Actiknow
Custom Software Development

Approval Workflow Design: How Many Controls Does a Business Process Actually Need?

Learn how to design approval workflows with risk-based thresholds, segregation of duties, escalation rules and audit trails without slowing operations.

Finance professional reviewing a business approval document before authorization
Contents hide

The question is not how many approvals you can automate

A purchase request waits in a manager’s inbox. The manager approves it, finance approves it again, and a director signs off because the previous spreadsheet required three signatures. Nobody can explain what the third approval checks. Meanwhile, urgent requests are handled through phone calls and retrospectively marked approved. The process looks controlled on paper but is neither fast nor consistently governed.

That is the central challenge in approval workflow design: a control must reduce a meaningful risk, not merely add another name to a chain. An approval can prevent an unauthorized commitment, protect a budget, resolve an exception or establish accountability. It can also become a rubber stamp that delays work while creating the illusion of oversight.

For COOs, finance leaders and product owners, the objective is to give ordinary, low-risk work a predictable path while directing consequential or unusual decisions to people with the right authority. That requires a business policy, a clear decision model and a software workflow that can enforce the policy without trapping operations.

Start with the decision, not the org chart

Before adding an approver, write down the decision that person must make. Is the reviewer checking budget availability, confirming the commercial rationale, validating regulatory evidence or accepting an exception to policy? If two reviewers perform the same check against the same information, consider whether one approval can be removed or replaced by an automated validation.

A useful design workshop starts with five questions: What outcome is being authorized? What could go wrong? How large is the exposure? What evidence does a reviewer need? Who can accept the residual risk? Those answers are more valuable than copying a paper form into a digital system.

This is also why process discovery matters before development. An organization may believe that a six-day cycle is caused by slow approvers, when missing purchase-order references or unclear ownership create most of the delay. The correct intervention may be a mandatory field, an integration or a policy clarification rather than another notification.

Classify requests by risk before defining approval levels

One approval rule for every transaction is easy to describe and expensive to operate. Risk-based routing separates standard cases from exceptional ones. Monetary value is an important input, but it is not a complete risk model: a low-value payment to a newly changed bank account may warrant more scrutiny than a higher-value recurring payment to an established supplier.

A practical four-tier decision model

Tier 0, routine and policy-compliant: The request meets predefined rules, has complete data and creates limited exposure. It can proceed automatically or through a lightweight acknowledgment when policy allows. Examples include a preapproved catalog purchase within an allocated budget or an internal access request covered by an existing role template.

Tier 1, standard discretionary decision: One accountable manager assesses a legitimate business judgment, such as a non-catalog purchase within delegated authority. The reviewer should see the purpose, budget, alternatives where relevant and the consequence of approval.

Tier 2, material or cross-functional exposure: A second independent check is justified because the transaction crosses a financial threshold, affects another budget, changes contractual obligations or has a sensitive data implication. The two approvals should test different things, such as business necessity and financial authority.

Tier 3, exception or high-consequence decision: Route policy overrides, unusual counterparties, sensitive access, material contract deviations or elevated fraud signals to a specifically authorized function. Include a written rationale, additional evidence and an explicit exception owner. Some actions should be blocked rather than approvable through an ordinary exception path.

These tiers are a design template, not universal thresholds. The organization must set values and triggers using its own loss exposure, regulations, delegated authority, operational frequency and risk appetite. A system should store the rule version that classified each request so historical decisions remain explainable.

Purchase requests organized into different approval priority and risk categories

Use monetary thresholds carefully

Thresholds can be useful because they are objective, testable and easy to communicate. But a simplistic rule such as ‘every request above $5,000 needs a director’ invites threshold gaming and may ignore context. A better policy can evaluate amount, cumulative spending, department budget, supplier risk, category, contract terms and whether a request is split into related transactions.

For example, a procurement workflow might route an approved supplier invoice that matches an existing purchase order directly to accounts payable; send an unmatched but low-value invoice to the budget owner; and require an additional finance review for an unusual payment destination or a material variance. The specific amounts are organization-defined. The important design principle is that each branch corresponds to a distinct risk.

Include aggregation rules. Ten small requests from the same initiator or supplier over a short period should not automatically evade a threshold intended for the total commitment. Decide how the system handles currency conversion, tax-inclusive amounts, amendments and recurring obligations. Document whether approval applies to the transaction, the contract ceiling or the full expected lifetime cost.

Separate authority from mere participation

Not every stakeholder needs an approval button. A requester supplies information; a reviewer may advise; an approver accepts responsibility; a processor executes the authorized action; and an auditor needs access to evidence. Confusing those roles creates long chains and ambiguous accountability.

Segregation of duties is particularly important when one person could initiate, authorize and execute a sensitive transaction. NIST’s security and privacy control catalog provides a recognized reference for access controls and separation-of-duties considerations; the practical implementation must reflect the organization’s risk and systems. See the NIST SP 800-53 control catalog for the broader control framework.

For a high-risk vendor bank-detail change, the person editing the payment instructions should not be able to independently approve the same change. For a low-risk internal content update, requiring a second person may offer little protection. Enforce conflict rules across roles and identities, not just by displaying two buttons on a form.

Separate finance employees performing payment verification and authorization tasks

Delegation also needs guardrails. A temporary delegate should inherit only the relevant authority for a defined period, and the audit record should show both the original approver and the acting delegate. A person must not use delegation to approve their own request when self-approval is prohibited.

Choose sequential, parallel or single approval deliberately

Single approval is appropriate when one owner has both the information and delegated authority to decide. Sequential approval makes sense when a later reviewer genuinely needs the earlier decision or evidence. Parallel approval can reduce waiting when independent functions can review the same request simultaneously. Avoid forcing parallel reviewers into a sequence merely because the workflow tool’s default template does so.

The distinction is supported by Microsoft Power Automate’s documented approval patterns, including first response, everyone must approve and sequential approvals. Those are platform capabilities, not recommendations that every business process needs every option. Review the official Power Automate approvals guidance when comparing implementation behavior.

Define the decision semantics explicitly. Does one rejection stop the request? Must every member of a group agree? Can one designated representative act for a team? Does an abstention count? What happens if one of two parallel approvals becomes obsolete after the request changes? Ambiguity in these questions is a common source of production disputes.

Business approval documents arranged for different review and decision paths

Build the approval policy as a decision table

A decision table is easier to review and test than a paragraph of informal instructions. Each row should define the request type, conditions, approver role, required evidence, response deadline, exception path and outcome. Order overlapping rules or assign explicit precedence so the same request cannot match two contradictory routes.

Illustrative approval matrix

Scenario A: A standard purchase with an approved supplier, available budget and matching catalog item. Required control: automated policy and budget checks; no discretionary approval if the organization’s policy permits.

Scenario B: A non-catalog purchase within delegated authority. Required control: one budget-owner approval with purpose and cost-center evidence.

Scenario C: A purchase above the delegated threshold or with material contract terms. Required control: budget owner plus a separate finance or commercial authority, each reviewing a defined concern.

Scenario D: A supplier bank-account change. Required control: independently verified supplier change process, restricted execution rights and a separate authorization path; a simple amount threshold is not sufficient.

Scenario E: An urgent request outside normal policy. Required control: a time-limited exception decision by an authorized role, recorded justification, explicit expiry and retrospective review where appropriate. Urgency should not silently bypass non-negotiable safeguards.

These scenarios illustrate logic, not a prescribed compliance policy. A useful acceptance test is whether a manager can read the table and predict exactly what will happen for a given request without consulting the developer.

Treat the approval as a stateful business transaction

A reliable approval application does more than send messages. It records a request’s identity, current state, version, owner, approver assignments, evidence and decision history. A practical state model might be Draft → Submitted → Validation Failed or Pending Approval → Approved, Rejected, Withdrawn or Expired → Executed or Execution Failed. Resubmission should create a clear new decision context rather than rewriting history.

Every approval must refer to a specific version of the request. If the amount, payee, access scope or contract terms change after approval, the workflow should re-evaluate its policy and determine whether previous approval remains valid. Silent edits after sign-off undermine the very control the workflow was designed to provide.

Execution also needs a separate outcome. ‘Approved’ means an authorized person accepted a request; it does not prove the ERP posted the invoice, the user account was provisioned or the contract was sent. Record the downstream system response, retry policy and reconciliation status. If an integration times out, do not blindly repeat a financial or provisioning action without checking whether the first attempt succeeded.

For teams building tailored processes, Actiknow’s custom solutions practice describes application development, API integrations and business-process automation. Those capabilities are relevant when approval rules must connect several systems and cannot be safely implemented as an isolated email flow.

Design for absence, escalation and exceptions

A process that works only when every approver is online is not production-ready. Define response service levels by request category. A routine procurement request might receive a reminder after an agreed period; a critical operational exception may need immediate routing to an on-call role. The deadlines must reflect business urgency and staffing, not arbitrary software defaults.

An escalation should be a controlled reassignment or visibility change, not an automatic approval. The new decision-maker needs sufficient authority, and the system should record why the case was escalated. If no eligible approver is available, put the request into an explicit blocked state with a visible owner instead of silently letting it disappear.

Exceptions need their own structured path. Capture the rule being overridden, justification, business impact, authorizer, compensating control and expiration. Do not make ‘Other’ or ‘Urgent’ an unreviewed shortcut around policy. Periodically analyze exception rates; a high rate may mean that policy is unrealistic, data is poor or people have found a convenient bypass.

Make the reviewer experience good enough to support judgment

Approval quality depends on the information presented. A reviewer should see the request’s purpose, amount or exposure, relevant history, supporting evidence, policy trigger, available alternatives and the precise action being authorized. If reviewers must open six systems to understand a request, many will approve from habit or delay it until they have time.

Present concise summaries with links to source records and an accessible detail view. Use explicit actions such as Approve, Reject and Request Changes, with reason codes where they improve analysis. Require meaningful comments for exceptions and rejections, but avoid forcing a lengthy justification for every routine decision.

For a custom web application, design approval screens for desktop and mobile use, but apply the same identity and authorization checks regardless of channel. Actiknow’s custom web application development service is relevant when the workflow requires bespoke screens, business logic and integrations rather than a generic form.

Preserve an audit trail that explains the decision

A useful audit record answers: who submitted the request, what information was reviewed, which policy version applied, who had authority, what decision was made, when it occurred, what changed afterward and whether execution succeeded. Store immutable or tamper-evident events according to the system’s security and retention requirements. Limit access to sensitive evidence.

Auditor examining financial approval records and authorization evidence

Capture both the human decision and the machine evaluations that preceded it. An automated approval should not appear in reports as if a named manager personally reviewed the request. Label policy-based auto-authorization distinctly, record the checks performed and preserve the rule version. This makes control testing and incident investigation much more credible.

Auditability is not a reason to keep every attachment forever. Define retention based on legal, contractual and business needs. Avoid putting sensitive personal or payment data into broad notification emails or verbose application logs. A workflow can provide a complete decision history while minimizing unnecessary exposure.

Test the controls before release

A happy-path demo proves little. Build a test suite from the decision table, covering values immediately below and above thresholds, conflicting conditions, incomplete evidence, changed approvers, role changes, delegation, self-approval attempts, withdrawn requests, expired tasks, retries and downstream failures.

Quality assurance specialist validating business approval rules using testing checklists

Include adversarial cases: a requester splits one commitment into multiple transactions; an approver loses authority after receiving a notification; a user edits an amount after approval; two reviewers act simultaneously; a webhook is delivered twice; or the destination system completes an action but its acknowledgment is lost. The correct outcome should be defined before implementation.

Test policy changes against historical representative requests before deploying new thresholds. A seemingly small rule adjustment can redirect hundreds of transactions or accidentally eliminate a required independent check. Use versioned configuration, peer review, staged rollout and rollback procedures. Keep a named business policy owner accountable for approving changes to the decision table.

Measure whether controls work, not just whether requests move

Executives need a small set of operational and control metrics. Track end-to-end cycle time by request tier; time waiting for approval versus time spent preparing or executing; percent of requests routed automatically; exception and rework rates; overdue cases; unauthorized-action attempts; and the proportion of approvals that later require correction.

Segment results by department, transaction type, risk tier and approver role. An improving overall average can conceal serious delays in high-risk requests. Similarly, a rising auto-approval rate may be positive only if policy violations, errors and downstream corrections remain controlled.

Measure review quality through targeted audits and sampled evidence, not by rewarding approvers for clicking faster. A five-second approval is not inherently efficient if the reviewer lacked the information to make a sound decision. Balance speed with control effectiveness and business outcomes.

A 30-day implementation plan for a better approval workflow

Days 1–7: Discover and quantify

Inventory the highest-volume approval processes and interview requesters, approvers, processors, finance, security and audit stakeholders. Map the actual paths, not just the documented ones. Extract available timestamps to identify queues, rework and common exceptions. Identify the business owner who can approve a policy redesign.

Days 8–14: Simplify the policy

Create a risk-tier model and a decision table. Remove duplicate reviews, define monetary and non-monetary triggers, specify segregation-of-duties conflicts, and document escalation and emergency paths. Have business and control owners sign off on the intended behavior before the development team configures it.

Days 15–21: Build and test a narrow pilot

Implement one representative workflow with clear request states, policy-version capture, reviewer screens, notifications and downstream integration checks. Run edge cases and authorization tests. Include an audit view and an exception queue from the first release rather than treating them as future enhancements.

Days 22–30: Pilot, measure and adjust

Release to a limited business group with a documented fallback procedure. Compare cycle times, approval volume, exception frequency and control outcomes against the baseline. Review cases that were delayed, overridden or returned. Refine the rules and ownership model before extending the workflow to more departments.

What leaders should decide before approving the project

Ask for six deliverables: a documented business objective; a request and risk taxonomy; an approved decision table; a role and segregation-of-duties matrix; a state and exception model; and a measurement plan. Without these, the team may build a polished interface around an unresolved policy problem.

Technology selection comes afterward. A straightforward internal approval can fit an existing workflow platform. Complex cross-system authorization, unusual business rules or a differentiated user experience may justify custom software. Either way, the organization owns the policy and should be able to change it through governed configuration rather than undocumented code edits.

Frequently asked questions

1. How many approval levels should a workflow have?

As few as necessary to address distinct, documented risks. Some requests need no discretionary approval, others need one accountable decision-maker, and a limited set need independent or specialist review. The number should follow the risk and authority model, not the organizational hierarchy.

2. Should low-value purchases be automatically approved?

They can be, if policy permits and controls such as budget availability, supplier eligibility, category restrictions and cumulative-spend checks are reliable. Low value alone does not eliminate fraud, security or compliance risk.

3. What is segregation of duties in an approval process?

It means assigning potentially conflicting responsibilities to different authorized people or roles so one person cannot unilaterally initiate, approve and execute sensitive activity. The exact conflicts depend on the process and risk.

4. Is parallel approval better than sequential approval?

Parallel review is usually preferable when reviewers can make independent decisions at the same time. Sequential review is appropriate when a later decision genuinely depends on an earlier one. Neither model is universally superior.

5. What should happen when an approver is unavailable?

Use an authorized delegate, defined reassignment rule or escalation path. Preserve the reason and decision history. Never silently convert a missed deadline into approval unless a formally approved policy explicitly allows that outcome.

6. Can an employee approve their own request?

Only where the policy explicitly allows it and the risk is appropriate. For sensitive transactions, enforce self-approval restrictions in authorization logic and test them across delegation and administrative roles.

7. What should happen when a request changes after approval?

Re-evaluate the approval conditions. Material changes to amount, payee, scope or contractual terms should normally trigger reapproval by the appropriate authority. The previous approved version must remain traceable.

8. What is the difference between approval and execution?

Approval authorizes an action. Execution is the subsequent business or technical operation, such as posting an invoice or provisioning access. A complete workflow tracks both and reconciles downstream results.

9. What metrics show that approval automation is successful?

Track cycle time, waiting time, exception and rework rates, overdue approvals, policy compliance and downstream corrections. Faster routing is useful only when the controls remain effective and the business outcome improves.

10. When should a company build a custom approval application?

Consider custom development when the workflow spans multiple systems, has unique authority rules, needs specialized user interfaces or requires detailed operational reconciliation that standard tools cannot support cleanly. Start with the business rules, not a preference for a particular technology.

A proportionate control is a stronger control

The best approval workflow is not the one with the most signatures. It is the one that lets standard work proceed safely, identifies the cases that genuinely need judgment and leaves evidence that a decision was authorized and executed correctly. Reducing redundant steps can improve control clarity as well as speed.

If your team is redesigning a multi-system approval process, Actiknow can discuss the application, integration and automation requirements. Share the current workflow, decision rules and pain points through the Actiknow contact page to explore a practical scope.