Agentic AI Capability

Five steps turn a question into the right query.

Between a typed sentence and a correct answer sit five steps: recognise the intent and the business objects, frame the goal, find the tables your team mapped, write the query for one source or several, and run it with the asking person’s access.

THE QUESTIONWhich components are short for production orders startingin the next ten working days at plant 1210?LTWHAT HAPPENS UNDERNEATH01Recognise theintenta shortage list, forwardten working days, one plant02Frame the goaleach short component, thequantity short, the date itstarts03Find the mappedtablesrequirements, stock, openorders, supplierconfirmations04Write the queryfour tables in threesystems, joined on materialand plant05Run it with youraccessplants 1210 and 1220; therows stay in the sourceREAD IN PLACESAP S/4HANAOracle FusionPlanningsuiteTHE ANSWER14 components are short9 of them inside five working days7741-X short by 4,200 pieces on 2 Octobertwo receipts confirmed for dates after the needshowcase estate · 4.2 s from the question to the answer
The question at the top, the five steps underneath, the three systems read in place, and the answer. Showcase estate.
5
steps between a typed question and the answer on screen
product
4
kinds of question, told apart before the answer is built
product
3
systems joined in one question in the showcase estate
showcase estate
4.2 s
average time for the whole sequence, shown on the answer
showcase estate
Where this sits

Stage 3 of six: ask

Step by step

One question, five steps, in order

Two steps settle the meaning; three steps fetch the facts.

1
Recognise the intent and the objectsThe sentence names a shortage, a horizon of ten working days and one plant. Component, production order, stock and plant are objects in the supply chain ontology: Lima’s built-in map of supply chain things and how they relate.
intent: shortage list
2
Frame the goalWhat is asked for: each short component, the quantity short and the date the shortfall starts. Grouped by component, ordered by date, for the plant named in the question.
goal framed
3
Find the mapped tablesThe knowledge index names the registered tables behind each object and the columns your team confirmed for them.
4 tables · 3 systems
4
Write the queryRequirements and stock come from one ERP, open orders from two, the working calendar from the planning suite. The federated query engine joins them on material and plant.
1 question · 3 queries
5
Run it with the asking person’s accessL. Tan is cleared for plants 1210 and 1220, so that filter travels with every part of the query and the rows stay in the source systems.
run as L. Tan

The first two steps are about meaning. Recognising the intent settles what kind of answer the person wants: a list, a comparison, a cause, a forecast. Recognising the objects settles what the answer is about: components, production orders, plants, suppliers.

Framing the goal makes both explicit. Juno states the measure, the period, the grouping and the filters it will apply, and that statement travels with the answer. A reader who disagrees with the frame can see where it went wrong before looking at a single number.

The last three steps are about data. The knowledge index holds every registered table and column, the values inside each column, and the names your team gave them, so the step from an object to a table is a look-up.

The query is then written against the mapped columns, run through the federated query engine, and filtered by the rights of the person who asked. Juno keeps the question, the frame, the query and the answer, so the same question can be re-run tomorrow and compared.

Four kinds of question

Each question is told apart before it is answered

Telling the four apart keeps a rule from being answered as if it were a measurement.

A policy question

About a rule the company has set, such as the safety stock policy for a material class.

A plan question

About a published plan, such as what the cancellation plan will do this week.

A question about what Juno knows

About the estate itself: which tables are registered, what is mapped, when a source was last read.

A data question

About the business: stock, orders, deliveries, suppliers. Answered by a query over your data.

Kind of questionExample someone typesWhere the answer comes from
Policy“What is our safety stock rule for A-class materials at plant 1210?”The policy register, with the named policy and its version
Plan“What will the purchase order cancellation plan do this week?”The versioned rule store, with the planning table
What Juno knows“Which tables hold the supplier confirmed date, and when were they last read?”The knowledge index and the registration record
Data“Which components are short in the next ten working days at plant 1210?”A query over the registered tables, read in place

The kind of question is shown on the answer, so a reader knows whether they are reading a rule, a plan, a fact about the estate or a measured figure. Product.

One question, three systems

How the parts of a sentence reach different tables

Each phrase in the sentence lands on a business object, and each object on tables your team confirmed.

Part of the questionBusiness object it namesWhere it is readWhat comes back
“components … for production orders”Production order · component requirementSAP S/4HANA · reservations and order itemsrequired quantity and date for each component
“short”Stock · open purchase order · supplier confirmationSAP S/4HANA and Oracle Fusionstock on hand and receipts due before the need date
“the next ten working days”Production order start datePlanning suitethe working calendar of the plant
“at plant 1210”Plantall three sourcesthe same filter applied in every part
SELECT c.material_id, c.material_name,
       SUM(r.required_qty) - SUM(COALESCE(s.on_hand_qty, 0) + COALESCE(p.due_qty, 0)) AS short_qty,
       MIN(r.requirement_date) AS shortfall_starts
FROM   ontology.component_requirement r
JOIN   ontology.material c ON c.material_id = r.material_id
LEFT   JOIN ontology.stock s ON s.material_id = r.material_id AND s.plant_id = r.plant_id
LEFT   JOIN ontology.purchase_order_line p ON p.material_id = r.material_id AND p.plant_id = r.plant_id
                                          AND p.confirmed_date <= r.requirement_date
WHERE  r.plant_id = '1210' AND r.requirement_date <= CURRENT_DATE + INTERVAL '14' DAY
GROUP  BY c.material_id, c.material_name HAVING short_qty > 0 ORDER BY shortfall_starts

The query ChatLima wrote for the question in the hero figure, shown the way a reviewer sees it. Table names are ontology names; the engine resolves each one to the registered tables behind it. Showcase estate.

When a sentence allows two readings

One clarifying question, then the answer

A single clarifying question costs seconds; a silently wrong assumption costs a quarter.

ChatLima · D. Okafor · buyershowcase estate
DO
Which of our suppliers were late last month?
L

Two dates carry the meaning of “late” in your estate. Which basis would you like?

a. The delivery date you requested on the order. b. The date the supplier confirmed in the acknowledgement.

one clarifying questionthe choice is kept for this thread
DO
Against the confirmed date.
L

11 suppliers received goods after their confirmed date in August. Delta Fasteners Ltd carries the largest share of the late lines.

SupplierLines dueLate linesLate rate
Delta Fasteners Ltd162148.6%
Corvid Plastics7745.2%
9 other suppliers612101.6%
supplierpurchase order schedule linesupplier confirmed dategoods receiptSAP S/4HANAOracle Fusion
3.8 s2 sources read in placebasis: supplier confirmed datewindow: August, one month of the quarter
Compare the same month last yearSplit Delta Fasteners by plantSave this as a weekly question
Ask about orders, stock, suppliers, shipments…

Some words carry two meanings in one estate. “Late” can mean after the date you requested or after the date the supplier confirmed; vendor evaluation often uses a third date kept for statistics. A product that guesses produces a number that looks right and is wrong for half the room.

ChatLima asks one question, offers the readings it found, and keeps the choice for the rest of the thread. When your company settles on one meaning, it is taught once in the vocabulary and the clarifying question stops appearing.

When the registered data reaches only part of a question, Juno answers that part, says which part is missing and names what it would need: in this estate, a source that carries supplier confirmations for the acquired division.

The running example

Wednesday 09:12: one sentence, three systems, 4.2 seconds

Showcase estate.

The running example

L. Tan’s question named open purchase order lines, a cancellation window and a week. Juno framed the goal as open lines inside their window before Friday, grouped by supplier, valued at the open quantity times the net price.

The mapped tables sat in three systems: order lines and windows in SAP S/4HANA, the acquired division’s lines in Oracle Fusion, projected demand in the planning suite. One question became three queries, joined on material and plant, run with L. Tan’s rights on plants 1210 and 1220. The answer arrived in 4.2 seconds with the query one click away.

Next capability

Answers the way you meant them

The same result as a table, a chart or a short narrative, with walkthroughs to re-run it and a route into your reporting tools.

Terms on this page
IntentWhat kind of answer the question asks for: a list, a comparison, a cause, a forecast.
Business objectA thing in the supply chain ontology, such as a material, a plant, a supplier or a purchase order line.
Goal frameJuno’s statement of the measure, period, grouping and filters it will use, shown with the answer.
Knowledge indexJuno’s record of every registered table, column and value, held in business language.
Federated query engineThe engine that runs one question across several registered sources and joins the results.
Clarifying questionThe single question Juno asks when one word in the sentence carries two meanings.
Questions
Does ChatLima write the query itself?

Yes, against the ontology names your team confirmed. The engine resolves each ontology name to the registered tables behind it, so the query a reviewer reads stays in business language.

How does it know which tables to use?

From the mapping your team confirmed and the knowledge index built at registration. Each business object points at the tables and columns that carry it, with the values profiled for recognition.

What happens with a question about several systems?

The federated query engine sends one query per source, reads each one in place, and joins the results on the keys the mapping recorded. The showcase estate joins three systems in about four seconds.

Can a person edit the query?

A person with the right role can open the query, copy it and run it in their own tool. Corrections to the meaning belong in the vocabulary, so every later question benefits.

What happens when the data will support only part of the question?

Juno answers the part it can, states which part is missing and names what would be needed, such as a source carrying supplier confirmations.

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.