Was It Just You That Had a Quiet Friday?

A quiet Friday raises two questions, and a till can answer only one of them.

The first is what happened, and your POS system knows that in remarkable detail: how many bills there were, what was on them, when each drink was rung through and where the evening started to sag.

The second is why, and above all whether it was you or the market. Was the whole street down because of the rain, the footy match or a rail strike? Or (and more seriously) did everyone else have a decent night while yours fell flat, which points somewhere closer to home: the menu change, the rota, the new place round the corner?

Those answers lead to very different Monday mornings. One says hold your nerve. The other says something needs fixing.

Most operators answer the second question with a few messages to friends who run places nearby, and a feeling. That isn't a criticism; it's all there is. Your till knows everything about your own Friday and nothing at all about anyone else's. So the most useful question you can ask about a bad night is the one no single system can answer.

Same score, different game

In a recent project building a mobile application, I worked with Opta's football statistics feeds. Before that, like most people who watch football, I'd seen Opta everywhere without giving it much thought: in the graphics during a live match, in the fixtures and results on the news, in the analysis on Monday Night Football. Working with the feeds, I could see the raw data underneath. What struck me wasn't the volume. It was that a corner, a card or a shot on target means the same thing in every league and every country the data covers.

That consistency is what lets football tell games apart. Take two teams that both beat Manchester United 3-0. In the first game the opening half is wide open, with chances at both ends worth more than four expected goals and three of them going in before the break; then United change formation and keep a clean sheet for the rest of the match. In the second, it's tight and goalless until the 80th minute, a red card changes everything, and all three goals arrive in the last ten minutes. Same score, completely different games. Expected goals measures how good a side's chances were rather than how many went in, and statistics like it are how analysts now explain that difference, instead of relying on instinct and the scoreline.

I put two and two together because I'd seen the pattern before. In construction I spent years working with COBie, a structured format for handing over information about a building, and watched shared definitions start to change who won work. Football has the same thing in a different form: statistics people trust because they can be compared. Hospitality has the gap without the answer. The closest comparison I've found for where Databiite could go is what Opta did for football. Not more data, but data that means the same thing wherever it came from. That's the part that excites me as an engineer. Storing a lot of data is cheap now. Agreeing what it means is the hard, interesting problem, and it's the one that decides whether anyone can trust the answers.

The hard part is agreeing what counts

An assist only means something because someone wrote down what counts as one. That's the work the first two essays in this series were about. The first argued that a venue's trading history should survive a change of POS provider. The second described the shared structure we're developing, so that a sale, a refund or a modifier can be described the same way whichever till recorded it.

It's harder than it sounds, and the obvious statistics are often the least comparable. Take covers, the number of guests served. It sounds like the most basic figure in hospitality, yet at least one major POS system doesn't record it at all, and among those that do it can mean different things: guests seated, guests on a bill, or whatever a member of staff typed in when the till asked, if it asked at all. Put those side by side and you'd be comparing different things under the same name. So any comparison worth trusting has to start from the figures we can define consistently, such as how many transactions there were, what they were worth and when they happened, and be candid about the ones we can't yet.

Expected takings

Football's expected goals suggests a hospitality equivalent. Call it expected takings: given everything we know, what should this Friday have taken? The gap between expected and actual is where the story is, and it splits into two parts.

The first is your own baseline. Is this Friday unusual for you, once the time of year, the day of the week, the trend over recent months and that menu change have been allowed for? This part needs no network at all. A venue with its history in Databiite already has the raw material for it, including the years brought across from a previous till.

The second is the market. Were similar venues nearby quiet as well? That needs other venues' trading, defined the same way, for the same night. It also needs protection. We won't produce any figure that draws on too few businesses, or too few locations, for any one of them to be picked out, and that rule applies whoever is asking.

Then there's the shape of the night. Two venues can each be down a fifth on a Friday for completely different reasons. One was steadily quieter all evening: people came in, had a drink and went home. The other traded normally until nine and then fell off a cliff when the trains stopped. Same score, different game. The takings alone can't tell you which Friday you had. Sales by the quarter-hour can, and alongside comparable figures from venues nearby, they could tell you whether the street had the same night.

Who else is asking

Operators aren't the only people who want to know whether it was the venue or the market.

Lenders are one group, and I should be clear that this is a working theory rather than something we've tested. When a hospitality business applies for finance, its bank statements show money arriving. Its trading history shows how that money was earned: how many transactions, how the week is shaped, how heavily it leans on one big night, and how far it swings between August and December. A venue choosing to share its own verified history with a lender is the nearer possibility. Setting it alongside similar venues in the same area is further off, and depends on the network growing.

Brands are another. Suppliers want to know where, when and how their products sell, and whether a product is genuinely selling more in each venue or simply stocked in more of them. Chris, my co-founder, spent more than a decade selling POS software and saw that transaction data already has buyers. What it rarely has is the venue in the room. Databiite was founded on the opposite principle: venues hold a great deal of valuable data, and the value of pooling it should be shared among them. Each venue chooses what it contributes, and a venue that holds more back takes a smaller share. To be plain about where this stands: until the network reaches the size we've set, no buyer can query it at all, so nothing has been sold and no venue has been paid. A good many questions about what should and shouldn't be on offer are still open.

Some of what this data could show belongs to everyone. One day I'd like to publish something like a pint index: the average price of a pint, region by region, tracked over time. It would go through the same privacy check as a paying customer's query. If our own marketing can't pass it, it doesn't get published.

Invited in, not watching

There's an important difference between Opta and us. Opta covers every match in the competitions it follows. We will only ever see the venues that choose to join, through the tills we've connected to, in the places we've reached. For a while that will be a partial picture, weighted towards certain kinds of venue and certain parts of the country. Early figures will describe venues on Databiite, not the market as a whole, and we intend to check what we see against published figures before we call anything an index.

There is a quicker way to gather this kind of data: from the top down, through the providers. It would arrive faster and in far greater volume. But the venues that created it wouldn't know, wouldn't have a say and wouldn't be paid. Building from the ground up is slower, and the picture stays patchy for longer. It's also the only way I can see of earning the trust of the people whose trading it is, and of paying them for it. If we get it right, the result is something no single till can see: transaction-level trading across many providers, including the independents so often missing from industry figures, described consistently and shared by choice.

Back to Friday

None of this exists yet in the form I've described. What exists are the foundations: integrations with a growing number of POS systems, history that carries across a switch, a draft of a shared structure for describing a sale, and a privacy rule built into the system rather than promised after the fact. The rest depends on venues joining.

The part I find most interesting is that it could work in both directions. The questions buyers pay to ask about the market could become questions operators get answered about themselves: how did my Friday compare with places like mine, near me?

So the next time you have a quiet Friday, try writing down what you'd need to know to tell whether it was you or the street. I suspect most of the list will be things your till can't tell you on its own. That's what we're building towards.