# Business automation for workflows that still depend on people copying work.

> 7Unit engineers the operational workflow: routing, approvals, notifications, CRM and ERP handoffs, and the software that makes the process visible. AI is used where it improves the workflow. We do not force AI into a process that deterministic software can solve better.

Provider: 7Unit
Service: Business Automation
Audience: For operators and founders whose problem is a manual process, whether or not a model belongs in it.
Canonical page: https://www.7unit.tech/solutions/business-automation

## The process runs. It just runs on effort.

People move status by hand, chase approvals in chat, and keep a spreadsheet because the official system does not match the work. The fix is workflow engineering. A model is optional.

- **Copying between tools** — The same customer, job or document is retyped from WhatsApp or email into a CRM, ERP or sheet.
- **Approvals with no state** — Sign-off happens in a message. Nobody can see what is waiting, or what was agreed.
- **Notifications that are not a system** — Reminders depend on a person remembering. When they are away, the process stops.
- **No view of the operation** — Leadership asks where the work is. The answer is a round of calls, not a record.

## Outcomes

- **The process has a state** — A request, deal or job moves through stages the team can see, with an owner.
- **Less retyping** — Data moves between the systems that already participate, on purpose.
- **AI only where it helps** — Unstructured language or documents can use a model. Exact, regulated or rule-shaped steps stay deterministic.

## What business automation covers

This is operational software and integration. It overlaps with AI only when the workflow needs it.

- **Workflow engineering** — The real path a lead, job, learner or dossier takes, written as states, roles and handoffs.
- **System integration** — CRM, ERP, messaging and internal tools sharing one record instead of parallel copies.
- **Approvals** — A defined sign-off, recorded against the work, not buried in a chat thread.
- **Notifications** — The next person is told because the state changed, not because someone remembered.
- **Business rules** — Routing, eligibility and exceptions expressed as rules you can review. No model required.
- **Operational software** — When the off-the-shelf tool cannot hold the process, we build the operational system around it.

## How we engineer the workflow

We use AI where it improves the workflow. We do not force AI into processes that deterministic software can solve better.

1. **Walk the process** — Who touches the work, in which tool, and where it waits. Including the unofficial spreadsheet.
2. **Name the system of record** — One place owns the state. Other tools sync to it or read from it.
3. **Separate rules from judgement** — Rule-shaped steps are software. Language and documents may use a model. Ambiguous commitments stay with a person.
4. **Integrate** — Move data and events between the systems already in the operation. Replace a system only if it cannot hold the workflow.
5. **Make it visible** — Queues, owners and exceptions are inspectable. Automation you cannot see is a new operational risk.
6. **Measure the baseline** — Cycle time, missed handoffs or manual steps removed. Compared with the process you described, not a generic benchmark.
7. **Leave it operable** — Runbooks and a team that can change a rule without calling us for every exception.

## Deterministic where it must be. AI where it helps.

We use AI where it improves the workflow. We do not force AI into processes that deterministic software can solve better. A regulated document that must resolve to one exact value is a rules engine. A messy inquiry that must be understood is a candidate for a model, with a person on the exception path.

- Workflow state
- Business rules
- System integration
- Approvals
- Notifications
- AI only if it helps

## Operational systems we have shipped

These cases are workflow and systems work. AI is present only where the case says so.

- **Bloom** — Educators stayed in Odoo. Learners got a portal. Curriculum publish became event-driven, and certificates generated on completion instead of by hand. [Read the Bloom case](https://www.7unit.tech/work/bloom)
- **AML and document compliance** — Compliance questionnaires as configuration, and document tokens that resolve to exactly one value. The system refuses to guess. No model in the path that must be exact. [Read the compliance engine case](https://www.7unit.tech/work/aml-compliance-engine)
- **Closer CRM** — A sales operation moved into the WhatsApp thread the team already used: pipeline, assignment and follow-up, with human takeover on AI-assisted steps. [Read the Closer CRM case](https://www.7unit.tech/work/closer-crm)
- **7Hub** — A multi-tenant loan origination platform: the operational system itself, not a layer of prompts on top of spreadsheets. [Read the 7Hub case](https://www.7unit.tech/work/7hub)

## Case studies

- [Bloom](https://www.7unit.tech/work/bloom)
- [AML and document compliance engine](https://www.7unit.tech/work/aml-compliance-engine)
- [Closer CRM](https://www.7unit.tech/work/closer-crm)
- [7Hub](https://www.7unit.tech/work/7hub)

## Start with one process

We map the current path, agree the outcome, and build the smallest operational loop that achieves it. Developer hours are a consequence of that scope, not the product. We do not promise a cost saving before the baseline exists.

- Current process, including the tools people actually use.
- A baseline: volume, delay, handoffs, or where work gets lost.
- The operational outcome you want, written before a model or a build is chosen.
- Systems involved, and which one remains the system of record.
- Risk and control boundaries: what may be automatic, what needs a person.
- Success criteria the engineering can be judged against.

## Processes that usually need this
- **A back office beside the real work** — Staff do the job in one tool and report it in another. Bloom’s educators and learners were on opposite sides of that gap.
- **Regulated paperwork assembled by hand** — The same facts are copied into documents, with a real cost when the wrong value lands on the wrong person.
- **A pipeline that is not the conversation** — Sales or service happens in WhatsApp. The CRM is a fiction updated at the end of the day, if at all.
- **Handoffs between teams** — Sales, operations and finance each keep a version. The customer feels every gap.

## Systems this work usually touches

We integrate what the process already depends on, then build only the missing operational layer.

- CRM
- ERP and Odoo
- WhatsApp
- Email
- Spreadsheets
- Approval roles
- Internal databases
- Notification channels

## FAQ

### What is business automation if it is not an AI project?

It is workflow engineering: state, rules, integrations, approvals, notifications and software that matches how the company operates. AI is a tool inside that, used when the input is unstructured. Many of the systems on our work page were automated without making a model the centre of the design.

### When do you use AI in a workflow?

When the step needs language or document understanding, and a wrong answer can be caught by a rule or a person. We do not use AI to generate a value that must be exact, or to avoid writing a business rule you already know.

### Can you automate approvals and notifications?

Yes. Approvals become a recorded state with an owner. Notifications fire because that state changed. The policy of who may approve stays yours. We implement it. We do not invent your authority model.

### Do we have to replace our CRM or ERP?

Usually no. Closer was built because the conversation and the CRM were different products. Bloom kept Odoo. We recommend replacement only when the current system cannot represent the workflow safely.

### Can you connect WhatsApp, email and the system of record?

Yes, when that is where the work already enters. The design question is which system owns the record, and which messages are allowed to change it. WhatsApp Business API rules still apply to how a business may message.

### Can we start with one process?

Yes. One path from trigger to done, with a baseline and success criteria. A second process is a later decision, after the first loop is operable.

### How do you know a step should stay manual?

If the cost of a wrong automatic action is higher than the cost of the delay, or if the rule cannot be stated, the step stays with a person. Discovery is where we write that down. We would rather leave a step manual than hide a judgement inside a prompt.

### How long does a workflow engineering engagement take?

Discovery is typically one to two weeks. The build follows in milestone loops once the process, the system of record and the success criteria are agreed. We do not quote a single duration for every operation.

## Related

- [When WhatsApp is the operating surface](https://www.7unit.tech/insights/whatsapp-product-surface)
- [AI automation when the workflow needs a model](https://www.7unit.tech/solutions/ai-automation)
- [AI integration into systems you already run](https://www.7unit.tech/solutions/ai-integration)
- [Custom CRM and operational systems](https://www.7unit.tech/solutions/crm-operational-systems)
- [Send a workflow brief](https://www.7unit.tech/rfq)

## Contact

Discuss the work: https://www.7unit.tech/contact
Send a brief: https://www.7unit.tech/rfq