Focus Areas

Focus areas are the specialists on the AI Strategist’s team. Each one keeps an eye on one specific thing for you - a rhythm that might break, a risk about to become real, a moment worth acting on - and speaks up only when it finds something worth your time. What it finds arrives as an insight with an explanation and a recommended action, and the important ones make it into your AI Daily Briefing.

How focus areas work

The Strategist discovers and sets up the focus areas that fit your data on its own; there is nothing you have to configure to get started. You stay in charge of the roster, and the way to manage it is to talk to the Strategist: in any strategy session you can ask what it is watching, what else it could watch, put a new analyst to work, pause or resume one, retarget one at a different process, or adjust how one behaves - it handles all of it in the conversation. For administrators, the Focus Areas page (under Administration → AI Strategist) remains available as a manual overview of everything on the roster.

Sometimes a focus area fits your data but needs one answer only you can give - which date field holds the promise you make customers, what a good month looks like in numbers, whether the companies named in one of your tables are relationships worth following, or whether a title field lists roles you actually hire. Rather than staying silent or guessing, the Strategist asks you the one question and explains why it matters, and your answer puts the analyst to work - or points it at a better target (“watch the companies in Projects instead”). Answer no and the matter is settled: it won’t ask about the same thing again.

Add a focus area yourself

You should not often need to: the Strategist staffs what your data warrants on its own. This is here for when you want something watched that it has not picked up yet, or want it pointed somewhere specific.

Just ask the Strategist. In any strategy session, tell it what you want watched - by name (“start watching commitment reliability on Order Fulfillment”) or by outcome (“keep an eye on whether we deliver when we promised”). It resolves the details from your data, plays the exact plan back to you - which analyst, which process, which fields, how often it checks - and sets it up only after you agree. Every new analyst starts in observe-first mode: it surfaces what it finds for your review and never acts on its own.

Administrators can also add one by hand from the Focus Areas page (under Administration → AI Strategist):

Select Add Focus Area.
Choose the focus area type. The catalog is grouped by category, and each entry explains what that analyst watches for.
Point it at its target. Focus areas come in three shapes:
- Whole account - some analysts watch across your entire organization, so there is no process to choose.
- One process - some do their best work on a single process; pick the one that matters.
- One process or all - the rest let you choose a single process, or leave the process blank to watch all of them.
Optionally, narrow which records it follows. The analysts that watch the companies, suppliers, customers, or roles named in one of your process tables (the news, competitor, supplier risk, customer churn, and talent market watchers) offer an Only include records that match setting. Add one or more conditions - a Status field that equals “Active”, for example - and the analyst follows only the records that match, using the same filters as your process table views. A record that stops matching drops out of its coverage on the next check, and one that starts matching joins automatically. Leave it blank and every record is followed. (You can also just describe the set you want in a strategy session - see Narrow which records an analyst follows.)
Optionally, route findings to whoever owns the record. Any analyst that watches a process offers an Also notify the people named in this field setting, listing that process’s people fields. Point it at the field that holds the account manager, the category owner, or whoever else is named on the record, and a finding about that record reaches them as well as everyone on the watcher list - so each finding goes to the person who owns it rather than to one fixed list. Leave it blank and only the watchers are told.
Optionally, give it standing guidance: seasonality you expect, relationships you already manage deliberately, expectations you hold. The analyst weighs what it finds against what you tell it. You can add the same guidance anytime just by telling the Strategist in a strategy session.

Who gets emailed about a finding

Only the people someone deliberately put on an analyst’s watcher list. Nobody is added to that list automatically, and nobody is emailed because of work they happened to do: editing a process, creating it, or being named its owner does not sign you up for mail about it.

Two settings decide who is written to, and both are somebody’s explicit choice:

The analyst’s watchers - set when the analyst is added, or anytime by telling the Strategist who should hear about its findings.
Also notify the people named in this field - routes a finding about one record to whoever is named on that record, such as its account manager (step 5 above).

Leave both empty and the analyst emails no one at all. Its findings still reach you everywhere you already look: your AI Daily Briefing, the decision journal, and any strategy session.

What happens to a finding after it is raised

A finding is not a note that sits there until someone deals with it. Every time an analyst runs, it re-checks the findings it has already raised and tells you where each one stands, so your list stays current without anyone tidying up after it.

The situation is gone: the finding closes. When the numbers behind a finding come back inside normal range, the analyst closes it and records what changed, so you can see that it ended rather than just wondering where it went.
The situation is still there: the finding stays open and refreshes. It picks up the latest measurement each time, which is what lets you see whether it is getting better or worse since you first saw it.
It could not measure this time: the finding says so and stays open. If the process an analyst watches has no process assigned to it, a field it was measuring has been removed, or there is simply too little recent activity to judge, it records that on the finding in plain language and leaves it open.

That last point is deliberate: an analyst never closes a finding because it failed to measure it. Closing means the situation genuinely improved, so a closed finding is always something that got better.

Findings you dismiss yourself are a separate thing, and the analyst learns from those instead (see below).

Steer an analyst by talking to it

Direction you give the Strategist in conversation becomes standing guidance, not just chat history:

Teach one analyst your business. Tell the Strategist something a specific analyst should always weigh - “for churn risk, accounts under $1,000 a year aren’t worth flagging” - and it records the guidance, reads back exactly what it saved, and that analyst honors it on every future run.
Withdraw guidance that no longer applies. Instructions do not have to pile up. Tell the Strategist that something you said before is no longer true - “we do chase the small accounts now” - and it retires that entry rather than stacking a contradiction on top of it: it stops weighing the old instruction immediately, tells you exactly what stopped applying, and records the change in the decision journal so the history of what you asked for is still there. If you are replacing one instruction with another, both happen in one step. Ask what an analyst has been told (see below) if you want to see the list before changing anything.
Teach all of them at once. Standing facts about your business (“December is our busy season”, “we never contact customers directly”) reach every analyst. The Strategist listens for these in conversation and offers to remember them; you confirm the wording before anything is recorded.
Tune the sensitivity. Ask an analyst to be more conservative (fewer, higher-confidence findings) or more aggressive (flag borderline findings too), and the Strategist moves it one notch at a time and tells you where it landed.
Dismissals teach it too. Analysts see what you dismissed and won’t re-raise equivalents unless the facts materially change. If several dismissals share a pattern, the Strategist checks in once - “should I treat December as seasonal?” - and your answer becomes standing guidance. It never adjusts its judgment from dismissals without asking first.

Quiet an analyst for a while

Some reasons for silencing an analyst have an end date: you are in your busy season, a process is mid-migration, a known problem is already being worked on. Say so and give it a date - “stop flagging late shipments until January 5th”, “give it a month” - and the analyst goes quiet and then picks itself back up on that date. Nobody has to remember to switch it on again.

That is different from pausing, which has no end: the analyst stays off until someone asks for it back. The Strategist prefers the dated version and tells you which one it did, because a pause nobody remembers to undo looks exactly like an analyst that has nothing to say.

Existing findings are untouched either way, and you can end a quiet spell early just by asking. A date more than a year out is really a decision to stop watching, and the Strategist will say so rather than hide it behind a date.

Ask how an analyst is set up

Ask the Strategist how any of its analysts is currently configured - “how is commitment reliability set up?” - and it tells you the whole picture in plain language: what it watches, how often it checks, how sensitive it is, which records it includes, who hears about it, what it does when it finds something, and the standing guidance you have given it over time. That last part is worth asking for now and then: guidance you gave months ago is still shaping what it flags, and this is how you see it again and change your mind about it.

Everything in that answer can be changed in the same conversation, including renaming an analyst - useful when several of them watch different processes and you want to tell them apart at a glance.

Narrow which records an analyst follows

For the analysts that follow the companies, suppliers, customers, or roles in one of your process tables, you can describe the set you want in words: “only watch the customers whose status is active”, “leave out the ones we’ve marked dormant”.

The Strategist works out the conditions, then runs them before saving anything and shows you what they actually selected - how many records matched out of the total, and the most recent ones by name. That check matters: a filter that quietly matches nothing (or everything) looks perfectly reasonable written down, and you would only find out weeks later from findings that were never right. Read the list back, and if it is the set you meant, tell the Strategist to go ahead. If it is not, say what is wrong and it adjusts and shows you again.

Have an analyst act on what it finds

An analyst does not have to stop at telling you. Tell the Strategist what should happen when it finds something - “when it finds a customer at risk, start a retention review” - and it wires that up in the conversation: which of your processes should run, and which of that process’s fields should receive the details of the finding (its headline, the summary, the full detail, the process it concerns, the people to notify, and the record it is about). Anything you leave unmapped is simply not filled in.

Then you choose how far it goes:

Tell me only - it surfaces the finding and does nothing else. This is where every analyst starts.
Bring it to me for approval - it prepares the action and waits. You get the finding with the recommended action attached, and the work starts only when you approve it.
Act on its own - it starts the work and tells you afterwards. This one has to be earned; see below.

If you ask for an analyst to propose or launch work without saying which process should run, the Strategist asks you which one rather than switching on a setting that would silently do nothing.

Earning the right to act on its own

The Strategist will not put itself in charge. Ask it to start acting without approval and it declines, and tells you where its record stands instead: “you have approved two of these and neither has verified yet.” That is its own standard, not a restriction on you. It runs unsupervised on an analyst only once there is a track record you can see: a run of recommended actions you approved, with enough of them measurably improving the numbers afterwards and none of them making things worse.

Until then it offers the middle step, which is exactly how the record gets built: let it bring each action for your approval. When the record is there, it comes to you with the offer itself, and taking it up is one word.

If you would rather not wait, an administrator can switch an analyst to acting on its own directly in its settings on the Focus Areas page. That is a legitimate call to make, and it is recorded in the decision journal the same way the earned route is, noting who granted it and that it was granted directly rather than earned.

Either way you can put an analyst back to asking first at any time, and anything already waiting on your approval keeps waiting for you.

You never have to approve an analyst

Staffing is the Strategist’s job, not yours. When it sees a pattern one of its analysts is built for, it puts that analyst to work - it does not wait, and it does not ask you to approve the hire. Every one starts observe-first, each is recorded in the decision journal so you can see what was set up and why, and any of them can be paused or retired with a word in a strategy session.

Approval belongs on what an analyst proposes to do, not on whether it may look. A proposal to change a process, launch work, or automate a step still waits for you every time.

Two things work differently, and both for the same reason: they reach outside your account or outside the data.

The watchers of outside companies (news, customer churn, competitors, suppliers, and the talent market) usually ask you one question before they start, because they research named companies on the open web and only you know which list is worth following. The question is which records to watch, not whether the Strategist may watch them, and when your data makes the answer obvious it skips the question and simply starts.
Handing over a process to the AI Process Owner stays your decision, because that one does not just watch: it runs the queue and contacts your people.

Meet the analysts

Here are some of the analysts you can put to work, grouped the way the catalog groups them. The in-app catalog is always the complete, current list.

Executive Focus Areas

AI Process Owner

The handover, not just a watcher: give it one process and the Strategist runs that process’s queue for you. It checks in personally with whoever an overdue or unusually slow step is waiting on, follows up if nothing moves, and escalates a genuinely stuck run to you with the decision teed up - a specific reassignment recommendation it carries out on your approval. Approved moves build a track record that can earn it the right to route stuck steps itself, and its results are measured against how the process ran before the handover. It never changes how the process works without your approval, everything it does is journaled, and you can take the process back anytime. See Process Ownership for the full picture.

Setup - choose the one process to hand over. Unlike most focus areas, the Strategist never sets this one up by itself: ownership is always your explicit decision - approve its standing offer, ask for it in chat, or add it here.

Year in Review

Marks the anniversary of your first activity with an honest look back: how your volume of work moved year over year, whether work got faster, what you built, and which processes carried the year. It tells the story once, warmly and truthfully (a down year is narrated as a down year, leading with whatever genuinely improved), then steps aside until next year.

Setup - watches your whole account; it speaks up in the week after each anniversary.

Financial Focus Areas

Negotiation Prep

Walks into your next negotiation with the folder already prepared. Tell it who you are negotiating with and it pulls together your own record of the relationship: how much business actually flows through it and which way that is trending, whether their turnaround and outcomes improved or slipped, and the friction along the way. Then it hands you a prep brief with your strongest cards, the facts that favor them (named honestly), and the questions to ask. Turn it on before the conversation and retire it after.

Setup - choose the process your relationship runs through, type the name of the vendor, customer, or partner in Who are you negotiating with?, and describe the negotiation itself (a renewal, a rate review, a scope change, and what you want out of it) as standing guidance - tell the Strategist in conversation, or type it into the analyst’s guidance panel. The brief is framed against what you put there.

Operational Focus Areas

Anomaly Sentinel

Notices when today is broken, while there is still time to fix it. Every few hours it compares today so far against your own rhythm for the same weekday and hours over recent weeks. A morning running far below normal usually means an intake quietly died (a web form, an integration, an inbox), and it tells you before lunch instead of at month end. A surge far above normal gets a heads-up too, with the processes where the change concentrates named. Normal days stay quiet.

Setup - watches your whole account; no process to choose.

Commitment Reliability

Keeps score on the dates you promise your customers. You point it at a process and tell it which field holds the committed date (a promised ship date, a delivery date, a go-live, a target resolution date), and it tells you how often you actually hit it, how far off the misses run, whether you are getting better or worse, and which promises are already past their date with nothing delivered yet. Those last ones are the ones you can still save, so it names them rather than waiting to file them as history.

It also cares about which deadline it is judging. Often a promised date is not about the whole process finishing: you promise a delivery, and the run carries on afterwards with invoicing, closeout, or a follow-up call. Measured against the end of the process, a delivery you made two days early looks weeks late. So when the numbers say the process keeps running well past the promised date while one particular step lands right on it, the Strategist tells you what it suspects and asks you directly: which step has to be done by the promised date? Name the step in your reply and it starts judging the promise there, so correcting it is a conversation rather than a hunt through settings. If your answer is that the promise really is about the whole run finishing, it leaves the measurement alone and stops raising it.

Setup - pick the process and the field holding the promised date. If the promise is about one step rather than the whole run, name that step too; if you do not, the Strategist works out that something is off and asks you about it.

Institutional Memory

Remembers how it went last time, and says so at kickoff. When new work starts for a customer whose past runs on the same process went badly (worse outcomes, more rework, blown timelines), it speaks up the moment the work begins, so your team can give it a closer eye from the start. It is the 20-year employee’s memory applied the instant it is useful, and its advice is always vigilance, never blame.

Setup - nothing to do; the Strategist runs this one across all your processes. If you would rather one process were watched on its own, ask for that in a strategy session and this analyst steps aside for it.

Post-Mortem Debriefs

Writes the debrief nobody else will. When a run ends badly (far past its normal cycle time, heavy with rework, or canceled after real effort), it puts the facts together while they are fresh: how long it ran versus normal, exactly where it sat, and the one change worth a ten-minute conversation. Blame-free by design: it talks about steps and structure, never people, and it is honest about what the data cannot show.

Pre-Mortem Guard

Notices when something unusual just entered a process (the largest order it has ever handled, a value far past anything in its history) and checks how runs like it actually ended before. If past big runs collected extra rework or blew their timelines, it says so at kickoff and suggests extra care for this run only, such as added checkpoints or early confirmation of capacity and dates. It never proposes redesigning a process over one unusual run.

Process Stability

Learns the range each process normally runs in - how long runs usually take, how often a run hits a problem along the way, how often work is abandoned before finishing - and speaks up only when the numbers genuinely leave that range: a reading beyond anything ordinary week-to-week variation produces, or a sustained drift that almost never happens by chance. Everyday ups and downs stay quiet on purpose, so when this analyst speaks, something real changed and it is worth finding what. It flags changes for the better too: when a process suddenly runs faster or cleaner than its own normal, that is worth understanding so it can become the standard way the work is done rather than a lucky stretch.

Setup - nothing to do; the Strategist puts one of these on each process that runs often enough for a control range to mean something, so no process gets left out because busier ones came first. Optionally, have it also watch whether steps finish by their due dates (only useful where steps carry due dates). Judging which runs hit a problem needs AI Strategist Signals; the pace and abandonment outcomes work without them.

Cycle-Time Predictability

Watches whether each process finishes on a timeline people can actually promise against. A process can run steadily and still be impossible to quote: a typical run takes three days, but the slowest one in twenty takes two weeks, so every commitment gets padded for the slow tail. This focus area reads the real run-by-run finishing times and speaks up only when that gap is genuinely wide, then points at the slowest runs so you can find what they share (a step that waits on someone, a handoff, one kind of request) and make whatever the fast runs do differently the standard. When you act on it, the Strategist keeps measuring and tells you, with real numbers, whether the spread actually tightened.

Setup - nothing to do; the Strategist puts one of these on each process that completes enough runs to judge the spread honestly.

Repeatable Decision Automation

Finds the steps where the answer is always the same, and handles the two kinds of “always” very differently.

Where a particular field predicts the answer, it offers to take the step off your team’s hands. It reads how the step was actually completed over the past year, only counting runs where it can tell what the person was looking at when they decided (if a field was edited after the step closed, that run is left out rather than guessed at), and it replays the proposed rule over that history so you can see how often it would have matched and exactly which runs it would have got wrong.

Before it offers you anything, it asks one question: is that field really what decides this, or are you checking something it cannot see, such as a call, another system, or something you just know? That question matters more than the numbers. A step can come out the same way a hundred times because of a phone call nobody records, and a rule built on that would confidently do the wrong thing the first time the call went differently. Only a confirming answer becomes a proposal. If you name something outside the process as the real decider, it says so plainly, leaves the step alone, and tells you what it would need to be able to take it over later.

Approving the proposal creates a live automated action on that step. It completes the step with the response you confirmed, the moment the step is assigned. Anything other than the value the rule depends on and the automation stands down, so a person completes the step exactly as they do now. You can see it, change it, or switch it off in the process designer at any time, and the Strategist keeps measuring the hands-on time on that step so it can tell you whether the change actually saved the hours it expected.

Where nothing in the process predicts the answer and the step simply always goes the same way, it will not offer to automate it, and it will tell you why. An automation with no condition on it would complete the step on every single assignment from the moment you approved it, with nothing left for it to get right or wrong. So it asks the people who do the work a different question instead: what is this step catching? A step that always comes out the same way is either a real check that would go the other way if something were wrong, or a habit the process has kept out of momentum, and only the people doing it can tell those apart. If the answer is that it is a genuine check, the Strategist closes the question and leaves the step exactly as it is. If the answer is that it has become a formality, it tells you what the step is costing in hands-on hours a year and leaves the decision with you. It never removes a step.

Setup - nothing to do; the Strategist starts this one once your processes have enough completed history for a pattern to be real. Steps that already have an automated action are skipped.

AI Delegation Opportunities

Finds the steps where someone is doing work that does not need to be done by hand - researching a request, drafting a reply, classifying or triaging, pulling the figures out of a document that came in, summarizing something long, checking details against your rules, or re-keying data between systems - and ranks them by the hands-on hours they cost each year. For the candidates it is confident about it does more than name them: it writes the working instructions and brings you a proposal to turn on, so there is nothing for you to configure.

It picks the cheapest approach that can actually do the work, because they cost very different amounts to run:

A single AI call where the work only needs what the run already holds (its field values, or a file attached to it). It reads those, writes its answer into the field you were shown, and stops there: it does not complete the step and it does not choose an outcome. The answer is simply waiting when the person opens the task, and they decide. This is the cheapest way to take work off someone, and the safest place to start - nothing moves without a person. If it earns your trust and you later want it to complete the step as well, that is a setting on the same action in the designer (see AI Prompt Action Type).
A full AI agent where the work genuinely needs to look something up, reach another system, or proceed in several steps. An agent does the work and then completes the step itself, choosing the outcome that fits what it found. If you want a person to check its work first, that is a review step after it, which the Strategist can add for you.
A request for a screen recording where the work is not done in ProcessPlan at all - somebody logs into a portal, downloads a statement, re-keys figures into another system. This is often the most expensive work in a process, and no AI can be written for it from the outside, so the Strategist does not try. See below.

For the first two, the proposal always shows you the exact instructions before you agree to anything, because those instructions are the whole of what the automation will do. Approving turns it on immediately, from the next time that step is assigned. You can read, edit, or switch it off in the process designer at any time, and the Strategist keeps measuring the hands-on time on that step so it can tell you whether the hours actually came down.

When the work happens on someone’s screen

Where a step’s real work is done by hand in another system, the Strategist will not invent an automation for it, and it will say so. It has never seen that screen, and guessing at a procedure it cannot observe is how screen automations end up brittle. What it does instead is ask for a recording.

The proposal tells you what the step is costing in hands-on hours a year and who it wants to ask. The request goes to whoever is accountable for the process, and it names the person who completes that step most often as the suggested recorder, so it can simply be passed on - they can hand it to somebody else if that is the wrong person.

Approving changes nothing about your process. It sends a real task, not a message that can be missed: it asks who is going to record the procedure and when, the Strategist follows it up if it slips, and it comes back to you with the answer. From there the recording goes through the Screen Recorder in the normal way. Nobody is committed to anything by approving it - you are agreeing to ask for an hour of somebody’s time.

Because a recording costs real time to make, the Strategist only asks where the annual hours clearly justify it, and it asks once per step, once only. Whether the recording arrives or not, it will not come back and ask the same thing again. The step stays on its list while the hours are still being spent, so you will still see it named in a strategy session or a briefing - but the request itself is made a single time.

A step that already has AI on it is left alone rather than given a second automation. And where the answer to a step is always the same, this analyst stays quiet and leaves it to Repeatable Decision Automation, which can handle that with a plain rule and no AI cost per run at all.

Setup - nothing to do; the Strategist puts one of these to work across all your processes once there is enough completed work to rank the candidates honestly. If you would rather one process were looked at on its own, ask for that in a strategy session. Turning on an automation it proposes needs edit access to that process; approving a recording request does not, since it changes nothing.

Supplier Risk

Sees a supplier becoming a problem before your supply does. It puts together two things that are normally watched by different people, or by nobody at all: what is happening to that supplier out in the world (distress, layoffs, a plant closure, a recall, a breach, a change of ownership) and how they are actually behaving inside your own processes (quality slipping, turnaround stretching out, engagements stalling). Either signal on its own is easy to explain away. Together they are a reliable early warning, and it goes to whoever owns that category with a concrete next move rather than a vague worry: confirm buffer stock, or qualify an alternate source now instead of during a crisis.

Setup - point it at a vendor-facing process and tell it which field holds the supplier name; it then follows every supplier in that process table and checks in every couple of weeks. You can narrow it to part of the list with Only include records that match, and anything you tell it about which suppliers or risks matter most, such as sole-source parts or critical lead times, shapes what it flags.

People & HR Focus Areas

Absence Coverage Planner

Thinks ahead about vacations and absences so nothing slips while someone is away. When a person on your team sets an out-of-office window, it looks at what will land in that window: the tasks already in their queue, the due dates that fall while they are out, and the pace at which new work usually reaches them. It speaks up before the window opens if no backup is named (so nothing will reroute) or if the named backup is about to receive a heavy handoff. Well-covered absences stay quiet.

Setup - watches your whole account; no process to choose.

Departure Shield

Acts the moment someone leaves. When a user is removed from your account, it immediately takes stock of what walked out with them: the open tasks still stranded in their queue and the steps where they did nearly all the recent work, the places their know-how leaves a hole. Then it hands you the departure checklist their manager never wrote: reassign this, name an owner for that, capture what they knew while memory is fresh.

It also offers to do the first part for you. When there is stranded work and the departed person’s record names someone who can pick it up - the backup they designated, or the person they reported to - the card comes with the handover ready to approve: one approval moves their whole open queue to that person, who is then notified exactly as they would be for any other assignment. It re-reads the queue when you approve, so anything already picked up by hand is left alone, and it tells you plainly if any task could not be moved and why. Approving it needs permission to change task assignments, the same as reassigning those tasks by hand. If someone else should inherit the work, say so in a strategy session and it moves it there instead. When the record names nobody, it does not guess - the card stays a checklist for you to act on.

Setup - watches your whole account; no process to choose.

Quiet Overload

Watches for the thing good people never say out loud: too much on their plate. It speaks up about a person only when three signals point the same way at once - a heavy open queue, work arriving faster than they can clear it, and their turnaround stretching against their own normal pace (never a comparison to colleagues). Any one signal alone stays quiet. What arrives is an early check-in prompt for a manager, never a performance judgment: their load has been climbing for a month, worth a conversation, and maybe moving a few things. It always names the innocent explanations too (a backlog after time off, a run of harder work), and the conversation decides, not the Strategist.

Setup - watches your whole account; no process to choose. Unlike most account-wide focus areas, the Strategist does not turn this one on by itself: putting a people lens to work is deliberately your call - ask the Strategist for it by name, or add it from the catalog.

Performance Focus Areas

Effort vs. Outcome Steering

Answers the steering question no single view can: is your team’s effort pointed at the work that actually ends well? For one process, it weighs how much of your team’s hands-on time each segment of the work consumes (by lead source, region, category, or any other field) against how that segment’s outcomes actually turn out, and flags the big mismatches, like a third of the hours going to the work that ends worst. When that allocation is not deliberate, steering existing effort is improvement without new spend.

Setup - point it at the process whose effort allocation matters most. Needs AI Strategist Signals and time tracking on tasks.

Entity Cadence Radar

Learns each relationship’s own rhythm in a process (the customer who orders every 90 days, the site that gets a monthly visit, the filing that recurs quarterly) and speaks up when that rhythm quietly breaks: a customer who has ordered like clockwork for two years is now three weeks past their usual gap. Overall volume can look perfectly normal while one relationship goes silent; it catches the one. Only relationships with a genuinely regular history are judged, each against its own pace.

Setup - point it at the process whose per-relationship rhythm matters (orders, service visits, recurring filings).

Entity Participation Gaps

Finds the customers, vendors, or other entities that skip a process most of their peers go through: if 14 of 17 companies in your fulfillment process also go through your annual review, it names the 3 that never have. On a sales pairing that reads as untapped opportunity; on an operations pairing it reads as a skipped obligation (onboarded but never audited, ordering but never surveyed). It discovers the pattern from your own data and asks whether the gap is deliberate.

Setup - point it at the process whose participants it should start from.

Response-Time Decay

Learns your own numbers behind “answer fast, win more.” From your finished work it measures how outcomes fall away as the first touch slips, finds the point where they drop off, then flags the open work currently aging past that point, with your historical odds attached. Working the aging list first is often the cheapest outcome improvement available on any given day.

Setup - point it at a process where responsiveness plausibly matters (quotes, inquiries, applications, support). Needs AI Strategist Signals.

Research Focus Areas

Opportunity Watcher

Watches the outside world for new business you did not have to go find: open RFPs and tenders, procurement openings, grant programs, and partner solicitations that match your company profile. It brings you the pursuable ones with deadlines front and center, each with its source, because an opportunity you hear about after it closes is worthless. It works from your company profile from day one; keep the profile sharp and the matches sharpen with it.

Setup - watches on your company’s behalf; no process to choose. Telling it what you sell, where you operate, and what contract sizes are worth your time - in conversation, or in the analyst’s guidance panel - sharpens the matches considerably.

Sales Focus Areas

Capacity Windows

Connects the two sides of the business that never talk: what is being sold and what your team can absorb. It weighs each process’s demand trend against the throughput your team has actually demonstrated and names the windows: slack, when you are running well below proven capacity and a push for more work fits right now, and walls, when demand is pressing against capacity and work sold today lands in a queue that cannot absorb it. Capacity is inferred from your best delivered week, so figures are directional.

Setup - nothing to do; the Strategist puts one of these on each process with enough history to trend demand against delivery.

Customer Churn Risk

Catches a customer drifting toward the exit while there is still time to keep them. It puts together two things that are normally watched by different people, or by nobody at all: what is happening to that customer out in the world (layoffs, budget cuts, a change of leadership or ownership, signs of distress) and how they are behaving inside your own processes (engagement tailing off, outcomes getting worse, work stalling or being canceled). Either signal alone is weak and easy to explain away. Together they are a strong warning, and it goes to the account manager with a specific save play while the relationship is still savable.

Setup - point it at a customer-facing process and tell it which field holds the customer name; it then follows every customer in that process table and checks in every couple of weeks. You can narrow it to part of the list with Only include records that match, and anything you tell it about these customers or the churn signals that matter most to you shapes what it flags.

Moments of Delight

Catches the moments that go unusually well, and makes sure you act on them before they cool. When a customer’s just-completed work scores well above the process’s own norm, it tells the relationship owner that this week is the window: a personal thank-you always fits, and where the relationship supports it, this is the natural moment for a referral, testimonial, or review ask. Everything else watches for problems; this one watches for goodwill and puts timing on your side.

Trend Focus Areas

Per-Entity Value Drift

Catches one relationship’s numbers quietly drifting while the overall picture stays flat: the discount that crept from 6% to 17% for a single repeat customer, the rate that slid for one vendor, the score eroding for one site. It compares each relationship’s recent values against its own history and flags only the drifts the process-wide trend cannot explain. Averages hide exactly this pattern; looking one relationship at a time is what makes it pop out.

Setup - point it at a process that captures numbers about recurring customers, vendors, or other entities.