# The work is already happening. It is just not in one system. > 7Unit puts governed AI inside the software a business already operates. One workflow at a time, wired to your CRM, ERP, WhatsApp or documents, with human approval where consequences matter and deterministic rules where a model is not needed. Provider: 7Unit Offer: Automate — AI Automation & Integration Audience: For operators and product leaders whose requests, documents and follow-ups live in WhatsApp, email, spreadsheets and a CRM that nobody updates. Canonical page: https://www.7unit.tech/services/automate ## What you get - **Workflow map and automation boundary** — The current process, the systems involved, the system of record, and a written line between rules, model and human decision. - **The automation, integrated** — Built into your CRM, ERP, WhatsApp number or document store. Not a chat window beside them. The existing system stays primary. - **Controls on the loop** — Human approval for anything that changes money, compliance or a customer commitment. Audit of every state change. Defined handoff when confidence drops. - **Observability and fallback** — Latency, errors and token cost visible per workflow. A tested path for when the model provider is slow, wrong or down. ## Shape and price - **Duration** — Workflow diagnosis 1 week · First automation in weeks, not quarters - **Cadence** — One workflow, measured against its baseline, then the next - **Team** — Senior engineer-led, with the people who run the process - **Workflow diagnosis** — Starts from USD 2,500. Covers the workflow map, boundary and success criteria. The automation build is scoped from that map. Quoted in AED or INR for UAE and India engagements. ## What this covers - **Customer conversations** — WhatsApp and messaging agents that keep session context, qualify a request and hand off to a person with the thread intact. Or buy Pingbot, the product we operate. - **Documents in, fields out** — Classification, extraction and routing for document-heavy operations. A person reviews the exceptions, not every file. - **AI inside existing software** — Assistants, classification and retrieval built into the CRM, ERP or product your users already open. Your tenancy, your permissions, your data. - **Governed agents (AgentOps)** — Agents wired to Slack, GitHub, Jira, CRMs and internal systems with structural approval gates and attributable actions. Run on our own engineering workflow first. - **Business automation without AI** — When a rule is enough, we do not wrap it in a prompt. Routing, notifications and data movement engineered deterministically. ## Proof - **Practical AI assistants** — Production LLM assistants including WhatsApp inquiry agents: session memory, human handoff with context, per-tenant cost attribution, graceful degradation when the model fails. [Read the AI assistants case](https://www.7unit.tech/work/ai-assistants) - **AML and document compliance engine** — 80+ documents per batch, unusual clause language flagged before human review, SLA breach rate from 34% to 3%. Deterministic where a value must be exact. [Read the compliance engine case](https://www.7unit.tech/work/aml-compliance-engine) - **Closer CRM and Pingbot, UAE** — WhatsApp-native pipeline, qualification and follow-up for UAE sales teams. 340+ daily conversations handled without agent involvement across our UAE deployments. [Read the Closer CRM case](https://www.7unit.tech/work/closer-crm) ## How it starts 1. **Book the call** — Thirty minutes. Bring the workflow that leaks: the inbox, the spreadsheet, the approvals stuck in chat. We tell you whether AI belongs in it. 2. **Workflow diagnosis** — One week with the people who run the process. Baseline, systems, boundary, success criteria. Written down before a model is chosen. 3. **Build, control, measure** — The automation goes into your systems with approvals and audit. Measured against the baseline. Then the next workflow, if the first one earned it. ## FAQ ### What can an AI automation company automate for us? Repeat operational work with a clear input, a system to write to, and a rule for when a person must decide. For 7Unit that has included inquiry handling, document intake, CRM follow-up, routing and internal assistance. We do not automate a process we have not mapped. ### Can you integrate AI with our existing CRM or ERP? Yes. The default design keeps your CRM or ERP as the system of record and lets the automation read and write through its API. We have done this with messaging-led CRMs and with back-office systems such as Odoo. Replacing the system is a separate decision, and usually the wrong first one. ### How is AI automation different from traditional workflow automation? Traditional automation moves data and triggers actions with rules. AI adds a model where the input is unstructured: language, documents, a judgement that can be drafted then checked. Most production systems need both. If a step is already a rule, adding a model only adds cost and a new failure mode. ### What happens when the model API fails? The workflow degrades on purpose. Live conversations escalate to a person with the thread intact. Asynchronous work queues and retries. Nothing that changes money or a customer commitment runs without the approval step. We test that path before we call the automation production. ### How do you handle sensitive business data? We design the control path before a model sees operational data: what may be sent, what stays in your systems, who approves an action, how calls are logged. Client data is not treated as training material. Regulatory interpretation stays with your counsel. ### Do you build custom AI agents? Yes, when an agent is the right shape for the workflow. Custom means wired to your tools, with approval gates, audit and a fallback. We run this pattern, which we call AgentOps, on our own engineering workflow before recommending it to clients. ### How long does an AI automation project take? Diagnosis takes about a week. A contained workflow with a clear system of record can be live within weeks. One that crosses several systems or handles sensitive data takes a deeper look first. Milestones are agreed after the map, not before. ### Can we start with one workflow? That is the engagement we prefer. One process, a baseline, the systems involved, the control boundary and success criteria. The engineering is scoped to that outcome, not to a bucket of hours. ## Related - [Build: product and platform engineering](https://www.7unit.tech/services/build) - [Pingbot, the WhatsApp assistant we operate](https://www.7unit.tech/products/pingbot) - [The four ways AI agent implementations fail](https://www.7unit.tech/insights/ai-agent-failure-modes) - [Send an automation brief](https://www.7unit.tech/rfq) ## Contact Book a 30-min call: https://www.7unit.tech/contact Send a brief: https://www.7unit.tech/rfq