On this page

Suggestions & Order: analysis and steering

In short

TensorPM gives you a second opinion on your project. It works from what is confirmed in the project context: goals, scope, dates, budget, risks, action items, responsibilities. Out of that it produces two things.

  • Suggestions. Short cards in the Pulse view that name what stands out, why it matters, and what you could do about it.
  • Order. A tab inside the Action Items view that puts the work into a recommended sequence along the critical path.

There is no separate Guidance view in the sidebar any more. Its two halves moved to the places where you already work: the suggestions to the dashboard you open anyway, the sequence next to the action items it talks about.

Neither replaces your judgement. They make sure you decide against an ordered picture of the project instead of a gut feeling.

When you need this

Typical moments from a working week:

  • Monday morning, and you want to know where this week's risk sits.
  • Before a client meeting or a steering committee, when you need three defensible statements instead of twenty impressions.
  • After a week away, with a lot of mail that arrived while you were gone.
  • When the project feels restless and you cannot name the reason.
  • Before you release the next phase and need to know whether the planned work covers the goals.

A bad moment is day one of an empty project. Without content there is nothing to analyze.

The Suggestions card in Pulse

Open your project and click Pulse in the sidebar. Its tooltip reads What needs your attention today?. The Suggestions card sits at the top of the status column on the right of the dashboard, above health, progress and metrics. On a narrow window that column moves below the agent area instead of beside it.

The card header shows the number of open suggestions and, next to it, a {count} new badge. "New" is counted per device against the last time you looked at the card, and looking at it is enough: moving the mouse onto the card marks its contents as seen.

What a suggestion says

Every suggestion is three short paragraphs in a fixed order, without headings:

  1. The observation: the plain finding, for example "13 action items are overdue".
  2. The impact: why it matters, for example "Schedule slip threatens delivery and project credibility".
  3. The suggestion: the next step, for example "Review deadlines and set realistic dates".

The colour of the row follows the severity. Read the strongly coloured ones first; that is what the colour is for.

Filters and order

Under the header sit filter chips, one per category that currently has an open suggestion, each with a count: Context, Strategy, Coverage, Execution. A chip you do not see has nothing open behind it. Click a chip to narrow the list, click it again to clear the filter. A single suggestion can belong to more than one category.

The list itself is not grouped by category. It is sorted by weight: how severe the finding is, whether it can be applied directly, whether it is tied to a goal or a success criterion, and how old it is. Older suggestions sink slowly. The order is frozen while you are on the project, so the list does not reshuffle under your cursor.

Three suggestions are shown at a time. Show {count} more opens the rest, Show less folds them back.

Apply, Discuss in chat, Dismiss

Each row carries icon buttons on the right. Hover for the label.

  • Apply only appears on suggestions that carry a concrete change to an action item. One click creates, edits, or removes that item. Everything the suggestion sets lands in the item's AI+ priority slot and never overwrites a value you set by hand. A suggestion that is only prose has no Apply button, on purpose.
  • Discuss in chat opens the project agent with a prepared message containing the observation and the recommendation. The suggestion keeps its status: talking about something is not answering it, so it stays open until you apply or dismiss it.
  • Dismiss asks twice. The first click changes the label to Click again to dismiss for about three seconds; a second click inside that window removes the suggestion from the open list.

Applied and dismissed suggestions are not gone. Below the list, {count} already answered opens the history of the ten most recent ones, each with a Bring back button that reopens it.

When the card is missing or empty

  • No card at all: the project has never produced a suggestion. That is the normal state of a fresh project, and TensorPM deliberately does not show an empty box for it. Run a project review first, see below.
  • No open suggestions. — the card exists, you have answered everything. That is the goal.
  • No suggestions in this selection. — a filter chip is active and its category is empty. Click the chip again.

The Order tab in the Action Items view

Click Actions in the sidebar. The view is titled Action Items and carries a tab bar with List, Board, Order, and Recurring. Order is the sequence view. The dropdown next to the tabs switches the board between Kanban and Priority and does nothing for Order.

What it shows

The heading inside the tab reads Critical path, with the subtitle "Cards in the recommended execution order, focused on the critical path, dependencies and parallel opportunities".

The work is grouped into collapsible steps, each headed Step 1, Step 2, and so on, optionally with a group name and an estimated duration. A step that can run alongside the previous one is marked Can run in parallel. The icon in the step header summarises the state of the items inside it: blocked, in progress, or fully completed. Some steps carry a short rationale explaining why they sit where they sit. Which steps you leave folded is remembered per project.

The cards themselves are your normal action items with extra badges:

Badge Meaning
Critical Path this item is on the critical path; if it slips, the end date slips
{count} dep(s) it depends on other items; the tooltip names them
Blocks {count} other items wait for it; the tooltip names them

This is the same data as in List and Board. What you complete here is complete everywhere.

Read it as your scheduling logic: whatever sits in the first steps decides the end date. Make sure those items have an owner and a realistic date.

Where the order comes from

The tab computes nothing itself. It renders the result of the execution analysis, which runs as part of a project review in Pulse. That also means it does not refresh on its own. If you added a dozen items yesterday, run a review before you trust the sequence.

You do not have to enter dependencies by hand. They come out of the analysis. Items the analysis did not place in a step do not appear here.

When it stays empty

The empty state reads No order available, with the explanation that the order comes out of the project analysis in Pulse. Two things produce it:

  • no execution analysis has run yet, or
  • the project has fewer than two action items, so the analysis has nothing to sequence.

Where the analyses are triggered

All analyses now start from one place: the agent area in Pulse. The button with the magnifier icon reads Review project, and the status line next to it tells you where you stand — Project not reviewed yet, Last reviewed …, Go through the project again, Add context first, or, while it runs, Going through the project … (2/6).

One review runs several steps in sequence, and TensorPM picks them by staleness rather than running everything every time. The Details disclosure lists what it intends to do and why, with reasons such as Missing, Changed context, or Outdated. Cancel stops it.

The steps are labelled Status, Progress, Context Analysis, Strategy, Coverage Analysis, and Execution. The review is owned by the main process, so reloading the window does not interrupt it.

The four analyses

Analysis Question it answers Where the result shows up
Context Analysis Is important information missing, or does it contradict itself? suggestions with the chip Context
Strategy Do the current choices actually serve the project goal? suggestions with the chip Strategy
Coverage Analysis Does the planned work cover the goals and success criteria? suggestions with the chip Coverage
Execution What is the right order, and what sits on the critical path? the Order tab, and Execution suggestions

Prerequisites

The analyses build on each other. The context analysis is the door opener: its quality rating decides what else can run.

Analysis Requirement
Context Analysis none, it can always run
Strategy context analysis rated Decent or better
Coverage Analysis context analysis rated Decent or better, and at least one action item
Execution at least two action items and a rating of Needs improvement or better

The possible ratings are Excellent, Good, Decent, Needs improvement, and Poor.

A step whose prerequisites are not met is not an error. The review skips it and says so afterwards: "Project review complete. … was skipped — its prerequisites are not met yet." The status area shows Prerequisites missing for the same case, and No suggestions yet before the first review.

To move forward, fill the gaps the context suggestions name — usually a goal, the scope, success criteria, a timeframe, a budget, milestones, or risks — in the Context area, then review again. As the rating rises, the rest opens up.

Priorities: AI+ versus User

Every action item carries two priority values, one calculated by the AI and one you set yourself. A small badge next to the number shows which one currently applies:

Badge Origin Editable
AI+ calculated by the AI, with a rationale in the tooltip no, while AI+ is displayed
User set by you yes

Clicking the badge switches between them, and AI+ is only offered once an AI value exists. As soon as you change the number by hand, your value applies.

In practice: the AI raises "Get fire safety certificate approved" to 100 because several items depend on it and the date is approaching. But you know from a phone call this morning that the authority will not answer for four weeks. So you switch to User and set 40. The item moves back and the list reflects reality again.

The same pair exists for the project status in the health card of Pulse, where a pill toggles between User and AI+.

How this connects to the rest of the project

Suggestions and the order have no data of their own. They read and write exactly what the other views show.

Action items

The cards in Order are real action items from the Actions area. Status, priority, and time tracking are the same values. When you apply a suggestion, it becomes a new action item or a change to an existing one.

Work packages

Every action item belongs to a category. When the work breakdown structure (WBS) is active, that category is either a structure node or a work package. Every suggestion you apply therefore lands in the right place in your structure instead of floating free. Details in Work Packages & Work Breakdown Structure.

Risks

If the project has no risks or no mitigations, the context analysis reports it. Conversely, recorded risks feed into the strategic assessment. You maintain them in the Context area under Risks.

Trail

TensorPM makes no decisions and records none. What you decide because of a suggestion, you record in the Trail area under Decisions. Every change to the project appears there under Changes together with its origin.

What this is not

Four things you might expect that TensorPM deliberately does not do:

  1. Not an automation. No analysis changes your project by itself. Every suggestion needs your click.
  2. Not a decision. The card names observation, impact, and suggestion. Whether the suggestion makes sense for your building project is your call.
  3. Not a guarantee. The analysis uses a language model on the state of your project. It can miss something or overweight something. Check the evidence before you act.
  4. Not a substitute for knowledge that exists nowhere. What lives only in your head cannot be taken into account. That is exactly why maintaining the project context is not busywork.

The responsibility stays with you. TensorPM makes sure you carry it with a better view.

A practical example: an office building on Königstraße

An existing office building is being refurbished and extended by one storey. Heritage protection is involved, asbestos was found, and the client sits in the Friday meeting. The site manager steers the project in TensorPM.

Monday: taking stock

She opens Pulse. The status line says Project not reviewed yet, so she clicks Review project. The Details list shows six steps; two of them are marked as skipped afterwards, because the context is still thin.

The Suggestions card appears with two entries under the chip Context: "13 action items are overdue", impact "Schedule slip threatens delivery and project credibility", suggestion "Review deadlines and set realistic dates"; and "No recurring action items for a long-running project".

Tuesday: straightening the context

She walks through the 13 overdue items. Four were finished long ago and simply never ticked off. Five get new, realistic dates. Four stay overdue because they really are stuck, among them the question of who bears the asbestos costs.

In the Context area she adds two missing success criteria and records the asbestos issue under Risks with a mitigation. Then she clicks Review project again. This time nothing is skipped, and suggestions with the chips Strategy and Coverage appear.

Wednesday: strategy and coverage

The strategic suggestions say, in essence, that the schedule holds as long as the permits keep moving, but that the unresolved cost question can block the new storey. She takes two of them seriously and dismisses the rest. One proposes preparing a decision paper for the client; it has an Apply button, so one click turns it into the action item "Prepare decision paper on asbestos cost responsibility" under the work package "Hazardous material removal".

Under Coverage one suggestion names a gap: no action item for the sign-off of the sample bay. She uses Discuss in chat to think it through with the project agent before she decides.

Thursday: the order

In the Actions view she opens the Order tab. Step 1 holds the fire safety certificate for the timber storey, marked Critical Path and Blocks 3. Step 2 carries the variation order for the sandstone portal and is marked Can run in parallel.

The certificate card shows priority 100 with the AI+ badge. From the phone call she knows the authority will not respond before the end of the month, so she switches the badge to User and sets 40.

Friday: the client meeting

She presents three points: the open cost question with the decision paper, the gap around the sample bay, and the schedule-critical fire safety certificate. Afterwards she records the decision in the Trail area under Decisions.

The AI did not save the project. The site manager decided. But she decided against a checked picture, she found the gap around the sample bay before the client did, and the reasoning is written down in the project.

A routine for the steering meeting

  1. The day before, update blocked and overdue action items.
  2. Click Review project in Pulse and let it run.
  3. Work through the Suggestions card: apply what fits, dismiss what does not, discuss what is unclear.
  4. In the Order tab, check that the leading steps have an owner and a date.
  5. In the meeting: present observation, impact, and suggestion. Do not read the analysis aloud.
  6. After the meeting: turn agreed interventions into action items with an owner and a date.
  7. Record decisions in the Trail area under Decisions.

Frequently asked questions

Where is the Guidance view? It no longer exists. Its suggestions live in the Suggestions card in Pulse, and its critical path is the Order tab in the Action Items view.

Why do I see no Suggestions card? The project has never produced a suggestion. Click Review project in Pulse.

Do I have to run all analyses every time? No, and you do not choose. One Review project picks the steps that are stale and skips the ones whose prerequisites are not met.

Does a review cost credits? Every analysis is an AI call. Through the TensorPM provider it consumes credits. With your own provider key (Business only), billing runs through your provider. Details in Account, Subscriptions & AI Modes.

How old is the result I am looking at? The status line in Pulse shows Last reviewed …. Nothing recalculates on its own.

Can I get rid of a suggestion that does not fit? Yes, with Dismiss, confirmed by a second click. It moves to already answered, where Bring back reopens it.

Does TensorPM change anything without me? No. The analysis produces suggestions. Only Apply changes data.

Why does a recommendation differ from my own judgement? Because TensorPM only knows what is in the project. If your knowledge comes from a phone call, record it. Then the next review will take it into account.

If something does not work

Symptom Cause Fix
No Suggestions card in Pulse the project has never produced a suggestion Click Review project
No suggestions in this selection. a filter chip is active Click the active chip again
A suggestion has no Apply button it is a recommendation, not a concrete change to an action item Act on it yourself, or use Discuss in chat
The review reports skipped areas prerequisites not met: rating too low, or too few action items Close the gaps the context suggestions name, then review again
Order shows No order available no execution analysis yet, or fewer than two action items Create action items, then click Review project
The order looks outdated analyses do not run automatically Click Review project again
An error instead of a result AI provider unreachable, credits exhausted, or the request was cancelled Check the connection and your balance, then run it again. See Troubleshooting
Suggestions keep coming back the underlying gap is still open Fix the cause in the Context area instead of dismissing the suggestion

Next steps

Previous
Navigation & views
Next
Work packages & WBS
Was this page helpful?