A production CRM, designed and built end-to-end by a small AI-native team
erpost.com is not a slide about the future of AI-native delivery - it is the B2B sales CRM Tequma and its sister advisory firm run their own businesses on, built the way we now offer to build for clients.
Client
Two advisory businesses under common ownership - Tequma and its sister firm TylkoAdvisors - each running a small, roughly 20-person B2B sales team.
The challenge
Both businesses needed a proper CRM to run their sales funnel - lead capture through to closed deal - tightly integrated with the Microsoft 365 tools their teams already used for calendar, Teams meetings and email. Off-the-shelf CRM platforms meant recurring per-seat licence cost, generic workflows that did not match a company-centric B2B sales motion, and yet another system to integrate rather than one built around Microsoft 365 as the centre of gravity. Building it the conventional way - a scoped RFP, a system integrator, a multi-quarter programme - was disproportionate for a 20-user team.
The approach
- A single, senior-led AI-native engineering pod - not a multi-vendor delivery team - working through eight phase-gated stages: scope, architecture, data model, backend and integrations, frontend, CI/CD, infrastructure, rollout
- A data model built and verified against a live PostgreSQL database before a single screen existed - 14 tables, 31 foreign keys and 65 indexes, proven end-to-end rather than assumed correct
- Deep, delegated Microsoft 365 / Graph integration - calendar, Teams meetings and logged email - built directly on APIs already paid for, with no new iPaaS or per-integration subscription
- Engineering governance enforced from day one: automated schema-safety checks, a single enforced data-entry path, and a rule that every customer-visible change is documented in the same pull request that ships it
- One codebase deployed as an isolated instance per client, each with its own database, backups and health monitoring - so a second brand went live on the same platform in days, not a rebuild
The outcome
- ✓In production with real customer data since July 2026, serving two companies at ta.erpost.com and tq.erpost.com
- ✓Multiple releases shipped on a disciplined release-and-hotfix cadence, each one documented for the sales team that uses it
- ✓Zero new recurring subscription or integration-platform cost added by the build
- ✓A governance model - enforced invariants, in-transaction audit trail, phase-gated delivery - that mirrors the assurance discipline Tequma applies to clients' SAP and Salesforce programmes
This is why our AI-native delivery practice exists: we are not proposing an unproven delivery model - we run our own businesses on it, and we bring the same senior-led rigour, verified data models and enforced governance to the AI programmes we run for clients.