Dynamic pricing that runs (close to) real-time, not once

Here is our point of view: a pricing engine is only worth what it earns while it is left switched on. And engines get switched off, formally or informally, far more often because of trust than because of maths.

Vendors, boards and data teams tend to focus on the model: how clever it is, how much data it uses, how often it updates. Those things matter. But the businesses that get lasting value from dynamic pricing are rarely the ones with the most sophisticated algorithm. They are the ones where the commercial team believes the price, understands why it moved, and knows exactly what happens when something goes wrong.

How engines quietly die

Dynamic pricing seldom fails with a bang. It fades. A seller overrides a price because a customer pushed back, and nobody notices. A floor is ignored "just this once" for a big deal. A data feed stops updating and the engine keeps producing prices that look reasonable but aren't. Google's engineering guide Rules of Machine Learning warns about exactly this kind of silent failure. A few months later, prices are being set in spreadsheets again, and the engine is still running in the background, faithfully recommending prices nobody uses.

Illustrative example

Illustrative example. A travel business launches an engine that reprices hundreds of routes several times a day. The first month looks promising. Then, late one Friday, a competitor's website publishes a fare at a fraction of its usual level, almost certainly an error. The engine, set to follow the market, matches it across a whole region. By Saturday morning, social media has noticed, the commercial director has been phoned at home, and the engine has been switched off "until further notice".

Nobody disputes that the model worked as designed. The problem was everything around it: no limit on how fast prices could move, no line between what the engine could decide alone and what needed a person, no clear owner for the off switch. Six months later the engine is technically live, but every price change goes through a manual approval queue. The investment is still on the balance sheet. The benefit is not.

What you'd notice

• The engine is "live", yet a large share of prices are overridden, and nobody can say how many.
• Overrides are agreed by email or chat rather than recorded where the price lives.
• Sellers describe system prices as "what the computer says", not as prices they can explain to a customer.
• Nobody can say who is allowed to pause the engine, or how quickly.
• A data problem was discovered by a customer before it was discovered internally.
• Each new exception becomes the next precedent.
• The benefit case is quoted from the original business plan, not measured.

Why it matters

Dynamic pricing can be one of the most valuable tools a business owns, particularly where demand, capacity or cost move quickly. But the value only arrives if prices move when they should, stay within limits the business is comfortable with, and are trusted enough that people don't route around them.

The stakes are rising. Regulators are paying close attention to algorithmic pricing: the UK Competition and Markets Authority's paper Algorithms: How they can reduce competition and harm consumers covers personalised pricing and the risk of algorithms facilitating collusion. Consumer law on misleading prices, such as the EU's Unfair Commercial Practices Directive, applies however often the price changes. An engine that nobody fully controls is a commercial risk and, increasingly, a regulatory one.

Done well, this is worth a lot. Our team's work on a dynamic pricing engine added around £5M EBITDA a year. These figures come from roles our team held directly, not from client engagements.

Takeaway

Treat dynamic pricing as an organisational change that happens to involve a model, not a model that happens to need an organisation. The model sets the ceiling on what the engine could earn. Trust, control and clear ownership decide how much of that it actually earns, month after month. For those interested in the underlying science, Talluri and van Ryzin's The Theory and Practice of Revenue Management remains the standard reference.

Figures come from roles our team held directly, not from client engagements. Examples marked illustrative are hypothetical; third-party figures are from the sources linked.