Your approvers should not spend their day searching for policy details, request status and approval history. Human in the loop AI can help people find the information they need faster, while the financial decision remains with the authorised person.
Techspire IT helps teams scope a controlled Claromentis pilot around one real approval process: clear boundaries for AI, clear ownership for people, and a practical first use case.
TL;DR: Human in the loop AI can take approval chasing off your approvers without taking over the approval. Use it for the repetitive questions that slow an approval process down: what is waiting, which policy applies, where a request has reached, and who approved it, and when. Keep approval rights, financial commitments and workflow changes with named people and agreed controls, and start with one approval process as a focused Claromentis pilot.
The practical answer
Do not begin by asking whether AI can make approval decisions. Begin by finding the repetitive questions that slow your approval process down:
- What is waiting for me?
- Which policy applies?
- Where has this request reached?
- Who approved it, and when?
Those are useful first tasks for AI assistance. Approval rights, financial commitments and workflow changes should remain with named people and agreed controls.
If you have one approval process causing delays, Techspire IT can help you turn it into a focused Claromentis pilot.
Use AI for answers. Keep people responsible for decisions.
A sensible first AI use case is not “approve this request for me.” It is helping the people in your process get the context they need before they decide.
For example, AI may help someone find a policy, locate a request, explain its current stage or retrieve its history. Your approval process should still make clear:
- who is allowed to approve;
- what they are approving;
- which limits and controls apply; and
- how the final decision is recorded.
That is the starting point for a controlled AI rollout: faster access to information, without handing over financial authority.
Human in the loop AI: keep these decisions with a named person
AI can make an approval process easier to navigate. It should not become a shortcut around the people and controls your organisation relies on.
Keep these actions with an authorised person:
- approving or rejecting spend;
- changing a request’s status;
- creating a financial commitment;
- overriding an approval route;
- deciding when an exception is justified.
The value of an AI pilot is not removing accountability. It is reducing the time people spend gathering information before they exercise it.
Interactive: Agent or person?
Sort each step: an agent or a person
Your sort
The line is short. The agent can find, answer and report. It cannot sign.
An agent can help
A person decides
Nothing is stored or sent.
Where AI can help before an approval
| Question from the business | Useful role for AI | Decision that stays human |
|---|---|---|
| “What policy or limit applies?” | Help people find the relevant guidance and source material. | Maintaining policy and deciding how it applies to an exception. |
| “What is waiting for my approval?” | Help an approver locate the information they need to review. | Approving, rejecting or escalating the request. |
| “Where has this request reached?” | Help an authorised person understand the current position. | Changing the request or workflow status. |
| “Who approved this, and when?” | Help retrieve the relevant history for review. | Relying on the organisation’s formal record and deciding what happens next. |
The exact scope should be agreed, configured and tested before it is used in a live process. Start with information access; do not assume an AI assistant should have authority to change outcomes.
Inside Claromentis: when we asked the agent to approve a request
A person asks who approved a request
A person asks, in the AI AssistantSample data
“Who approved PO0019, and when? Show each step”
The answer, the request's history, and the refusal
CheckedThe form's own list still shows PO0018 as Awaiting Management Approval, with nobody assigned.
What happened, step by step
On our demo portal, a person asks the AI Assistant who approved a purchase request.The answer names who approved it, and when.And it shows the request's history, step by step.Then the same person asks it to approve a different request.It says it cannot approve, and which status the request is in.On our own demo portal, an agent we built to read purchase requests, asked to approve one, declined, and the request stayed unchanged.
What a safe first pilot should prove
A first pilot should answer one business question: can we reduce approval chasing without weakening control?
Choose one process and agree the boundaries before configuration begins:
- The questions AI may help answer.
- The information it may access.
- The actions it must never take.
- The people who remain accountable for decisions.
- The evidence your team needs before expanding the pilot.
Do not try to automate every approval process at once. A small, controlled pilot gives Finance, Operations and IT a shared way to assess the opportunity before making a wider commitment.
Three questions before you add AI to a workflow
Before including AI in any approval process, ask:
- Is this an information task or a decision that changes an outcome?
- Could this action commit money, create an obligation or bypass a control?
- Is there a named person who remains responsible for the final action?
If the answer is unclear, do not automate the step. Start with the questions people repeatedly ask instead.
Turn one approval bottleneck into a controlled Claromentis use case
The best first use case is usually familiar: an approval process where people lose time finding policy, status or history before they can act.
In a scoping session, we will help you map:
- the current approval journey;
- the questions causing delay;
- the people who need access to information;
- the decisions that must remain human; and
- the smallest pilot worth testing.
The outcome is a practical recommendation: proceed with a defined pilot, improve the underlying process first, or leave AI out of that workflow for now.
Start with one process, not an AI programme
You do not need to redesign every workflow before exploring AI. Bring one approval process that is creating friction today.
Good starting points often involve repeated questions about:
- policy and approval limits;
- request status;
- missing information;
- approval history; or
- the next person responsible for action.
We will help you decide whether that process is suitable for a controlled pilot and what must be in place before it goes live.
Frequently asked questions
Can we use AI around approvals without giving it approval rights?
Yes. That is human in the loop AI: start by limiting the pilot to information tasks. Agree the boundaries, test them in your environment and keep approval decisions with nominated people.
Which approval process should we start with?
Choose a frequent process where people spend too much time finding policy, status or history, but where approval roles are already clear.
What should we bring to a scoping call?
Bring one real approval process, the people involved and the point at which work slows down. We will help you identify a sensible first scope.
Do we need to commit to a full AI rollout?
No. The useful next step is a defined conversation about one process and whether a controlled pilot is justified.
Find out whether your approval process is a good first AI pilot
Bring us one approval process that is slow, unclear or overloaded with status-chasing.
In a focused session, we will help you identify:
- where people are losing time;
- what information AI could help surface;
- which actions must remain with people; and
- whether a Claromentis pilot is the right next step.
You will leave with a clearer first-use-case recommendation, not a generic AI pitch.
Prefer to discuss your situation first? Contact Techspire IT.
Not sure approvals are the right place to start?
If your team is also exploring AI for internal knowledge, employee questions or process automation, tell us what is creating the most friction. We will help you identify the use case that is most practical to test first.




