Locket Intelligence · In development
We’re developing Locket Intelligence to interpret customer conversations, behaviour and purchase history, understand how a person’s situation is changing, and determine how a business should respond.
What to recommend. What to change.
When to follow up. When to wait.
A more human commerce
Built for what comes next01A different starting point
A click becomes an interest. An abandoned basket becomes a reminder. Four months without a purchase becomes a discount.
But behaviour alone doesn’t explain the situation.
Has the customer’s goal changed? Has something interrupted it? Do they need help, a different product, or time?
Those distinctions determine whether the next interaction is useful and whether the business earns another purchase.
02Locket Intelligence · In development
Locket Intelligence is being developed
around three outputs.
Identify the customer’s goals, circumstances, barriers and intent as they change. Separate observed evidence from inference. Weigh conflicting signals and make missing information explicit.
Understand the personEvaluate possible responses against the customer’s situation, the business’s objectives, available products and operating constraints.
Recommend a product. Offer guidance. Adapt the experience. Ask a question. Follow up. Wait. Request human review.
Apply commercial judgementReturn a structured recommendation with supporting evidence, unresolved questions and conditions for execution. Give an application the information it needs to create an appropriate customer experience.
Give software a basis to act03Context changes the decision
Two customers bought running equipment.
Neither has purchased in four months.
Illustrative scenario · Planned model
Support the next relevant purchase while respecting what the customer needs now.
Respect contact permissions. Recommend only available products. Require brand approval before sending.
Customer ASame goal. Changed circumstances.
Input · Customer message
“I’m still planning to run the marathon, but shin splints have stopped my training.”
The goal remains. An injury has changed the immediate need.
Pause race-preparation promotions. Offer relevant recovery support. Ask before scheduling a return-to-training follow-up.
Draft email · Awaiting brand review
“Thanks for the update. We’ve paused your marathon product reminders. Would you prefer to get in touch when you’re ready, or would you like us to check in later?”
Customer BA new goal. A different direction.
Input · Customer message
“The marathon’s done. I’m switching to swimming.”
The original goal is complete. The customer has chosen a new activity.
Stop marathon follow-ups. If the brand can serve the new interest, offer relevant swimming products or guidance.
Draft email · Awaiting brand review
“Thanks for sharing what’s next. What kind of swimming are you getting into, and is there anything you’d like help choosing?”
Authored examples showing the intended input, decision and application response. These are not live Locket Intelligence outputs or measured customer results. People and images are illustrative.
Explore the interactive demo04The ambition
One-to-one marketing has always required someone to understand the customer and decide what to do next.
Doing that manually for millions of people is economically impractical.
Our aim is to make those decisions affordable enough to run for every customer: a specialist model that interprets the situation, selects a response and gives software the information it needs to act.
The unit of marketing becomes
the individual decision.
05Our research
Our planned training approach will specialise open-weight foundation models for customer understanding and commercial decisions.
Our behavioural framework will guide how training examples are annotated: the customer’s stated goal, circumstances, barriers and intent; the evidence behind an interpretation; and what remains uncertain. It will also define criteria for evaluating whether a response fits the situation.
We plan to post-train open-weight foundation models on permissioned, curated examples connecting customer evidence to interpretations and appropriate commercial responses. The aim is to specialise the model for the decisions that shape customer relationships.
Expert and brand-reviewed comparisons will help us teach the model which responses are appropriate and why. Corrections will be reviewed and curated for future training, rather than changing the model automatically after each interaction.
Before releasing model updates, we plan to compare interpretation quality, decision appropriateness, constraint compliance, latency and cost against general-purpose models. Separately, controlled evaluations in applications will assess customer and commercial outcomes.
The goal is a shared model that learns how to interpret customer situations and select appropriate responses across consumer businesses.
Training and evaluation will be deliberate, reviewed processes. Proposed capabilities will be tested before they become part of a released model.
Understand the situation.
Decide the response.
Create the experience.
06From intelligence to experience
Sophei is the first application of Locket’s decision architecture.
Today, it combines existing foundation models with that architecture to interpret customer conversations and commerce data, recommend a response, and prepare individual emails, storefronts and landing pages for the brand to review and approve.
Sophei is live with paying brands and an approved Shopify app.
As Locket Intelligence develops, Sophei will be its first application, turning structured decisions into customer experiences. It provides a commercial setting for reviewing interpretations, brand corrections and outcomes.
Change the situation in our interactive demonstration. Follow the proposed decision through to an illustrative email or storefront, then try a customer preference that calls for no message at all.
Try the interactive demoCurated fictional scenarios illustrating the planned model and a possible application experience. Locket Intelligence and its API remain in development.
07The model and API · In development
We’re developing Locket Intelligence as a shared model for customer understanding and commercial decisions, with access planned through an API.
Sophei will be its first application. Other software will be able to use the same intelligence to decide how to respond to a customer, then create and deliver the appropriate experience.
Current evidence and business constraints.
Interpret the situation and select a response.
A recommendation with evidence and conditions.
Review, create and deliver the customer experience.
The model will learn how to interpret situations and select responses. A separate context and memory layer will supply each business’s current customer history at the time of a decision. The shared model will work with that context while each business’s customer information stays separate.
08What we’re building
A shared foundation model in development for customer understanding and commercial decisions, with access planned through an API.
In development02Our first commercial application, bringing individual decisions and experiences to customer retention.
Live03Our planned dedicated computing environment for running Locket Intelligence.
Planned
Locket One · Planned
A dedicated computing environment for Locket Intelligence.Concept visual. Product in development.
09A more human commerce
What to show someone discovering a product. How to respond when they hesitate. What to change when their circumstances change. When to reconnect after a purchase.
Retention is our first application.
We’re building towards shared intelligence across the software that manages the customer relationship. Our planned API will give each application a basis for deciding how to respond to the individual.
Let’s build what comes next
We’re interested in speaking with consumer brands, research collaborators and software teams who want to build with customer intelligence.