Feedzai builds real-time fraud and financial crime prevention for banks and payment firms. Augusta Labs rebuilt two of its expert workflows, NDA review in Legal and client QBR preparation in Revenue, on one platform inside Feedzai's own cloud.
Feedzai's FY26 plan moves the company from copilots to agentic workflows, with a person approving what agents produce. Contract redlining and QBR decks were both on its internal AI wishlist, with no tool assigned to either.
Legal reviews every NDA that Feedzai's commercial teams send or receive. Revenue prepares a quarterly business review for each of its roughly 100 clients. Both depended on a few experts and on data held by other teams.
Each counterparty NDA was opened beside the template, the guidelines and a personal file of notes. The rules were applied from memory, and negotiation context was copied in from email.
Account teams copied last quarter's deck, chased six or seven systems for data and waited on a risk consultant to pull each client's use-case metrics.
The starting point, as the Legal and Revenue teams described it during discovery.
All first-pass reviews sat with a single lawyer, with no structured backup.
Most of it spent gathering data, not building slides.
Use-case metrics came from a risk consultant, fitted around other work.
From the CRM and data warehouse to support, project and HR tools.
Two agents on one platform, built in weekly feedback loops with the lawyers and account teams who use them, and deployed inside Feedzai's own cloud.
Rebuild first-pass NDA review and client QBR preparation as AI-native workflows, and run both inside Feedzai's cloud on foundations that later workflows can reuse.
An NDA review agent that redlines each contract against Feedzai's own rules, and a QBR generator that drafts each client's quarterly review from Feedzai's data.
Each workflow was split into steps with their own inputs, rules and checks. A person reviews every output before it is used.
On written request, the Receiving Party shall return or destroy all Confidential Information and certify the destruction in writing, save for copies kept in routine backups.
These obligations survive for five three years, with no limit for trade secrets and personal data.
Splits the NDA into clauses, compares each with Feedzai's template and flags missing clauses.
A focused reviewer per clause sees only that clause and the guidelines that apply to it.
Code drops any redline that cites no registered rule or rewrites more than half a clause.
The lawyer accepts, rewrites or rejects. Each decision is stored and recalled on similar clauses.
The account team picks the client, quarter, use cases and language in four screens.
24 queries maintained by Feedzai's data team pull the client's metrics from the data warehouse.
Fixed rules pick the headline metrics. The AI writes the narrative and never a number.
The draft opens as a slide deck with last quarter carried forward, ready for the team to edit.
Both agents run on one platform inside Feedzai. The next workflow starts from these parts.
Three phases from discovery to handover, with a status review every week.
NDA rules lived in one lawyer's files, and nothing linked a QBR slide to its data.
Both workflows mapped step by step, with the architecture agreed with IT Security.
The first prototype struck whole clauses, and the first QBR flow needed a guide.
Redlines change only what must change, and the QBR flow was redesigned around its users.
Both agents ran outside Feedzai, on redacted contracts and generated figures.
Lawyers review real NDAs in Feedzai's cloud, and QBR figures were checked cell by cell.
What Feedzai kept at handover: two working agents in its own cloud, the full source code and the documentation its engineers need to run them.
Everything built during the engagement belongs to Feedzai.
Each result is set against the baseline in section 01. Both agents now work on real NDAs and live client data inside Feedzai's cloud, and a person still approves every output.
One request builds the full deck from Feedzai's data, ready for the account team to edit.
Metrics come straight from the data warehouse for every use case with live data.
Each redline cites the Feedzai rule behind it, and uncited suggestions are dropped.
This is where we really see Augusta's difference. I know everything works in PowerPoint. But then you have to make it work in practice.
Delete return, keep destruction with written certification.
Three years, unlimited for trade secrets.