Tinu AI: financial memory that continues

The Problem

Finance teams already have all the data, but not the continuity.

When analysts leave, so does the context in their heads. Even with hundreds of documents, teams lack the connective layer that explains why a number changed or how a decision was made.

This leads to rework, delays, and repetitive “Can you remind me what happened last quarter?” cycles.

Testimonial from a VP of Product at a fintech startup: “Quarterly book ends are a huge pain in the a**. I just want a tool that keeps the context so we stop wasting hours redoing work.”

And the data backs her up — Research shows that missing context and repeated rework impose a significant cost on large finance teams.

Two statistics on a purple-to-teal gradient. Left: “$12M — Avg annual cost of decision delays in large finance teams caused by missing context and repetitive rework. Gartner 2023.” Right: “68% — of financial analysts report spending more time hunting for explanations rather than performing analysis. PwC 2023.”

To better understand the problem space and its impact on our primary audience, we interviewed professionals whose roles include financial planning and analysis responsibilities, as well as leaders who oversee financial planning and decision-making.

The Opportunity

Why does critical context fail to turn into action?

The real problem isn’t just that context lives in people’s heads. It’s that context decays at three specific points:

Capture FailureContext never gets written down or gets written down too late.
Structure FailureEven when context is written, it’s unstructured, unorganized, or difficult to reuse.
Retrieval FailureEven documented context can’t be found at moment decisions need to be made.

Context decay pipeline in five steps, left to right. 1. Context from decision in head: a decision is discussed verbally, including assumptions and tradeoffs, but nothing is written down in the moment. 2. Written late or not at all: the explanation is written days later from memory, after details and rationale have already faded or shifted. 3. Notes are fragmented or unorganized: pieces of context live in Slack messages, spreadsheet comments, and scattered docs, none linked together. 4. Unsearchable docs: when someone searches for past rationale, nothing relevant surfaces, forcing teams to guess where explanations might live. 5. Reconstruction work: time is spent reconstructing the past instead of confidently moving forward.

The opportunity comes when we can create a system that preserves context across these points of failure.

The Solution

Tinu AI: Financial Memory That Continues

Tinu AI functions as a shared memory layer for finance teams, ensuring decision context survives capture, structure, and retrieval and remains usable in day-to-day workflows.

Tinu AI “Welcome, Kaitlyn” screen with a New Context Entry form. The journal questions, created by Kaitlyn J., ask “Was there a moment you weren’t sure who owned a financial decision? (For example: approving a vendor invoice, reallocating a team budget, or validating a forecast number?)” and “What’s one financial insight, learning, or decision takeaway you want future you (or your team) to remember?” Text beside it: “Tinu AI prompts employees to explicitly write down decision context …while details are still fresh.” — Structured context capture (journal entry)

Tinu AI “Context Entries” page listing Previous AI Summarized Entries. Each entry has a date, an ID (CE-0488, CE-0487, CE-0486), a sentiment tag (positive, neutral, negative), a summary, linked files and stakeholders. Text beside it: “AI organizes raw reflections into a consistent format allowing context to be compared, searched, and reused across people and time.” — AI structuring & summarization

Tinu AI chat panel open over the Context Entries page. It answers with “Key Financial Metrics – Adyen N.V. (H1 2025)”, listing net revenue, processed payment volume and EBITDA with cited sources, above an “Ask Tinu AI…” input. Text beside it: “Employees retrieve past context by asking questions, rather than guessing where explanations might live.” — Context retrieval by intent (chat)

Excel spreadsheet of November 2024 actuals and variances with the Tinu AI for Excel panel open, showing “Analyzing document… Searching through ERP and CEs…” Caption: “Tinu AI is auto-detecting your file…” — In-workflow action (plugin)

Tinu AI panel shows “November Accruals Detected”: 12 recurring accruals expected, 3 show unusual patterns vs Oct, rent typically $25,000 since Sept, insurance due for annual increase, similar to CE-0488. Buttons: “Auto-fill from October” and “Review Changes”. The empty Nov_2024_Actual cells are highlighted. Caption: “File detected — ‘2025 November Accruals’ — AI insights summary generated.” — In-workflow action (plugin)

By preserving context across capture, structure, retrieval, and action, Tinu AI turns everyday financial decisions into durable organizational memory.

Impact

Tinu AI transforms fragmented financial work into a searchable, scalable memory layer.

Organizations

Knowledge becomes infrastructure, not tribal memory
Lower turnover cost
Faster analysis cycles

Teams

No more lost explanations
Faster onboarding
Seamless handoff when roles shift

Managers

Centralized visibility
Higher-quality weekly updates
Less back-and-forth

We also get a context reinforcement loop within the Tinu AI ecosystem that grows with usage.

Tinu AI builds a continuous context loop. A loop of five stages with arrows between them, the last feeding back into the first. Employees add new context entries manually. AI structures and integrates content. Managers gain visibility. The organization’s memory grows. Employees retrieve past context to support decisions, validate work, and move faster.

Over time, TinuAI shifts critical context from individual memory into shared infrastructure, making decision-making more durable as teams and roles change.

Iteration #1

What if people are hesitant to share their context?

Not everyone wants to share the context that makes them indispensable. For some, undocumented knowledge feels like job security, which can reduce participation.

Moving forward, this raises an important design consideration: this product must deepen trust, clarify value for contributors, and create incentives that make context-sharing feel safe, fair, and mutually beneficial.

Here are some possible solutions we considered:

Framed close-up of a New Context Entry form with a “Private” lock badge next to the title, journal questions created by Kaitlyn J. Panel label: “Context entries are automatically set to private to build trust.”

Private-to-public preview:

Contributions start private and users can choose what gets shared.

Framed close-up of the Tinu AI header with a notification that reads “Your entries helped 4 teammates complete tasks faster this week,” above a context entry tagged positive. Panel label: “Make it obvious how your context inputs help move work forward.”

Impact Feed:

We surface the impact of each contribution, making it clear how shared context drives work forward.

Instead of framing context-sharing as giving something up, we want to design the product to make it high-leverage for the individual employee.

Iteration #2

What if employees input low-quality context?

Not all context is equally useful.

In the moment, employees may log vague or incomplete context entries—especially when time is limited or stakes are unclear. Low-quality inputs can dilute shared knowledge and reduce trust in the system.

Here we can see an example of a low-quality vs high-quality context entry.

Two versions of context entry CE-0488 compared side by side. Low-quality, marked with a red X: “Vendor approval issue. Got stuck on a vendor invoice again. Wasn’t clear who needed to approve it, but we sorted it out eventually.” It lacks a decision owner, an outcome, guidance for next time, and any signal of why this mattered. High-quality, marked with a green check: “Vendor invoice delayed due to unclear approval owner. Invoice was delayed because it wasn’t clear who owned approval. Finance ended up approving after confirming it was over $25k. For future invoices over $25k, route directly to Finance AP.” It has clear ownership, decision and rationale, outcome, and reusable guidance for future scenarios.

One of the possible solutions we considered:

The “Welcome, Kaitlyn” New Context Entry form, marked Private, with “Journal questions created by Kaitlyn J.” Prompts include “Was there a moment you weren’t sure who owned a financial decision?” and “What was the outcome, and what should someone do next time to make the process more efficient?” An annotation points to the byline: “Managers can choose to create guided prompts to elicit high quality context entries.”

Manager-created prompts provide structure at the moment of capture:

By defining specific questions, managers help standardize what kind of context matters while leaving employees free to answer in their own words.

And by standardizing what information gets captured, these prompts help produce consistent, high-quality context entries that can be reused over time.

Reflection

What I Learned

Context loss is a systems problem, not a people problem.

I learned that missing context isn’t caused by people failing to document. It’s caused by systems that don’t support capture, structure, and retrieval of that context at the right moments.

Early on, it was pretty tempting to design a documentation app enhanced with AI capabilities to predict what the lost context might’ve been. However, we found through our interviews that the pain point extended beyond just one workflow.

Reframing the problem as context decay across a system led to a more effective solution: a cohesive memory layer on top of your pre-existing workflow.

Collage of six team photos. Three are video-call screenshots and casual study-room selfies from team meetings. Two show the five teammates in black suits on pitch day, posed in a sunlit atrium and on a staircase.

See our pitch deck(opens in a new tab)

Among 72 teams, Tinu AI earned a top-three spot in the Fintech track at the Tech Innovation Jam.

Thanks for reading!

Next Case Study

Designing a personalized learning system for soft skills

Mobile
B2C
Design Competition