How your operation makes the way the work gets done.
Build is the function that decides how the work gets done across two stacks. Macros, workflows, decision trees, the AI's prompt architecture, the handoff seam between bot and team. The infrastructure that turns "we have tools" into "the operation has a design."
v 02 · live
What Build means.
Build is the work of designing how the work gets done across both populations. Not the work of buying tools. Not the work of writing scripts. The work of architecting the workflows, macros, decision trees, routing logic, and AI prompt structure that turn a customer's request into a resolution. The human stack and the AI stack as one design, not two.
Most CX teams inherit a macro library nobody owns. Workflows that were correct 18 months ago. Routing rules written by an admin who's no longer there. Decision trees that don't match the product anymore. The team navigates around the parts that don't work and the operation calcifies.
Most operations also inherit an AI deployment configured by a vendor. The system prompt is a black box maintained by a CSM. Updates ship without operator review. The AI's behavior and the team's macros drift in opposite directions and nobody owns the gap between them.
That's the gap Build names. The tools have configuration; nobody owns it. The AI has a prompt; the vendor owns it. Workflows have logic; nobody maintains it. The operation runs on accumulated history on one side and vendor decisions on the other, rather than considered design. New product changes don't flow through. Drift becomes the default on both sides.
Haven's Build module starts with the workflow and prompt audit. What macros are used. What the AI is doing in the same scenarios. What's stale. What's drifting between the two. The macro library, the routing rules, the decision trees, the AI prompt structure: all named, all tracked, all owned in one place. A maintenance cadence replaces accumulation on both sides.
The operation gets simpler over time, not more complex. Each product change has a flow-through to the human stack and the AI stack. Each new capability has a place to live. The handoff seam becomes a designed surface, not an accident.
The progression. Four levels.
The macro library accumulates. The AI's prompt is a vendor black box. Workflows are written when problems appear and never retired. Nobody owns the configuration. The handoff between bot and team is whatever the AI decides. Drift is the default on both sides.
- Accumulated macros
- Stale workflows
- AI prompt owned by vendor, opaque to operators
- Handoff seam is accidental
Some maintenance happens on the human side, when there's time. The AI is still on its own schedule. Quarterly cleanups catch the worst macros. Most of the library is still leftover. AI prompt updates ship from the vendor without operator review. New product changes flow through inconsistently to humans and not at all to the AI.
- Quarterly macro cleanup at best
- Most library is legacy
- AI prompt drifts independently
- Maintenance is firefighting on both sides
The library and the AI's prompt are owned and audited together. Macros have owners, dates, and usage counts. AI prompt versions are tracked alongside macros. Stale workflows are retired routinely. Product changes have a defined path into both the human stack and the AI stack.
- Owned library + owned prompt
- Routine audits across both
- Clear retirement criteria
- Product change pipeline covers humans and AI
The library shrinks as quality grows. The AI's prompt evolves on the same cadence. Macros are consolidated. Prompts are versioned. Workflows get simpler. The operation responds to product changes within a sprint, on both sides. The handoff seam is a designed surface, not an accident.
- Library shrinking; prompt evolving
- Workflows simplifying
- Sprint-paced flow-through on humans and AI
- Handoff seam designed and monitored
What Build builds.
The macro & prompt audit
Every macro tagged with owner, last-used date, usage count. AI prompt versions tracked alongside. Stale on either side retired. The library and the prompt shrink first, then grow with intention.
- Owner attribution per macro and prompt section
- Last-used · usage count telemetry
- AI prompt versions tracked alongside macros
- Drift detection across humans and AI on every release
The workflow map
Top 20 customer journeys mapped end-to-end across humans and AI, including the handoff seam. Decision points named, owners assigned, drift detected. The operation made legible.
- End-to-end journey diagrams covering both populations
- Decision points named & owned
- Handoff seam captured as a designed surface
- Versioned per product release
The change pipeline
A defined path for product changes to flow into both stacks. Each release has named impacts on macros, processes, KB articles, and the AI's prompt. No more surprises on either side.
- Operational impact named per release
- Macro · process · KB · AI prompt updates queued
- Pre-flight checklist per change
- Post-release drift review across humans and AI
See it cascade.
A new contact cluster lands in the queue. Build is where the AI's prompt update gets drafted, alongside the team's process change in Improve and the knowledge article in Enable. One signal, three drafts, the team and the AI stay aligned by default. See how the cascade lands in Build →