RevLifter on the Inside Commerce podcast - Do targeted, personalised campaigns in the cart and checkout really boost sales and margin?
We joined James Gurd and Paul Rogers on the Inside Commerce podcast to dig into a question every retailer asks eventually: are on-site promotions actually working, or are they just quietly eating your margin?
Here are the highlights.
The problem with most promotions
Discounts are one of the most powerful levers in eCommerce. Pull it, and conversions go up.
But there are two costs that come with it.
- In the short term, you damage profit margin. Every discount you give away is money you don’t make.
- In the long term, you damage brand value. Give away enough discounts, and you stop being seen as a brand. You become a discounter.
Most of the damage comes down to one thing: targeting, or the lack of it.
Site-wide offers, shown to everyone regardless of whether they need one, are the biggest mistake we see. Add in clashing codes between different teams (acquisition running one promotion, retention running another, brand wanting none of it), and retailers end up giving away far more than they need to.
Three ages of promotions
We talked through how promotions have evolved:
- Age one: spreadsheets and instinct. Look at your CRM or CDP data and decide who should see an offer based on gut feel.
- Age two: rules-based. Someone gets an offer when they hit a trigger, a limit, or a threshold.
- Age three: data and AI, deciding dynamically who does and doesn’t get an offer.
We’re firmly in that third age now. And the shift matters, because humans are full of bias about who deserves a discount. Data isn’t.
How the intent modelling actually works
Here’s the mechanic, in plain terms.
We put a tag on a retailer’s website and track visitors anonymously. Every click, every page view, every scroll gives off a signal. Collect enough of those signals across thousands or millions of visits, and a model can work out which ones actually predict whether someone is likely to buy.
The results are sometimes surprising. The classic assumptions (search traffic converts better, social traffic converts worse) don’t always hold up for a specific retailer. The data tells you what’s actually true for your customers, not what’s true in general.
The important bit isn’t the score, it’s what you do with it.
“A lot of retailers ask us: " Okay, we’ve got a score, now what? So we built in actionability. Should this person get a discount, and if so, what discount should they receive?”
Most visitors, roughly 98 to 99%, won’t buy at all. Identify as many of them as possible and stop showing them offers, and you’ve already saved budget. Then identify the smaller group who would buy with the right nudge, and you can target them properly.
That’s where the incremental revenue comes from.
The SportsShoes result
SportsShoes came to us with a familiar problem: they were running offers, weren’t sure they were working, and knew they were costing money.
We took over their exit pop-ups and applied intent rules to stop showing offers to visitors who were already very likely to buy. The result: over 2,000 discounts suppressed and £25,000 saved, while keeping the conversion uplift the promotion itself delivers.
(Worth noting: the real results are even better now. Case studies are a snapshot.)
We also had to build rules to prevent offers from clashing with an existing health insurance partnership and test messaging by market. Softer copy performed better in France. Harder discount language worked fine in the US. Small differences, but a big impact.
Not every intervention needs to be a discount
High intent doesn’t always mean “show a discount.” Sometimes it means showing nothing at all, or showing something else entirely.
We’ve tested content, social proof, and reassurance messaging (e.g., accepted payment methods) instead of discounts. The gains are smaller, typically 1 to 2% rather than the 20%+ we see from well-targeted discounting, but for some retailers that’s exactly the right trade-off.
Geo and event targeting can work the same way. For Club L London, we built campaigns around specific events like the White Party in the Hamptons and spring break in Florida, timed to when people were actually planning around them.
How premium and luxury brands use this differently
Premium retailers tend to worry more about margins and brand risk, which makes them among the most interesting to work with.
- Radley wanted to pull back on discounting altogether, especially on low-margin product lines.
- Harrods wanted to nudge people toward buying experiences rather than products.
- Flow Kayaks needed help with cross-selling. Nobody buys two kayaks, but a Stretch & Save target pitched at the right level (not double the price of a kayak) can nudge someone into a paddle, a life jacket, and some gloves.
A word on gamification (and the spinning wheel)
Spinning wheels are lazy. There’s a discount behind the wheel either way. All the wheel does is add a few seconds of theatre in front of it.
Real gamification comes back to a genuine value exchange: the customer does something and gets something meaningful in return. Visualising progress toward a Stretch & Save target works. Prize draws instead of blanket discounts work. A spinning wheel that reveals a discount you weren’t even eligible for does not.
A spicier take: loyalty programmes
One from the podcast worth sharing directly: loyalty programmes, in their traditional form, might be overengineered.
“Most loyalty programmes end with the customer earning a discount anyway. So why not skip the cards, the admin, and the infrastructure, and just give the discount to the people who deserve it?”
It’s a fair challenge, especially for retailers without the scale or cross-channel reach to justify the overhead.
Measuring incrementality
We used to rely on holdout groups: show nothing to one segment, show the offer to another, and compare results.
We’re moving past that now. With enough data, the model can forecast what’s likely to happen for a given group of visitors, then measure how close that forecast was to reality afterwards. That’s a more accurate way to prove incrementality than a holdout test alone, and it gets sharper over time as the model learns.
Where this is heading
The old SaaS model hands retailers a platform and some instructions, then leaves them to figure it out on their own. We think that model is going away.
What comes next is a system that makes sensible, unbiased decisions on a retailer’s behalf: understanding what’s happening on the site, acting on it, and coming back to say “here’s what I did, here’s what else I could do.” Less clicking buttons, more getting things done.
Want to hear the full conversation? Catch the episode on the Inside Commerce podcast.
