Building-materials demand is made by thousands of small actors — dealers, contractors, sites — whose behaviour changes quietly and locally. By the time a quarterly review sees the change, the competitor's scheme has been running for two months. The doctrine — deterministic signals, AI delivery, human approval — is how the change gets heard in week one.
What the engines watch
- Quiet influencers surface themselves — contractors with steady accrual history who stopped earning points, clustered by territory, with their linked sites still active.
- Site follow-ups time themselves — stage-based requirement windows put the follow-up visit on the plan when the buying decision is live.
- Credit drift flags itself — dealers trending toward limits or ageing past terms, before despatches get held.
- Nudges publish overnight — scheme budgets burning fast, slab attainment stalling, coverage gaps — queued with evidence for the morning.
What agents draft for approval
Sense answers the territory's questions from its own records — "which enrolled contractors went quiet, and whose territory are they in?", "which sites hit finishing stage with no order?" — and drafts the win-back visits, the dealer outreach, the scheme tweak proposal. Writes wait for a human; answers show their working; permissions scope everything to the asker.
Governed enough for a loyalty program
Programs that move money need agents that can't: budgets enforced before model calls, bounded tools, approval-gated writes, full audit — and AI Studio to build your own agents on your product and program knowledge. Governance in full: AI trust.
Common questions
How do agents protect the influencer program?
By watching accrual patterns: sudden quiet from steady earners — the classic competitor-poach signal — surfaces as a clustered nudge with drafted win-back actions, weeks before program reporting would notice.
Can agents time site follow-ups?
Yes — construction stages imply requirement windows, and the follow-up work generates onto the right officer's plan for that window, deterministic and inspectable.
Do agents change schemes or prices?
No — they surface the evidence and can draft a proposal; scheme and price changes go through the same human approvals as always, with the agent's reasoning attached.
What does the sales head get each morning?
A daily brief per user — their territories' numbers computed deterministically, the overnight nudges ranked, and the ability to ask follow-up questions in plain language.