small pallet stones on a lever. One stone holds a tooth while the balance wheel swings. As the balance returns, a jewel on its underside engages the fork at the end of that lever. The pallet releases the tooth. The escape wheel moves, gives the balance a little energy through the lever, and comes to rest against the other pallet. On the return swing, the exchange repeats in the other direction. The tick you hear includes the sound of the next tooth being caught.

There is an elegant bargain at that point of contact. The balance spends a little energy unlocking the mechanism and receives enough in return to make up for losses. For most of its swing, it moves clear of the fork. The part that holds the gear train back also helps sustain the motion that will release it again. Make the lock too shallow and the next tooth may fail to catch properly. Give it too much lock and the balance has more work to do each time it comes round. The useful pause depends on what happens at both ends of it. The watchmaker who adjusts the pallets has to observe the movement running. A pallet can look immaculate while the watch gradually loses the afternoon.

The first tooth

At 09.12 on a Monday, an engineer at a fictional wholesaler proposes putting an invoice number on an error screen that already shows the customer's account and order. Failed invoice downloads currently present a blank box. Customers ring support, where the script tells agents to ask for the invoice number. When the caller explains why that might be difficult, an experienced agent asks for the order number and finds the invoice in the back office system. The engineer has a small patch and one question for security. Can the screen show the number without widening access to invoice data? The order number, at least, has already passed its audition.

Security quite reasonably asks for the access check to be documented. The platform team owns the service that performs it, but its next planning round is three weeks away. Product asks whether there is any way to bring the issue forward. There is: work associated with an active incident can interrupt the plan. Support has already logged eleven calls, none of which individually qualifies as an incident. The engineer asks whether they can count the calls together. They can, once support opens a parent ticket, which requires a customer impact statement. Support asks whether anyone knows how many customers are affected. The engineer suggests putting an invoice number on the error screen.

By Thursday, the patch has acquired a pre-read. The pre-read needs a cost estimate, and an estimate of three hours prompts a question about why the item merits the committee's time. The team adds the cost of support calls. The new figure passes the threshold at which Finance must attend. Finance asks when the work will pay for itself. On the team's figures, the patch would pay for itself in a week. Finance is next available in two.

Escalation driven design

The first escalation works. A director sees the support figures and secures half an hour with the platform engineer who owns the access check. Escalation makes sense when the people doing the work need somebody to settle a genuine conflict over shared capacity. Here, it also teaches everybody watching that a request in the platform queue waits three weeks and a request with a director attached waits forty minutes. By the following month, several teams have discovered that almost every technical dependency can be made to express a strategic concern. The platform team now has two queues, one for escalations and one for requests whose sponsors are still on holiday.

This is escalation driven design. In its mature form, the route by which a problem can obtain authority determines the shape of the thing the company builds. At team level the invoice issue buys a three hour patch, if the team can borrow a security reviewer. At director level it buys a cross-functional review of invoice reliability. At executive level it buys a reusable document access capability across invoices, credit notes and delivery receipts. Each version is easier to fund at its respective altitude. Architects are asked to preserve the flexibility to handle document types that the company does not yet issue. The blank box has acquired a five year horizon.

When the proposed shared service comes back for estimation, it needs both capacity and an owner. Nobody wishes to own a service whose customers can now include every business unit, so ownership goes upstairs. Upstairs quite sensibly asks the business units to agree on what the service should do. Two of them had never needed anything from it and discover, during consultation, that they have requirements. A third offers its existing service, provided everybody adopts its account model. The invoice number can still fit on the screen. The meeting can no longer fit in an hour.

The load on the fork

In the watch, a small lever can hold back the whole train because its contact with the balance is brief and predictable. At the wholesaler, the equivalent of the fork hosts the call, invites the approvers, writes the minutes and carries the decisions back down. Three weeks later, another team asks the same platform engineer to review a routine permissions change. She declines the slot. She is preparing a board briefing on why routine permissions reviews take three weeks. The team leads with that fact in its escalation, which secures a place on the briefing agenda and costs the engineer the rest of Thursday. By Friday, the board has two examples of the problem and the platform team has two reviews it could have finished.

The transformation office times decisions from the moment a complete request enters governance. Teams respond by keeping requests in draft until they have secured the approval necessary to submit them. This cuts recorded approval time by half. Meanwhile the architects spend their mornings in early alignment calls about requests that, by the company's definition, have yet to arrive. The new workflow tool helps. It can circulate the same briefing to all approvers in seconds and issue reminders in parallel. For the first time, everybody can hold up the same decision simultaneously.

Support, meanwhile, has built a useful little routine. The agents who know to ask for the order number teach the others. Soon they can find the invoice and guide the customer through a fresh download before the script has finished asking them to verify the original fault. Call duration falls. The business case for fixing the screen becomes harder to make just as the work becomes more expensive. This is the only part of the company that has managed to reduce the cost of the defect, and its reward is to remain responsible for it. The board sees the improvement in average handling time. A customer sees a blank box and starts looking for the number to ring.

The grand complication

In watchmaking, a complication adds a function beyond ordinary timekeeping. A date display has to advance with the calendar. A chronograph has to start and stop on demand. Each puts demands on the movement, and the extra parts must earn their place by doing something the wearer can use. The wholesaler's new complication is a committee empowered to determine whether a change is small enough to proceed without a committee. Its remit includes pruning the list of bodies that can approve a change. All the bodies on the list have seats.

The committee's first practical exercise is the invoice number. The engineer demos the patch, including a request from an account that cannot see the invoice. The box stays blank. Security agrees that the existing access rules hold. “Can we put it on the screen this week?” the engineer asks. Finance agrees that the patch would save money. The difficulty is that those savings already appear in the platform’s business case. If the patch goes out on Thursday, the programme will lose its first measurable benefit before it has held its kick-off.

The chair asks whether the patch could go out as an early release of the platform. The architect checks the diagram. The patch uses the existing service, which the programme exists to replace. Finance could take the savings out of the business case, but that would send the case back for approval. The committee commissions a short paper on whether releasing the invoice number would prejudice the ability to release invoice numbers at scale. The engineer attaches her original patch. It has outlived two quarterly plans and now needs to be rebased.

The warranty

A watchmaker can put a movement on a timing machine and listen to the beats. The spacing of those beats helps reveal how the watch is running. At the wholesaler's year end, someone puts up a slide showing the brisk, regular pace of decisions. There is an approved architecture, a named sponsor, a funded discovery phase and an agreed route for changes small enough to proceed without review. The team receives a commendation for making a complicated problem legible to the business. Support receives a revised script.

At 09.12 the following Monday, a customer calls because the invoice page has failed again. The agent asks for the order number and brings up the invoice on her own screen. As she reaches the end of the script, the quality prompt appears. The call cannot count as fully resolved until she has encouraged the customer to use self-service next time. On her screen, the invoice number sits beside the order number. On his, there is enough room for both.

Technical reference: American Watchmakers-Clockmakers Institute, Horological Times, September 2020.