Agentic AI Capability

Every recommendation comes with its basis and its confidence.

A finding says that a rule matched. A recommendation goes further: it names the action to take, the evidence behind it, how well that evidence agrees and what is at stake.

WHAT THE RUN PRODUCESevery match produces thisA findingthe rule matched these factsThe facts as readvalues, systems, the momentThe rule versionversion 3, rows 01 to 03 matchedWHAT A RECOMMENDATION ADDSthis part waits for evidenceThe actioncancel line 00020 of order 4500067123The basisthe conditions that matched, in plain wordsThe confidence0.91, from evidence that agreesThe value at stake$8,420 · cover 11.4 weeks against a need of 8evidencesupports itshowcase estate · one line of the 38 the cancellation policy matched on Monday
A finding becomes a recommendation where the evidence supports an action. Showcase estate.
4
parts every recommendation carries: action, basis, confidence, value at stake
product
0.80
the confidence floor in the showcase estate, set by your team
showcase estate
38
lines the cancellation policy recommended on Monday, worth $412,000
showcase estate
9
of those lines stood at or above $25,000 and went to D. Okafor as buyer, for his judgement
showcase estate
Where this sits

Stage 5 of six: act

Step by step

How a rule match becomes a recommendation a person can judge

Five steps: match, basis, confidence, value at stake, rank.

HOW ONE RECOMMENDATION IS ASSEMBLED01A rule matchesthe conditions hold on thefacts of this run02The basis isassembledeach condition, the valueit read and where it camefrom03The confidence isscoredhow well the evidencebehind the action agrees04The value at stakeis worked outmoney, cover, service, andthe days left to act05The queue isrankedthe work of the morning,in the order it earnswhat a person accepts, edits or sets aside returns as evidence for the next score
The loop at the bottom matters: an edit or a set-aside is evidence, and the next score reads it.

The run gives the engine a set of lines where a rule matched. For each one, Juno assembles the basis: every condition in the rule, the value it read, the system that value came from and the moment it was read.

It then scores the confidence. The score measures how well the evidence agrees: whether each condition rests on a mapped field, whether two systems say the same thing, and how the same case turned out the last time a person saw it.

The value at stake follows. For a cancellation it is the open commitment and the cover the line returns. For an expedite role it would be the demand it protects. Every recommendation also carries the days left before the chance to act closes.

The queue is then ranked, so the first item on a planner’s screen is the one that earns the most attention. A person accepts, edits or sets aside each one, and every choice returns as evidence for the next score.

One recommendation

Eight parts, on one screen, for a single line

Action, basis, facts, rule, confidence, value, limits and what follows: one screen holds all eight.

Recommendation · DEC-2026-09-21-0417 · PO cancellation agent · drafted Mon 21 Sep 06:06 · showcase estate
Part of the recommendationWhat this one holds
The actionCancel line 00020 of purchase order 4500067123: 200 of material 7741-X, the open quantity, for plant 1210
The basisProjected excess 11.4 weeks · window open until Thu 24 Sep · supplier class standard
The facts as readRead Mon 06:01 from SAP S/4HANA and the planning suite
The rulePO cancellation policy, version 3; rows 01, 02 and 03 matched
The confidence91% three conditions agree, and the window is a mapped custom field
The value at stake$8,420 of open commitment · projected cover 11.4 weeks against a need of 8 · 3 days left to act
Limits checkedValue under $25,000 · daily budget 29 of 60 · confidence floor 0.80 all passed
What happens nextA draft for D. Okafor’s approval queue at A2, the level this policy runs at; a line of $25,000 and above reaches D. Okafor as buyer, for his judgement, with its reason A2 supervised

One of the 38 lines the policy matched on Monday. Showcase estate.

The basis is written the way the policy is written, so a buyer reads the reason in the same words the planner used when the rule was stated. There is one line per condition, and each one names the value it read.

Every part carries the same reference, so this recommendation, the approval that follows it and the answer from your system are one record on the audit trail.

The last row is what a reviewer looks for first: what follows from this recommendation at the level the agent runs at today. It carries the limits that were checked and the name of the person the action goes to when a limit sends it there.

A recommendation stays a proposal until a person approves it or a limit lets the agent act. Until that moment the line in your system stays as it is.

Confidence

What the score means, and what an empty recommendation says

The score is a measure of agreement, and an empty recommendation states what the agent waits for.

ConfidenceWhat it usually meansWhat Juno does with it
0.90 and aboveEvery condition rests on a mapped field, and the systems agreeActs inside your limits, or drafts for the approver
0.80 to 0.89The evidence agrees, with one value read from a single systemActs or drafts, with the weaker condition marked on the basis
Below 0.80A condition rests on a field two systems report differentlyReports the finding and asks a person to decide
Left emptyThe evidence is too thin to support an actionStates what it waits for, and hands the finding to a person
What Juno seesWhat it says on the findingWhat it leaves for a person
A fact the mapping has yet to coverNames the ontology attribute it waits forThe judgement, with the gap stated on the record
Two systems reporting different valuesShows both values, each with the system it came fromThe choice of which system to trust for this case
A score under your confidence floorReports the score and the condition that disagreesThe decision, with the working attached
History too short to test againstStates how many periods it holds todayThe judgement about acting now or waiting

A recommendation is left empty in these four cases, and the finding still reaches a person. Product.

An empty recommendation is a useful answer. It tells a planner that Juno read the case, found the rule matched, and holds evidence that falls short of an action. The finding still arrives, ranked with the rest, so the judgement stays with the person who has the context for it.

Ranking

The morning queue, in the order the work earns

Three things decide the order: the value at stake, the days left, and how well the evidence agrees.

RankLineWhat the agent proposesValue at stakeDays leftConfidenceRoute
01Order 4500067812 · line 00010Cancel the open quantity of 7741-X at plant 1210$61,40030.94To D. Okafor as buyer
02Order 4500067123 · line 00040Cancel the open quantity of 8802-K at plant 1210$38,90030.92To D. Okafor as buyer
03Order 4500068004 · line 00020Cancel 1,200 of 8802-K at plant 1210 · excess 9.3 wk · window Wed 23 Sep$24,10020.93Drafted for D. Okafor
04Order 4500067123 · line 00020Cancel 200 of 7741-X at plant 1210 · excess 11.4 wk · window Thu 24 Sep$8,42030.91Drafted for D. Okafor
05Order 4500067990 · line 00030Cancel 300 of 5513-B at plant 1220 · excess 8.6 wk · window Thu 24 Sep$7,16030.88Drafted for D. Okafor
06Order 4500068110 · line 00010Cancel 150 of 9021-M at plant 1220 · excess 12.1 wk · window Fri 25 Sep$5,90040.90Drafted for D. Okafor
07Order 4500068221 · line 00050Cancel 80 of 4410-T at plant 1210 · excess 10.2 wk · window Thu 24 Sep$2,28030.90Drafted for D. Okafor

Seven of the 38 lines, in queue order: two of the 9 at or above $25,000, which go to D. Okafor as buyer, and the five drafts chapter 5 shows in his approval queue. Showcase estate.

Ranking puts three things together: what is at stake, how long there is to act, and how well the evidence agrees. A line worth $61,400 with three days left sits above a line worth $8,420 with the same window.

You choose what ranking means for each agent. A team that cares most about cash sets the value first; a team working against closing windows sets the days left first. The setting is held with the agent and carries its own version history.

The running example

Monday 06:06: 38 recommendations, one ranked list

Showcase estate.

The running example

The 06:00 run matched 38 open lines worth $412,000, and by 06:06 each one carried its basis in the words of the policy, a confidence and the cover it would return. Twenty-nine sat under the $25,000 threshold and reached D. Okafor’s approval queue as drafts worth $187,300; nine ran above it and reached D. Okafor as buyer, for his judgement, with the reason for each.

One line outside the 38, for an allocation-bound supplier, came back as a finding with an empty recommendation, because two systems reported the supplier class differently. L. Tan read the finding, chose the value from the supplier master, and that choice became evidence for the next run.

Next capability

Approvals by a named person

How a drafted action reaches the person who owns the decision, what the queue shows, and what happens when a clock runs out.

Terms on this page
FindingA recorded match between a published rule and the facts at that moment.
RecommendationA proposed action with its basis, its confidence and the value at stake.
BasisThe conditions that matched, each with the value it read and the system it came from.
ConfidenceA score from 0 to 1 for how well the evidence behind an action agrees.
Confidence floorThe score below which an agent reports the finding and asks a person to decide.
Value at stakeWhat the action is worth: money, cover, service, and the days left to act.
Questions
Where does the confidence score come from?

From the evidence behind each condition: whether the field is mapped and confirmed, whether two systems agree on the value, and how people treated the same case before.

Why leave a recommendation empty?

Because a stated gap is more useful than a guess. The finding still reaches a person, with the attribute, the disagreement or the short history named on the record.

Can we change what ranking means?

Yes. Ranking is a setting on each agent: the value at stake, the days left and the confidence carry weights you choose, and a change becomes a new version.

Does a person see the working?

Yes. Every recommendation carries its rule version, the conditions that matched, the values read, the systems they came from and the moment they were read.

In the platform

Where Juno sits among the four capabilities

Each one builds on the one before it.

One shared foundationCatalogLineageAccess policyWrite path

See Juno answer and act on your own data

Bring two systems and one decision your planners make every week. In the first week we register the data, confirm the mapping with your team, answer your questions in ChatLima and run the first agent at the level you choose.