A RAID log for the AI native era

The standard process for managing RAIDs is old-school and manual, taking place in spreadsheets or SharePoint lists. This turns overwhelmed PMs into bottlenecks, since the entire team has to wait for them to review and update the log to get latest picture of how their project is doing.

To help empower our customers as they manage their RAIDs, I kicked things off on the project by doing a deep dive into existing RAID logging solutions. This helped me understand how inefficient current processes are, and how much opportunity there is to automate this work with agents. I then dove deeper into RAID management best practices, researched documentation, and learned how existing industry standard tools like SAP handle RAID management.

Anatomy of a RAID log

A RAID log is an interconnected set of Risks, Action Items, Issues, and Decisions. It documents the status of a project as it evolves, identifying threats to the critical path so they can be resolved.

Large digital transformation programs, can span three to five years. Cross project dependencies, large teams, and even larger budgets make a RAID log critical to success but exponential in complexity.

If the team is unable to adequately address a Risk as it intensifies, it will be promoted to an Issue. This means it is actively driving the project off the critical path.

By modelling this lifecycle and the associated relationships, we were armed with a framework to build a robust RAID log, and a set of intuitive human-in-the-loop workflows including:

• Status evolution based on relationships
• Risk promotion to Issue
• Automated follow-ups and escalation

Extraction and human-in-the-loop

RAIDs appear in meetings, chat messages, emails, reports, and any other surface where people interact. Their evolution is long and rarely clear-cut, and their status is implicitly updated in conversations. This complexity is an opportunity for Tato to put its deep project knowledge, integrations, and agents to work to extract, monitor, and document RAID items.

Meetings are the primary source of RAID item context, but the items themselves are managed in Jira, SharePoint, or plain old spreadsheets. Our first job was consolidating those sources into a normalized log that could act as a single source of truth.

Tato updates the log after any activity, and pulls from external tools periodically or on command through MCPs and bespoke integrations.

Initially these updates were exposed as simple sections showing the under-the-hood thinking of our agent. Exposing that machinery almost always overloads the user, because despite our best efforts LLMs produce volume and lead to cognitive load. I hid that machinery to cut a lot of noise, since a single meeting could produce upwards of 60 snippets for a handful of RAID items.

Then I reduced the amount of text and added icons and avatars, to improve legibility and hierarchy and make the list easier to parse.

0:00:00 / 1:04:19

1x

Risks

Recorder can't join authenticated meetingsNewE

Integration approval adds two weeks to the planNewM

Data migration sign-off still pending finance reviewUpdateR

Action Items

Confirm who owns the employee master recordNewD

Share the localisation gap list with the workstreamUpdateN

Submit the ticket to activate scope item J78NewA

Create users and replicate roles in the sandboxUpdateN

Send starter system access links to the new teamNewM

Book the enablement session for the finance leadsNewF

Draft the rollback plan before the next cycleUpdateD

Issues

Staging data is out of sync with productionNewR

Report format differs across workstreamsUpdateA

Sandbox refresh wiped the test usersNewN

Decisions

Roll out in phases, region by regionNewE

Standardise on one report formatUpdateF

Keep the legacy interface until the Q4 cutoverUpdateM

BeforeAfter

I exposed two primary actions on each item. This is where the human can review the agent’s assumptions and trigger additional workflows by tracking or dismissing.

Dismissing an item tucks it away and changes its status, letting the agent know that this extraction was not useful. Tracking an item marks it as exceptionally relevant and updates its status. This tells our agent to pay special attention to it moving forward.

To add delight I explored a subtle interaction to make this key action feel impactful and energetic.

Early feedback showed a perception that Tato was smarter and more nuanced, and that the system was easier to control. We've found our users' perception of quality to be as important as the quantitative improvements when it comes to interacting with an LLM.

As a final improvement to this workflow I reworked the entire summary page. A two-column layout gives RAID management its own first-class surface and keeps meeting context in view while the user grooms the outputs.

Share

Scope confirmation workshop

Projects

Phase 2 rollout

Date

Thursday, Sep 3, 2026 at 4:30 PM

Attendees

DNFA+1Programme lead, Integration lead, Finance workstream, Data migration

0:28:12 / 1:04:19

1x

DProgramme lead45%

NIntegration lead45%

FFinance workstream9%

AData migration1%

Skills

Meeting follow-up

Extracts next steps, assigns action items per participant, and proposes follow-up slots.

Recent meeting highlights

Open action items

SummaryRAID7ScopeTranscript

Risks

2

Recorder can't join authenticated meetingsNewE

Integration approval adds two weeks to the planNewM

Action Items

3

Confirm who owns the employee master recordNewD

Share the localisation gap list with the workstreamUpdateN

Submit the ticket to activate scope item J78NewA

Issues

1

Staging data is out of sync with productionNewR

Decisions

1

Roll out in phases, region by regionNewE

An auditable paper trail

A well managed RAID log is an accurate ledger of a project's evolution. This keeps teams and our agents accountable. Our history view was hard to reach, and focused purely on versions.

The improved history documents which changes were made by people and the agent and the reasoning behind them. It exposes related activities and reasoning, so the human can understand updates and correct them when necessary.

I tried several iterations for the history and landed on a drawer with a clickable timeline. The drawer offers the most real estate for understanding a risk and its history while letting the user stay in their current workflow, with the option to open a full page if necessary.

The ultimate control centre

The last piece of this project was an improved control centre. Much of the work of RAID management happens in the context of an activity summary, but we wanted a delightful primary surface for managing the full log.

I started with basic UX improvements: tightening space for more breathing room, grouping related controls, making the table contents scrollable rather than the page so the header, footer, and every available action stay visible. I ensured the table had robust and intuitive keyboard navigation, including the skip to main content link, row navigation with arrow keys, proper focus management, and more.

Finally, I built a collection of indicators to show impact, likelihood, and status. Each animation is driven by a single value so it springs smoothly between any two values, and supports additional intermediate states.

Impact and adoption

Although our user base is small, we've seen strong signals of adoption and value, including a 5x increase in items logged via chat in the 4 weeks since launch. Chat is one of the main ways users can generate new RAIDs and import existing items via MCP or document uploads.

We've also received positive qualitative feedback from early users, including:

This RAID log is one small step toward a future where humans and AI collaborate to ensure successful business transformations. The next step is to build upon our learnings and double down on clever automations that help large teams close the loop on their RAIDs.