Project Delta Digest in Microsoft 365 Copilot - A Status Report Agent Worth Building
Project status reports are the kind of work that nobody wants to do but everybody wants to read. The program manager spends Friday afternoon trawling through Teams chats, Outlook threads, SharePoint files and meeting notes to produce a one-page summary that the executive sponsor will glance at for 30 seconds. By Monday the report is stale. By Wednesday someone is asking the program manager to produce another one.
Microsoft's Project Delta Digest template for Copilot agents is aimed squarely at this problem. The template is a starting blueprint for building an agent that scans Microsoft 365 sources, pulls together what happened, and produces a structured digest with sections like TL;DR, shipped items, blockers, risks, and team load.
We've built three of these for clients in the last two months. Here's what we've learned about when it's worth the effort and when it's not.
What the template actually gives you
Out of the box, the Project Delta Digest template gives you an agent persona, an instruction set, a section framework, and prompts for the common digest types (daily change digest, this week's wins, current risks, leadership weekly recap, stakeholder update). It's not magic. It's a well-thought-out scaffold that saves you from having to design the digest structure from scratch.
The 12-section framework covers:
- TL;DR
- Shipped items
- In-progress work
- Upcoming milestones
- Recent activity
- Blockers
- Team load
- Risks
- Discussions
- Files
- Delivery signals
- Reference documents
That's the spine. The instructions tell the agent to assign a 1-to-5 risk score, to detect stale items, to adapt detail level by audience, and to offer draft Teams messages or follow-up emails based on what it finds.
The honest assessment is that the template by itself, with no grounding, produces fairly bland generic-sounding output. The value of the template is in the structure it imposes once you wire it up to real data. Don't deploy this thinking the template alone will magically produce a useful digest. It won't. You need to ground it.
Grounding is where the work is
Microsoft says the template "works best when grounded in additional context." That's a polite way of saying "this is useless until you connect it to your actual project data."
The realistic grounding options for an Australian business are:
A SharePoint site for the project. This is the easy one. If you've got a project site with files, status notes, decision logs and meeting minutes, you can point the agent at it and immediately get reasonable summaries of what's documented. The problem is that most teams don't keep their project sites that up to date. We did a digest agent for a financial services client where the project site had files from three sprints ago and nothing current. The agent produced a great summary of three-sprint-old work and missed everything important.
Copilot connectors for the source-of-truth system. This is where the value lives but also where the effort lives. The Azure DevOps connector, Jira connector and ServiceNow connector pull in actual work item state, assignment, and status changes. Combined with the digest agent, you get summaries grounded in current ticket data instead of in stale documentation. This is the configuration we'd recommend if you have the budget for it.
Microsoft Graph for email and chat. The agent can pull from Outlook threads and Teams chats associated with project participants. This is where the "risk" and "blocker" detection actually shines, because most of the real risk signals on a project are buried in chat threads, not in formal status reports. The downside is that you need permissions and policies sorted out before you let an agent loose on people's inboxes.
We typically combine all three for a serious deployment. SharePoint for documents, a Copilot connector for the work tracking system, and Graph for chat and email signals. Each one alone is okay. Together they produce a digest that catches things the program manager would have missed.
Where it's working well in client engagements
The use case we've had the most success with is the leadership weekly recap. A senior executive doesn't want item-level detail. They want TL;DR, milestone progress, top three risks, and a momentum signal. The template's audience-adaptive output handles this exactly the way you'd want.
We built one for a manufacturing operations client where the GM was burning two hours a week composing the executive update. The agent now produces a draft of that update by 8am Monday, sourced from the prior week's Azure DevOps state, Teams discussions, and SharePoint documents. The GM still reviews and edits, but the work has gone from 2 hours to about 20 minutes.
The other use case worth highlighting is the daily change digest for a delivery team. Project leads use it to figure out what shifted overnight, what's newly blocked, and who needs help. We had a software delivery team using this where the team lead was scrolling through six Teams channels every morning. Now they read a 5-minute digest and triage from there. It hasn't replaced the standup. It's made the standup shorter.
For our own agentic automation work with clients, the Project Delta Digest pattern is one we now reach for early in engagement scoping. It's a fast win that shows the client value before you start the harder integration work.
Where it falls down
The template has limitations, and Microsoft has been honest about them. The bits worth flagging from real-world use:
It's not multi-turn aware. The template is designed for one question at a time. If a user asks "give me the digest, then draft a stakeholder email based on what you found" it will sometimes do one and not the other. We've trained users to ask in two separate prompts, but it's friction.
The risk scoring is opinionated and not always right. The template assigns a 1-to-5 risk score based on priority, age and impact. That's reasonable as a default but it bears no relation to whatever risk register your project actually uses. We've had a few clients where executives push back on the auto-assigned scores. The fix is to either edit the instructions to align with the client's risk framework, or to remove the auto-scoring and present items without scores. Both work. Just know going in that the default scoring will not match your governance model.
The "team load" section is hit or miss. The agent tries to infer how loaded each team member is based on assigned items and activity signals. The data quality varies wildly. If your team uses Azure DevOps rigorously, it's decent. If half the work is invisible in tickets, the team load reporting will be misleading. Don't rely on this section for actual capacity planning.
The drafted Teams messages and emails are formatted in a Copilot-ish way that sometimes reads as artificial. We've ended up adjusting the instructions to be more terse and to match each client's communication tone. The default is too smooth. People notice.
Customising the section framework
The most common request we get after a first deployment is to add or remove sections. The 12-section default covers most projects, but specific industries want different things.
A financial services client wanted SLA breach counts and compliance flags. We added a dedicated section to the agent instructions and grounded it in their incident tracker.
A construction client wanted weather delay tracking and equipment availability. Same approach. Edit the instructions, add the section, ground it in the source.
A consulting client wanted hours-burned-versus-budget per workstream. This one was harder because the data sat in Xero rather than Microsoft 365. We added a custom API plugin to pull the numbers in, which is supported but adds complexity.
The lesson is that the template is genuinely extensible. The default 12 sections are a starting point, not a constraint. We've never delivered a digest agent that uses the exact default structure. Every client wants something different in there.
The compliance conversation you need to have
A Copilot agent that scans email and chat is going to touch private data. This is not optional and you cannot skip it. Before you deploy a Project Delta Digest agent in production, you need to have the conversation about:
Who is the agent running as? Is it running with the prompter's permissions, or with a service account's permissions? The answer affects what data it can see.
What data classification scheme is in place, and does the agent respect it? If your tenant uses sensitivity labels, the agent will inherit access controls but it can still summarise content that someone with broader access shouldn't be redistributing to a wider audience.
Where do the digests get stored? If the digest output is saved into a SharePoint site that's more broadly accessible than the source data, you've created a privacy leak.
We've had two clients where the compliance review took longer than the agent build. That's the right ratio. For regulated industries especially, this conversation needs to happen before you build, not after.
Whether it's worth building
If your organisation has a project management function that's spending real time on status reports, and you've got the Microsoft 365 Copilot licensing in place, the Project Delta Digest template is one of the better starting points for a useful internal agent. The structure is sound, the use cases are well-scoped, and the wins are real if you ground it properly.
If your projects don't live in Microsoft 365, or if your data is scattered across systems that don't have connectors, the value drops fast. You'd be better off building something custom against the source systems directly.
Our take on Microsoft 365 Copilot consulting work for Australian businesses is that this template is the highest-value starting agent we've found in the catalogue. Start here, measure the time saved, then expand into more specialised agents once people trust the pattern.