The sales you lose at night: how to measure them in twenty minutes instead of imagining them
It's 23:47. A customer — let's call her Sara — has the jacket in her basket and one doubt: are returns free? She finds a form promising an answer within 24-48 hours. She closes the tab.
It's the scene that opens every AI assistant pitch, including the one we'd give you. The problem is that none of them tells you how many Saras you actually have, and without that number you're buying a solution to a problem of unknown size.
TL;DR
- The number comes out of GA4 in twenty minutes, with two reports and no paid tools.
- Evening traffic is nearly always there; what varies a lot between shops is whether it converts worse than daytime traffic.
- If it converts the same, your problem isn't the hour: it's something else, and an assistant won't fix it.
- If it converts worse, you have a figure in euros, and you can decide.
How to measure it, step by step
1. Traffic by hour of day
In GA4, free-form exploration: dimension "Hour" (or "Hour of day"), metric "Sessions". Period: at least ninety days, because one week is noise.
You get a curve. Most e-commerce sites have two humps: one around lunchtime and one between 21:00 and midnight. The second is often the taller one, and it's the one where nobody is there to answer.
2. Conversion in those same hours
Same exploration, add the metric "Session conversion rate" (or purchases, if you prefer absolute numbers).
Here's where it gets interesting. Compare conversion between 21:00 and 24:00 with conversion between 10:00 and 18:00:
- Evening conversion equal or better. It happens, more often than people think: those buying in the evening have already decided. In that case a night assistant will give you less than you expect, and the money is better spent elsewhere.
- Evening conversion lower. You've found the loss. Step 3 turns it into euros.
3. The sum, with your own numbers
evening sessions × (daytime conversion − evening conversion) × average order value
It's a generous estimate, because it assumes recovering the whole gap, and nobody recovers the whole gap. Halve it and you'll have a figure to reason with. If that figure, over a year, doesn't pay for the tool, you've just saved money by reading an article.
What it can genuinely answer at 23:47
It's worth being precise here, because this is where pitches overreach. Night-time questions fall into two families, and an assistant handles them very differently.
Policy questions — it does these well. Returns, shipping, timing, warranty, payment methods, sizing and fit as described on the page. These are stable facts, written down somewhere, and an assistant connected to your documents retrieves and states them. They're the majority of doubts that stall a basket.
Availability questions — go carefully. "Do you have the M?" requires knowing the state of that variant, not of the product, and the answer depends on how often the assistant re-reads the catalogue and how far down it goes. I've written about both: on Shopify the problem is variants and how often it re-reads the catalogue.
The practical consequence: insist that on the last item the assistant says "it shows as available, I'll confirm in the morning" rather than "yes". The first sentence saves an order; the second cancels one and costs you a return.
The alternative nobody offers you
If step 2 reveals that evening conversion is low, before buying anything look at the three most frequent questions you get by email. If they're always the same three — and they usually are — writing them into the product page costs an afternoon and works at 23:47 too, with no subscription.
An assistant earns its place when the questions are many and varied, or when the conversation produces a sale the FAQ alone wouldn't have closed. Not when writing down the delivery times would have done.
FAQ
How many Saras am I really losing?
Do the three steps above. Anyone giving you that number without looking at your data is inventing it.
What if the assistant answers badly?
An assistant connected to your data gets it wrong less often, but what matters most is that it can say "I don't know" and hand over to a person. Better honest than brilliant and wrong — especially at night, when the person isn't there and the wrong sentence sits until morning.
How long does it take to switch on?
Connecting it, minutes. Making it useful — tone, rules on shipping and returns, when to hand over — an afternoon, plus a review after the first week of real conversations. Anyone saying "minutes" means the first part.
ShopBrain answers from your store's data and hands over to a person when needed. But do the three steps first: if the number that comes out is small, we'll be the first to tell you to spend that money elsewhere.
Related articles
- When the posting and the interview describe different rolesA hiring process produces four documents — the posting, the technical exercise, the final interview, the rejection email — and each says something different about the real role. How to read them before investing weeks.
- When the interview measures a conflict that isn’t yoursIt wasn't the first time I'd seen this pattern at a large, well-run company: whoever writes the posting wants one thing, whoever evaluates you needs another. How to spot it before you sign.
- The worst bug does not break the feature. It breaks the view26 tools connected and working in production, the dashboard showed zero. The bug had been there for months for another product too, and nobody had seen it because whoever could notice had admin permissions.