How does SwiftERM predict what each customer will buy next?

SwiftERM predicts each customer's next purchase by learning that individual's changing preferences and buying intentions continuously, then deciding autonomously which products are becoming most relevant to them and when the chance to buy is strongest. It works per customer rather than per segment, and it keeps operating after the customer leaves your site. The approach suits retailers who already have repeat customers and want more revenue from them without building campaigns or fixed journeys.

The core mechanism: one customer, one market

The site's framing is that "every customer becomes a market of one." Instead of grouping shoppers into segments and applying rules, SwiftERM builds its picture of each person separately and updates it as their behaviour changes. The page describes customers as continuing to decide what to buy after they leave the website: "their preferences continue to evolve. Desire strengthens and fades, priorities change, and what someone ignored yesterday may be precisely what they want tomorrow."

That is the input side of the prediction. The output side is a per-customer judgement of two things:

• Which products are becoming most relevant to that individual right now.
• When the opportunity to buy is strongest for them.

So the prediction is not a static "customers who bought X also buy Y" recommendation. It is a continuously revised estimate of relevance and timing for one person.

Why it can act after the customer leaves

Conventional ecommerce, as the page describes it, "waits for customers to return, groups them into segments, or sends campaigns according to rules, schedules and predetermined journeys." SwiftERM's stated difference is that it reaches customers after they leave the site, using what it has learned about them, without anyone creating a campaign or journey first.

A concrete example of the gap this addresses: a customer views a product, doesn't buy, and their interest fades for a few weeks before returning stronger. A rule-based setup only reacts if someone wrote a rule for that pattern. SwiftERM's claim is that it can recognise that individual's renewed intent and the right moment, and turn it into a sale "without anyone having to operate the process."

What "autonomous" means here, precisely

The page draws a sharp line between automation and autonomy, and it matters for understanding the prediction:

The prediction quality therefore depends on the system's own learning rather than on how well you have configured segments. That is the trade-off: less setup work, but also less manual control over exactly what each customer is shown.

What you need before it can predict well

The page does not publish technical prerequisites, data requirements or integration details. What it does state is the commercial context: the aim is to "generate more revenue from customers they have already acquired, while reducing dependence on indiscriminate promotion and discounting." That points to businesses with an existing repeat-purchase base and enough customer history for individual learning to be meaningful. If you are evaluating it, ask specifically about data sources, integration with your ecommerce platform, and how quickly predictions become usable for a new customer with little history — none of these are covered on the page.

Trying it and checking the claims

The site offers a free 30-day trial, described as having no lock-in and zero ongoing management, with terms applying. To start, you submit first name, surname, company, business email and your ecommerce website address through the trial form. The page also invites you to "examine your ecommerce opportunity together" via the same contact route.

Two claims worth verifying during a trial rather than taking at face value:

• The "1,500%" figure appears as a heading on the page but the excerpt does not explain what it measures or over what period. Ask for the definition and baseline.
• The Trustpilot quote ("No product out there I could more highly recommend at this moment in time") is a single customer review, not a performance measure.

A practical way to test the prediction itself in 30 days: pick a cohort of repeat customers, record what SwiftERM predicts and when it acts, and compare against what those customers actually bought and against your existing rule-based campaigns. That tells you whether the per-customer timing adds sales your current setup misses — which is the specific opportunity the product claims to capture.