# Turn email-and-spreadsheet processes into tracked workflows. > Approval, request and document workflows moved off email and spreadsheets, using the right tool per step: API integration, RPA or a supervised AI agent. Source: https://www.algofact.tech/en/services/solution-engineering/business-process-automation Language: en-US Updated: 2026-10-26 Languages: en-US, tr-TR ALGOFACT turns approval, request and document processes that run on email, spreadsheets and paper into tracked workflows with defined roles. We pick the tool step by step: API integration and a workflow engine where rules are clear, RPA where a legacy system has no interface, and an AI agent with human approval where documents need interpretation. We measure the process before we automate it and again afterward, with the same yardstick. ## Key facts - **Focus:** Approvals, requests and documents - **Methods:** Workflow engine, APIs, RPA, AI agents - **Control:** Role-based approval and audit trail - **Measurement:** Process metrics before and after - **Information security:** ISO 27001:2022 - **Starting point:** A pilot on a single process ## Which processes are worth automating? The best candidates repeat often, follow rules that can be written down and today depend on several people chasing each other by email: purchase requests, expense and leave approvals, contract signature rounds, customer applications, supplier onboarding. The delay is rarely the work itself. It is that nobody can see whose desk the work is sitting on. Automating a broken process only makes it break faster. We map the process first and strip out redundant approvals and duplicate data entry with you. Sometimes the biggest gain comes from removing an approval step before any software is written. ## RPA, API integration or an AI agent? Different steps in the same process often call for different tools, so we choose step by step. - **API integration and a workflow engine:** The first choice when systems have APIs and the rules are clear. It is fast, unaffected by screen changes and easy to monitor. Approval steps, forms and notifications live in a workflow engine. - **RPA bots:** Bots mimic the clicks a person makes in a legacy desktop or web application that has no API and cannot be changed. A useful bridge, but bots break when screens change, so we treat them as temporary and monitor them. - **AI agents:** For steps that rules cannot capture, such as reading and classifying an incoming document or email and spotting missing information. The agent proposes and a person approves; it does not complete steps that touch money or contracts on its own. - **People:** Judgment, negotiation and exception handling stay with people. A good workflow does not remove these steps; it brings the right person the right information at the right time. ## Example processes ### Procurement Request, budget check, quote collection, approval and the purchase order pushed into the ERP. The AI-supported RFP generator built for RFx Engine automates RFP drafting for supply chain teams. ### Work assignment and queues Incoming work routed to the right person or team by rules. For Türkiye’s Ministry of Health we built automatic assignment software for contact tracing during the pandemic. ### Document and contract flows Generated from a template, routed to signers in order, completed with e-signature and archived in searchable form. ### Collections and membership dues Recurring payment reminders, recording collections and reconciliation. The membership dues collection system built for KAFFED is one example. ### Onboarding and applications Customer, supplier or employee onboarding with document collection, checks, requests for missing information and approval in one flow. ## How the project runs ### Process map and baseline We capture today’s steps, durations, waiting points and exceptions from system records and interviews. Success metrics are written down as numbers at this stage. ### Redesign Redundant steps go, each remaining step gets a method (workflow, API, RPA, agent or person) and approval authority is defined. ### Pilot We start with one process or one department. Users run the old and new paths side by side for a short period, and issues are fixed inside the pilot. ### Rollout Based on pilot results, the workflow extends to other departments and processes, with short training sessions and per-process guides. ### Monitor and improve A process dashboard shows pending items, durations and bottlenecks. The workflow is built so new rules can be added without code changes wherever possible. ## Audit trail and authority Every approval, rejection and handoff is recorded with who made it, when and on what information, so an internal or external auditor can trace a decision from the screen. Approval authority is tied to roles rather than individuals, so the process does not stall when someone goes on leave or changes jobs. RPA bots and AI components run under their own service accounts, with access only to the systems they need. For teams spread across several countries, approval chains and data storage location are set per legal entity, and retention periods and access limits for personal data follow GDPR and KVKK inside the workflow itself. ## What you receive ### Process maps Current and target process, with the method and owner for each step. ### Workflow application Forms, approval steps, notifications and role-based screens. ### Integrations and bots API connections and, where needed, RPA bots and AI components. ### Process dashboard Pending items, timing metrics and the audit trail. ### Training and runbooks User guides, administrator documentation and bot maintenance instructions. ## Questions ### Should we use an off-the-shelf workflow tool or build our own? If you already run a workflow tool or a platform such as Microsoft Power Automate and the process is standard, we look at that first. Custom development fits better when the process is tied deeply into several internal systems, when outside users such as customers or suppliers take part, or when licensing cost climbs quickly with user count. ### Do RPA bots need maintenance? Yes. A bot stops when the screen it mimics changes. We set up monitoring and failure alerts for every bot and document which screen changes would affect it. Over time, replacing the same step with an API integration is usually more robust. ### How do we know the automation worked? The metrics we baseline at the start, such as step duration, waiting time, errors and rework, are measured again after automation using the same method. The results are reviewed with you at the end of the pilot. ### How do we get people to actually use the new workflow? Process owners are involved from the design stage. The pilot runs with real users, and screens get simpler based on their feedback. Short training sessions and per-process guides support the switch. ### What drives the timeline and cost? How many processes are in scope, the number of steps and exceptions in each, whether the connected systems offer APIs and whether an AI component is needed. After the discovery call we present the scope and plan for a single-process pilot in writing. ## Related - [Solution Engineering](https://www.algofact.tech/en/services/solution-engineering) - [ERP, CRM and HIS integration](https://www.algofact.tech/en/services/solution-engineering/erp-crm-integration) - [AI agents](https://www.algofact.tech/en/services/ai-ml/ai-agents) - [Use cases](https://www.algofact.tech/en/work) - [Discovery call](https://www.algofact.tech/en/contact)