Query Resolution

Challenges Query Resolution addresses

Done by hand, customer support means gathering inbound customer message, working through 5 separate passes over the same material, then producing classification, draft reply and account record. None of it is difficult and all of it is exacting, which is the combination people are worst at holding. The errors that matter are the ones a tired reader does not notice, and they surface later — in a reconciliation, or in somebody’s reply. It starts whenever something arrives, which means someone has to be watching for it to start at all. As volume grows the work does not get harder, only longer, and the first thing to go is the checking.

Query Resolution runs that same sequence end to end and returns the result as structured artefacts. What it cannot settle it hands over rather than guesses at, and your correction is kept: it asks “Would you have sent this?” after every run, and those answers become the set it is measured against. Nothing that moves money, alters a contract or reaches a customer executes without human approval, and every action is written to an audit log. The gain is in the volume that no longer has to be read, not in removing the judgement.

How it works

Step 1: Classifying the message

First of 5. It works from inbound customer message and feeds the step after it.

Key Tasks:

• Reading each item on its own terms: Classification is from the content rather than from a keyword list, so items phrased in a way nobody anticipated still reach the right customer support bucket.
• Applying the same boundaries every time: The definitions do not drift between runs, which is what makes a count from this month comparable to one from last.
• Keeping the confidence: A borderline item is assigned with its confidence recorded rather than filed away as certain.

Outcome:

• Confidently classified: The item is grouped and counted, and moves on without review.
• Below the confidence threshold: It is held for a person rather than placed in the closest bucket — the wrong bucket is worse than an unfilled one, because it disappears into a count.

Step 2: Looking up the account

Step 2 of 5. It takes what step 1 produced and hands its result to step 3.

Key Tasks:

• Covering the whole set: Every record in scope is examined, not a sample. The step before it narrowed the field; this one does not narrow it further by accident.
• Judging relevance by content: Whether something belongs in this run is decided from what it says rather than from where it was filed or how it was named.
• Discarding visibly: What is excluded is recorded as excluded, so "nothing found" can be distinguished from "nothing looked at".

Outcome:

• Relevant items found: They pass to the next step with the reason they were selected attached.
• Nothing relevant: The run reports that it found nothing and stops, rather than producing an empty artefact that reads like a failure.

Step 3: Searching the knowledge base

Step 3 of 5. It takes what step 2 produced and hands its result to step 4.

Key Tasks:

• Sweeping the knowledge base: The sweep is defined by the knowledge base, so what it returns is scoped to this step rather than to everything the connected system holds.
• Carrying the reason forward: Each item that survives arrives at the next step with why it was selected attached, so a false positive can be traced to the rule that admitted it.
• Covering the whole set: Every record in scope is examined, not a sample. The step before it narrowed the field; this one does not narrow it further by accident.

Outcome:

• Matches returned: They pass to the next step with the reason each was selected attached.
• No matches: The run says so and stops rather than producing an empty artefact that reads like a failure.

Step 4: Deciding how to handle it

Step 4 of 5. It takes what step 3 produced and hands its result to step 5.

Key Tasks:

• Working from the run so far: This step takes what step 3 produced and carries it toward step 5.
• Following the same rules each time: The behaviour is configuration rather than judgement made fresh per run, so customer support is handled the same way every time.
• Surfacing what it cannot settle: Anything ambiguous is passed on as ambiguous rather than resolved silently.

Outcome:

• Passed on: The result passes to the next step, with anything unresolved carried forward as an open item rather than dropped.

Step 5: Preparing the reply for approval

Last of 5. It takes what step 4 produced and produces classification and draft reply.

Key Tasks:

• Writing from the run, not from a template: The text is built from what this run actually found, so two customer support outputs differ where the underlying records differ.
• Leading with what needs a decision: The exceptions come first and the routine detail follows, because the reader is deciding rather than reading.
• Staying inside the evidence: Nothing appears in the text that is not supported by a record the run examined. Gaps are stated as gaps.

Outcome:

• Draft ready: It is held for approval. Nothing reaches a customer, a calendar or a channel until a person releases it.
• Not enough to write from: It says so instead of producing something plausible from thin evidence, which is the failure that is hardest to catch on review.

Step 6: Your review, and what it changes

The run ends with a person, not with a result being filed.

Key Tasks:

• Asking a specific question: It asks “Would you have sent this?” rather than for a rating. A question about this run is answerable; a score out of five is not.
• Keeping the correction: What you change is recorded against the case that produced it, so the disagreement is retrievable rather than absorbed.
• Building the evaluation set: Those cases become what the agent is measured on. It is scored against your judgement rather than against a general benchmark.

Outcome:

• A measured agent, not an assumed one: The cases Query Resolution handles well and the cases it does not are both visible, and the second list is the one that decides what changes. Nothing is retrained silently on the back of a single correction.

Why use Query Resolution?

• Answers from your own material: It retrieves from a knowledge base you populate — your policies, contracts and reference documents — so its decisions follow how your business actually operates rather than a general model’s assumptions. Sources are cited alongside the result.
• Nothing leaves without approval: Draft reply is drafted and held. A person releases them, so the agent's reach ends at your own review step.
• Every statement cites its source: Findings come back with the records behind them, so a reviewer can check a claim instead of deciding whether to trust it. An assertion with no source is the expensive kind to discover late.
• A batch is one run, not a hundred: It works the whole set in a single pass and returns a row per item with its verdict, so the volume that needs no attention never has to be opened.
• Works from the message itself: The body and its attachments are the input, so a request does not have to be transcribed into a form before anything can happen to it.