← Back to Blog

haggl.ai Blog

Upselling to AI Agents: Give a Saving, Get a Bigger Basket

|5 min read

Upselling to AI agents starts with an offer worth considering. Your product costs $100. You want to give a shopper a reason to choose your brand, but you do not want to sell that same product for $90 every time someone asks for a better deal.

There is another offer you can make: buy two for $180. Or buy the $100 product together with a complementary $40 product for $120 in total.

The shopper gets a saving. You get a larger order if they choose it. The condition is simple: the discount unlocks only when the basket gets bigger.

When a personal AI agent is comparing options for the shopper, that gives your merchant agent something useful to put on the table.

Give the agent a better purchase to consider

An upsell needs to make sense for the person buying. Two units might suit someone replenishing a product they use regularly. A complementary product might complete a task: the tool and the right accessory, or the coffee maker and a starter supply of coffee.

The agent needs a concrete offer it can compare with the original purchase. What is included? What does the whole basket cost? How much does the shopper save? Why would the extra item be useful?

A larger basket that exceeds the shopper’s budget or adds something they do not need is a weaker offer, even if the percentage saving looks impressive. Keep the original option available and explain the alternative in the context of the shopper’s request.

Two for $180

Two units of a $100 product would normally cost $200. The shopper saves $20 by buying two.

One unit remains $100.

Both for $120

A $100 product and a $40 complementary product would normally cost $140. The shopper saves $20 by buying both.

The saving requires both items.

For a brand avoiding standalone discounts, the baseline is doing nothing

Brands have different reasons for keeping their everyday prices intact: positioning, existing customer expectations, channel relationships or simply the economics of the product. A conditional bundle offers a different choice. The merchant gives a saving in exchange for a larger purchase.

For that brand, the practical question is whether offering bundles helps compared with leaving the ordinary offer unchanged. We examined that question using the same 234 bundle-eligible synthetic shoppers across 15 categories.

No Haggl versus bundles
MetricNo HagglWith bundles
Agentic recommendation rate25.9%31.5%
Recommended value per shopper$25.19$30.85

In this evaluation, bundles produced a 5.6 percentage-point increase in agentic recommendation rate and 22.5% more recommended value per shopper.

Recommendation rate means how often the agent recommends the store. Recommended value per shopper includes every shopper: the recommended basket at that store counts toward the average, and a recommendation for another store counts as zero.

The distinction matters. Average basket value among recommendations for the store was essentially unchanged, at $97.35 without Haggl and $97.92 with bundles. The observed gain came mainly from more recommendations, rather than larger baskets among those recommendations. These early results are a reason to test bundles; they do not yet establish a reliable improvement.

Make the conditions as clear as the saving

A useful bundle says exactly which products and quantities qualify. If the shopper removes the second item, the conditional saving no longer applies. A quantity deal should specify whether the price covers two units, additional units or a particular variant.

The merchant should also evaluate the complete basket. More revenue is useful only when the offer leaves enough contribution after product costs, fulfillment, payment costs and fees. Choose combinations that are relevant to shoppers and affordable for the business.

For an agent, these explicit terms make the offer easier to assess. For the shopper, they make the choice understandable. For the merchant, they define exactly what checkout needs to enforce.

Start with one offer worth choosing

Pick a product customers reasonably buy in multiples, or a pair of products they have a reason to use together. Set the saving and the minimum qualifying basket. Keep the ordinary purchase available.

Then compare that offer with your current approach. Does the agent recommend your store more often? Does recommended value per shopper improve? When the pilot reaches real shoppers, follow through to paid orders, order value and the economics after returns.

Upselling to an agent starts with a better reason for the shopper to buy more.

Explore bundles with Haggl or talk to us about a bundle pilot.

For more on offer design, read why the best offer is not always the lowest price and our guide to preparing your site for AI shopping agents.

Study notes

Source: Haggl bundle evaluation, 21 September 2026, run bundle-tradeoff-20260921-v2. Post-hoc comparison of 234 matched eligible shoppers with equal category weights across all 15 categories; 126 ineligible cases with discount fallbacks were excluded. These are synthetic recommendations, not purchases or revenue.

Descriptive 95% intervals from 4,000 paired bootstrap draws include no improvement: −0.2 to +11.4 percentage points for recommendation rate, and −5.5% to +60.2% for relative recommended value per shopper. Only 19 shoppers selected the exact bundle; other store recommendations selected ordinary baskets. The revised evaluation format also reduced the accepted-policy control, so these figures should not be compared directly with earlier full-policy Haggl benchmarks.

The examples illustrate offer mechanics, not measured results for those specific offers. Prices exclude taxes and shipping. Conditional bundle issuance remains subject to a scoped pilot with checkout enforcement and partial-return treatment validated before activation.