BACKGROUND
Hyprbots is a fintech startup focused on reducing manual work across invoices, accruals, and procurement to improve efficiency for accountants and CFOs.
01
Procurement Feature
PROBLEM
Procurement workflows in small and mid-sized businesses are slow, error-prone, and heavily manual; this requires accountants to reconcile multiple financial documents before taking action.
DESIGN BRIEF
How might we facilitate procurement reviews to help accountants identify and reconcile errors in the review process efficiently?
RESEARCH
I studied Hyprbots' approach to improving the review process: surfacing AI-parsed data to indicate errors and mismatches
Through shadowing calls with Hyprbots' CFO partners and working closely with senior PMs and design leadership, the core friction was identified as cognitive overload during the review phase and not the matching itself, which the AI handled
As the firm created AI co-pilots to automate procurement workflows, accountants still needed to review and validate AI-generated data.
SCOPE
Introducing the human review layer within the workflow (human-in-the-loop approach)
I focused on designing the human review layer within an automated three-way matching workflow — specifically how accountants scan, validate, and correct procurement data before taking action.
IDEATION
The real design challenge: AI trust in financial decision-making scenarios
When the system flagged errors after three-way matching, the workflow broke down in a predictable way. Junior accountants would scan the error list, fix the obvious mismatches, and pass the document up. Senior accountants would open it to be met with a full list of unresolved flags and panic: they couldn’t tell which errors had been addressed and which ones needed further digging.
The interface made human judgement invisible. This drove the design decisions.
Error states indicated with accordions open by default
If AI flagged a mismatch, it had to be the first thing the accountant saw and not something they had to hunt for. Collapsed error states put the burden of discovery on the user, which led to errors being missed in a hectic workflow. The relevant accordions being open by default made the AI's judgment immediately visible and actionable.
Including an acknowledgement and action section
This was the critical trust mechanism. Accountants needed a way to say: “I've seen this, I've assessed it, I'm choosing to proceed.” Without it, every unresolved lower-priority error looked identical to an unreviewed one, which caused senior accountants to panic.
The 'Acknowledge' interaction created a visible layer of human judgment on top of AI output, making the handoff between junior and senior reviewers legible for the first time.
Prioritizing high stakes fields while retaining other details
Working with backend engineers, I identified which fields appeared consistently as fields coming from the ERP system the platform connected to. GL fields and tax lines were brought to the top since errors in those fields carry the most downstream financial consequences.
This was the first time field priority had been explicitly mapped to ERP consequence rather than visual convention.
OUTCOME
Two-pane layout which includes a document drawer as well as key finance information with error indicators to help accountants scan and take action to rectify procurement errors
The procurement review screen gave accountants a structured way to verify AI-parsed data, surface critical errors first, and take role-appropriate action, reducing the ambiguity that previously caused errors to slip through between junior and senior reviewers.
IMPACT
$9M
raised through the seed funding & series A supported by MVP demos including this key feature shipped in early 2025.
