Marketing Bandit Optimisation

Marketing bandit optimisation is the use of multi-armed bandits to continuously allocate marketing decisions across subject lines, timing, creative, offers, channels, and other campaign variables while balancing exploration and exploitation.

Key points

• Hightouch frames the marketer's dilemma as "stick with winners or test everything": a campaign with 5 subject lines, 4 send times, 3 offer types, and 2 creative templates already creates 120 possible combinations ^[src-025].
• Traditional A/B testing struggles with this combinatorial space because testing one or two variables at a time can take weeks or months, often after the campaign's peak moment has passed ^[src-025].
• Multi-armed bandits solve the allocation problem by shifting send volume dynamically as performance data arrives, making thousands of small adjustments rather than using fixed traffic splits ^[src-025].
• The send-time example moves from equal 20 percent allocation across five times, to early exploitation of a 2 pm winner, to refined allocation across 2 pm and 5 pm while still reserving some exploration ^[src-025].
• Common marketing uses include email campaign components, send-time optimisation, channel orchestration, offer strategy, creative/content testing, and combinations of dimensions such as content theme, offer type, send frequency, and channel preference ^[src-025].
• The limitation is that standard multi-armed bandits find the best option on average. For true 1:1 personalisation, Hightouch positions Contextual Bandits as the next layer because they combine bandit allocation with individual customer data ^[src-025].
• Hightouch's contextual-bandit article expands that limitation: a "winning" strategy can work for price-sensitive Marcus while alienating premium Sarah, so average optimisation must be augmented with customer context ^[src-026].
• Braze frames the same marketing pattern as real-time experimentation embedded in customer engagement platforms: multiple bandits can run in parallel, each tied to a campaign, channel, or optimisation goal ^[src-027].
• In Braze's use cases, arms can include creative, offers, channels, timing, CTA design, subject-line themes, notification tone, onboarding forms, or retention incentives ^[src-027].
• Braze's Intelligent Selection layer treats each campaign interaction as feedback that can update confidence across variations and shift exposure toward higher-performing options while the campaign is live ^[src-027].
• Statsig's parallel A/B testing article provides the fixed-allocation counterpart to this combinatorial problem: multiple campaign or product variables can be tested simultaneously when interaction effects are modeled instead of avoided by default ^[src-029].
• Braze's whitepaper gives the clearest combinatorial example: four plans, five creatives, five days, four send times, and three frequency options create 1,200 possible actions per customer before personal context is even considered ^[src-152].
• Braze's proposed architecture reduces that burden by assigning bandits to separate dimensions, so channel, offer, timing, and other choices can be learned as cooperating decisions instead of one huge action table ^[src-152].

Related entities

• Hightouch
• Braze

Related concepts

• Multi Armed Bandits
• Dynamic Traffic Allocation
• Exploration Exploitation Tradeoff
• AI Decisioning
• Agentic Marketing
• Contextual Bandits
• Customer Feature Matrix
• Intelligent Selection
• Parallel Ab Testing
• Treatment Interaction Effects
• Community Of Bandits

Source references

• ^[src-025] Hightouch — "Under the hood of AI Decisioning, part three: Multi-armed bandits"
• ^[src-026] Hightouch — "Under the hood of AI Decisioning, part four: Contextual bandits"
• ^[src-027] Team Braze — "What is a multi-armed bandit? Smarter experimentation for real-time marketing"
• ^[src-029] Allon Korem and Oryah Lancry-Dayan — "You can have it all: Parallel testing with A/B tests"
• ^[src-152] George Khachatryan, Nathaniel Rounds, Victor Kostyuk / Braze – "A community of bandits" (source page: "From multivariate testing to AI decisioning", 2025-09-30)