Ask Duply

A system that does the job the training describes.

Ask Duply is the flagship interface of Duply, an execution operating system I designed for organisations whose revenue depends on thousands of independent field sellers. It combines a pipeline CRM, an AI coach, a content engine and a signal-detection layer into one system for daily field activity.

The industry it serves spends heavily on training, events and libraries of resources, and still watches most new sellers stall in their first weeks. Ask Duply's premise is that the missing layer isn't more knowledge. It's a system that sits inside the seller's actual workflow and turns knowledge into the next specific action.

Access to knowledge was never the constraint.

Field organisations have a structural execution gap. The company produces training, scripts and compliance guidelines; the field improvises anyway. Three failure modes repeat across the sector:

• New sellers stall because generic training doesn't tell them what to do today, with their contacts, in their words.
• Follow-up dies quietly. Interested prospects go cold not because of objections, but because nobody was prompted to act at the right moment.
• The company has no line of sight into field conversations, so messaging drifts and compliance risk accumulates where it can least be seen.

Generic AI chatbots don't solve this. A blank input box transfers the hardest problem, knowing what to ask, onto the least experienced person in the system.

Every layer, strategy to production.

I founded the product and carried it across the full lifecycle: problem definition and market positioning, system and product architecture, AI behaviour design, full-stack engineering, deployment, user onboarding and post-launch optimisation. Thirteen years operating inside field organisations supplied the domain model; the last stretch of the journey was making software behave the way great field leaders do.

Start with the behaviour, not the model.

The design principle throughout: the AI should steer, not wait to be asked. Every feature begins with a behavioural question: what should this person do next, and what stops them doing it? Then it works backwards to the technology.

• Diagnose before building. The product grew out of a diagnostic methodology (the Field Activation Audit) that maps where execution breaks down between company intent and field behaviour.
• Meet the work where it happens. Field conversations happen in social platforms and messaging apps, so the system extends into the browser rather than demanding a new destination.
• Ship small loops. Features released in tight cycles with live users, with feedback and usage shaping each next release, the same idea-to-production-to-optimisation loop enterprises need for any AI initiative.
• Design for governance from day one. Because the system generates the field's words, the organisation's standards can be applied where the words are made, supporting more consistent, policy-aligned messaging instead of reviewing it after the fact.

Anatomy of the system.

Ask Duply is delivered as a web application and a browser extension, built on an edge-serverless architecture with a large language model at the reasoning core. Described at pattern level:

• Signal engine Helps users identify and prioritise relevant relationship signals they choose to capture, life events, wellness goals, buying intent, surfaced as a prioritised queue instead of leaving follow-up to chance.
• Coaching layer A conversational coach with working memory that diagnoses each prospect situation, explains its reasoning, and hands the seller a ready-to-send message in their own voice, next best action, not blank box.
• Pipeline intelligence Stage-aware CRM for prospects, customers and team members, with heat decay, proactive alerts and who-to-focus-on-today guidance built into the data model rather than bolted on.
• Content engine Generates on-brand social content on a rotation system designed to avoid repetition and platitude, tuned to each seller's captured voice, with the organisation's standards applied at generation time.
• Learning loop Tracks which suggestions get used and what happens next, feeding outcomes back into future coaching, the feedback mechanism that keeps an AI system useful after launch.

These are platform services, not app features. The same decision engine, user memory, behavioural sequencing and automation layer powers the consumer wellness programs elsewhere in this portfolio, Ask Duply is one interface on an architecture built to carry many.

In production. In use. Improving.

Ask Duply is in production with active users and has moved through multiple releases shaped by live feedback. The platform anchors two commercial motions: a direct subscription product for individual sellers, and an enterprise conversation with direct-selling companies, where support for more consistent, policy-aligned field messaging has proven to be the lead, not the feature list.

The sharpest commercial lesson came from the enterprise side: what organisations fear about field AI is loss of control. A system that gives the company a role in how AI behaves in the field converts that fear into the business case.

Built lean, built to scale.

AI

Anthropic Claude, reasoning, coaching, generation

Backend

Edge-serverless workers with key-value storage

Clients

Web application · Chrome extension

Channels

SMS and email integrations for proactive coaching

Practices

Versioned deploys · prompt architecture · evaluation-led iteration

What building it taught me.

• Adoption is designed, not assumed. The features that changed behaviour were the ones that removed a decision, not the ones that added a capability.
• Steering beats chatting. An AI that proposes the next action outperforms an AI that answers questions, because most people don't know what to ask.
• Governance is a design input. Applying standards where content is generated is cheaper, safer and more effective than reviewing after the fact.
• The loop is the product. Launch is the midpoint. The value of an AI system is decided by how it learns from use, which means feedback capture has to be architecture, not afterthought.