Google’s Data Agent Kit Connects Business Data: Can AI Explain Why Quotes Aren’t Turning Into Jobs?

Google’s Data Agent Kit does not automatically tell a Sydney business why a quote was lost.

It gives data practitioners a governed way to query and combine supported Google Cloud data sources, then turn an investigation into reusable data models.

For NSW property and renovation operators, that could expose patterns across price, response time, site access, scope changes, strata constraints and scheduling. The commercial value depends on clean data, consistent quote statuses and human interpretation.

Google Is Bringing Root-Cause Investigation Closer to the Data

Google Cloud's latest Data Agent Kit demonstration is notable because it moves beyond the familiar idea of asking an AI assistant to summarise a spreadsheet.

In Google's example, an analyst investigates why average order value has fallen while revenue remains flat. The evidence is distributed across a BigQuery warehouse, a Cloud SQL database and files held in Cloud Storage.

Rather than manually moving between systems and repeatedly rewriting queries, an agent can work across those authorised sources and help investigate the underlying pattern.

The same analytical model raises a more commercially useful question for Sydney service, property and renovation businesses: why are apparently viable quotations failing to become booked work?

That question sounds like a sales question. In practice, it is frequently an operations question.

A quote may be commercially competitive but still fail because the required start date cannot be accommodated. An apartment project may appear straightforward until lift bookings, loading access, waste removal or strata conditions emerge.

A flooring quote may be revised because the substrate requires grinding or levelling. A client may be willing to proceed but wait too long for a revised scope. A project may be recorded as “lost” even though it has simply moved into a later programme.

The data needed to distinguish those outcomes rarely lives in one column.

Data Agent Kit Is Not a Plug-and-Play Sales Dashboard

This distinction matters.

Google describes Data Agent Kit as a preview product for data scientists, data engineers and data application developers.

Its supported environment includes services such as BigQuery, Cloud SQL, AlloyDB, Spanner, Cloud Storage, Dataflow and Managed Service for Apache Spark.

It does not mean a business can install the kit and immediately ask questions across every CRM, accounting package, quoting platform, inbox and project-management system it uses.

The operational data must first be available through an authorised and technically supported data architecture.

Customer records may need to be loaded into a warehouse. Quote versions may need to be normalised. Scheduling data may require a reliable project identifier. Lost-job reasons may need to be structured rather than buried inside notes.

Data Agent Kit can reduce the friction between the analyst and that data estate. It does not remove the work required to make the estate coherent.

The Conversion Rate Is Not One Number

Management dashboards commonly reduce quote performance to a simple ratio:

Accepted quotes ÷ issued quotes = conversion rate.

That number is useful for trend monitoring, but poor at diagnosis.

A Sydney renovation operator could have a 30 per cent overall conversion rate while producing completely different results across apartment work, occupied homes, commercial projects, insurance remediation, concrete preparation and straightforward installation work.

The commercial question is therefore not simply whether conversion has fallen. It is where the fall occurs and what operational conditions accompany it.

Are slower quotations converting less often?

Evidence required: enquiry time, inspection time, quote issue time and acceptance time.

Possible operational response: review estimating capacity and approval delays.

Are heavily revised quotes losing momentum?

Evidence required: quote version count, revision reason, revised amount and days between versions.

Possible operational response: improve site discovery before initial pricing.

Are apartment projects converting below houses?

Evidence required: property type, level, lift access, parking, strata requirements and waste route.

Possible operational response: move access qualification earlier in the process.

Does requested timing conflict with available capacity?

Evidence required: requested start date, first available date, crew capacity and project duration.

Possible operational response: change scheduling communication or resource planning.

Does conversion fall when floor preparation is added?

Evidence required: removal, grinding, adhesive removal, levelling and installation line items.

Possible operational response: review how substrate risk and preparation costs are explained.

Are some lead sources producing high quote volume but little work?

Evidence required: lead source, qualification status, quote value, acceptance and cancellation.

Possible operational response: reallocate marketing or qualification effort.

Sydney Project Conditions Can Distort What Looks Like a Pricing Problem

Property services are particularly difficult to analyse because the quoted product is often not a standard product.

Consider flooring removal in a third-storey Sydney apartment. The customer may initially compare the price against a ground-floor house, yet the contractor may need to account for restricted working hours, stair transport, building protection, loading restrictions, disposal logistics, parking and additional labour.

Alternatively, consider an existing timber floor that has been removed to reveal unexpected adhesive contamination or a slab outside installation tolerance. Concrete grinding or floor preparation may need to be added.

A basic pipeline might record the revised quote as more expensive and later mark the opportunity as lost.

A stronger operating dataset could show something different:

• The initial site information was incomplete.
• A second inspection was required.
• The quotation increased after substrate preparation was identified.
• The client's preferred construction window was approaching.
• The revised quote was issued four days later.
• The next available installation date no longer fitted the client's programme.

Pricing is part of the story. It may not be the root cause.

A Better Quote-to-Job Record Looks More Like a Project File

To investigate conversion properly, businesses need an evidence chain from initial enquiry to final commercial outcome.

A useful record could include:

1. Enquiry context: Service, suburb, property type, source and requested timing.
2. Qualification: Photographs received, plans available, site inspection required and decision-maker identified.
3. Site conditions: Access, occupancy, strata requirements, parking, loading, waste handling and known substrate risks.
4. Quotation: Original value, scope, exclusions, issue date and quote validity.
5. Revisions: Amount changed, scope changed, reason and turnaround time.
6. Capacity: Requested commencement, first available commencement and estimated project duration.
7. Follow-up: When contact occurred and whether further technical or commercial questions were raised.
8. Outcome: Accepted, declined, deferred, duplicate, out of scope, price objection, timing mismatch or genuinely unknown.

Elyment's existing work on enquiry routing between email, CRM and quoting systems deals with getting complete information into the commercial process.

Quote-conversion analysis begins one stage later. It asks what the business can learn after a valid quotation has actually been issued.

What AI Could Find That a Weekly Pipeline Meeting Might Miss

The attraction of agentic analytics is not that AI can pronounce a definitive reason for every customer decision.

Its value is the ability to interrogate a larger number of operational relationships than a manager can comfortably test by hand.

A data practitioner might investigate:

• Whether conversion changes by days between enquiry and quotation.
• Whether projects requiring more than one quote revision convert differently.
• Whether conversion declines as the gap between requested and available start dates increases.
• Whether specific property types repeatedly generate access-related revisions.
• Whether high-value quotes behave differently from smaller jobs after controlling for project type.
• Whether certain services produce more deferred opportunities rather than genuine losses.
• Whether quote follow-up timing has any measurable association with acceptance.
• Whether a particular period of low conversion coincided with unusually high operational backlog.

Those questions move management away from anecdotes such as “customers think we are too expensive” and towards testable operating hypotheses.

An Illustrative Sydney Conversion Review

Consider a hypothetical property-services business that issued 300 quotes over six months.

This is an illustrative scenario, not Elyment operating data.

Its headline conversion rate is 30 per cent. Management initially concludes that pricing must be too high.

A deeper analysis finds several clusters:

• Quotes issued within two business days of a completed inspection convert materially better than quotes requiring longer internal preparation.
• Projects with an unidentified access requirement produce more revisions.
• Opportunities where the requested commencement date is substantially earlier than available capacity are disproportionately recorded as lost.
• Jobs requiring additional levelling after inspection produce a lower acceptance rate when the extra preparation is introduced as a late revision.
• A group classified as lost actually contains customers who postponed work rather than selected a competitor.

None of those findings proves why an individual client made a decision.

Collectively, however, they could change management's response. Instead of discounting every quote, the business might improve inspection completeness, shorten revision turnaround, distinguish deferred work from genuine losses and expose realistic scheduling windows earlier.

That is a very different intervention from simply lowering prices.

The Most Important Field May Be the One Nobody Records

AI analysis becomes unreliable when operational statuses are designed for administrative convenience rather than management truth.

A CRM containing only open, won and lost may conceal commercially important states such as:

• Awaiting customer decision.
• Awaiting building approval.
• Awaiting revised scope.
• Client project postponed.
• Timing unavailable.
• Budget deferred.
• Duplicate opportunity.
• Outside service scope.
• No longer contactable.
• Competitor selected.
• Reason genuinely unknown.

Collapsing those outcomes into one “lost” category creates a dataset from which even technically sophisticated analysis can produce a commercially misleading answer.

The lesson is familiar from Elyment's analysis of automation economics: activity is not the same as completed business value.

The measurement layer has to represent the real operating outcome.

Scheduling Data Belongs in Sales Analysis

One of the most useful shifts for project businesses is to stop treating sales and delivery as independent datasets.

A client buying renovation work is also buying a place in a programme.

If the enquiry says work is required before settlement, before tenants return, during a school holiday period or before another contractor mobilises, capacity can become as important as price.

A quote system that knows the quoted amount but not the requested start date cannot measure that relationship.

The same applies to site access. A customer may appear to reject a quotation after the price changes, when the underlying problem was that the original scope assumed normal access and the building later imposed a narrow loading window.

For Sydney operators managing flooring removal, concrete grinding, levelling, installation, painting and broader renovation coordination, delivery variables therefore belong inside the commercial dataset.

What Data Agent Kit Could Change for the Analyst

Google's Data Agent Kit is designed to reduce several technical barriers in this kind of investigation.

Its tools can work with supported Google Cloud resources through Model Context Protocol connections, while its agent skills assist with activities such as querying, transformation, validation and data engineering.

In a properly designed environment, a data team could move through a process such as:

1. Verify the baseline quote and acceptance figures.
2. Segment the results by project, property and operational characteristics.
3. Join quotation data to scheduling or customer records.
4. Identify missing values, duplicates or inconsistent classifications.
5. Test specific commercial hypotheses.
6. Turn useful exploratory analysis into a reproducible data model.
7. Validate the model against known projects and management records.
8. Publish controlled reporting for operational decision-makers.

Google's own September 2026 demonstration is particularly relevant because it moves from a one-off root-cause investigation into a reusable dbt model.

That is the point at which agentic analytics becomes operationally interesting.

The business no longer has to rediscover the same problem every month.

AI Can Find Associations. Management Still Has to Establish Cause.

There is an important limit.

Suppose the system finds that expensive quotations convert less frequently.

It would be easy to conclude that reducing prices will improve conversion. But higher-priced jobs may also be larger, more disruptive, technically complex, subject to more approvals or harder to schedule.

Price may be correlated with the actual source of friction rather than causing the outcome by itself.

Similarly, the system might find that site-inspected jobs convert better than remotely quoted work.

That does not automatically prove site inspections create the improvement. Businesses may simply choose to inspect opportunities that were already better qualified.

Good commercial analysis therefore needs three layers:

• Descriptive analysis to establish what happened.
• Diagnostic analysis to identify relationships worth investigating.
• Operational verification to decide which relationship is credible enough to change the process.

AI can accelerate the first two. Experienced people still own the third.

Privacy and Permission Design Cannot Be an Afterthought

Quote data can contain names, addresses, phone numbers, correspondence, property photographs, access instructions and other personal information.

The Office of the Australian Information Commissioner advises organisations using AI products involving personal information to consider necessity, purpose, access, human oversight, accuracy, privacy risks and security as part of deployment.

Connecting more business data should therefore not mean giving every agent access to everything.

The data used for conversion analysis should be limited to what is required for the defined analytical purpose, with access controlled according to role.

NSW Government agencies have additional mandatory requirements under the NSW AI Assessment Framework.

That framework does not automatically impose the same requirements on private Sydney businesses, but its emphasis on risk identification, accountable ownership, security, privacy, documentation and lifecycle review provides a useful governance benchmark.

Build the Measurement System Before Automating the Response

Once management sees a conversion problem, the temptation is to automate immediately.

Send more reminders. Generate faster follow-ups. Automatically score leads. Trigger discounts. Ask AI to decide which opportunities deserve attention.

That sequence is backwards if the root cause has not yet been established.

A stronger implementation sequence is:

1. Define the commercial outcome. Decide precisely what counts as issued, accepted, deferred and lost.
2. Map the evidence. Identify which systems hold enquiry, quote, schedule, revision, project and customer data.
3. Standardise the identifiers. Make sure the same opportunity or project can be reconciled across systems.
4. Repair the missing fields. Add the operational variables management actually needs.
5. Analyse before intervening. Test where conversion changes and whether the relationship is commercially credible.
6. Change one part of the workflow. Improve the process rather than launching multiple automations simultaneously.
7. Measure the result. Compare later cohorts against the original baseline.

Businesses without this foundation may benefit from an AI readiness and operational data assessment before connecting AI tools to live quoting, scheduling or customer systems.

The Better Question Is Not “Why Did We Lose This Quote?”

Individual customers do not always provide reliable explanations. Some decline politely. Some stop responding. Some postpone. Some compare alternatives. Others change the project entirely.

Asking AI to invent an explanation for each lost quotation would create false precision.

A more defensible management question is:

Which measurable conditions repeatedly appear before quotation outcomes change, and which of those conditions can the business operationally improve?

That framing protects the analysis from becoming amateur psychology.

It also directs attention towards factors management can actually control: response time, completeness of scope, capacity communication, revision turnaround, qualification quality, project sequencing and follow-up discipline.

From Data Question to Operating System

The strongest use of connected analytics is not a spectacular one-off answer. It is a repeatable management system.

A well-designed quote-conversion model could allow an operations team to review:

• Rolling conversion by service and project type.
• Average time from qualified enquiry to quote.
• Percentage of quotations requiring revision.
• Conversion by requested versus available start date.
• Volume of genuinely lost versus deferred opportunities.
• Conversion after site inspection.
• Conversion where access information was complete before pricing.
• Quote value alongside gross delivery capacity rather than in isolation.

This is where technology starts to support physical project delivery rather than sit beside it.

Elyment's operating model spans physical works, project coordination, compliance-sensitive processes and digital systems. That makes the commercial data useful only when it reflects how jobs actually move from enquiry to inspection, pricing, acceptance, mobilisation and delivery.

Find the Leak Between Quote and Job

Review your quote data, workflow handoffs, scheduling constraints, project requirements and AI readiness before automating the next stage of the pipeline.

Request an Operations & AI Review

The Bottom Line

Google's Data Agent Kit points towards a more useful phase of business AI: not simply generating content, but helping data practitioners investigate operational questions across authorised enterprise data.

For Sydney property, construction and renovation operators, quote conversion is an especially useful test case because the answer rarely lives inside the quote itself.

Pricing, project complexity, access, strata conditions, response time, revisions, customer timing and operational capacity can all influence whether a quotation becomes booked work.

The opportunity is therefore not to ask AI to guess why customers said no.

It is to build enough operational evidence that management can see where the commercial process actually changes, investigate why, and improve the parts of the system it can control.

Sources and References

• Google Cloud: Agentic analytics with the Data Agent Kit
• Google Cloud: Data Agent Kit overview
• Elyment: AI flooring quotes and smarter intake
• Elyment: Enquiry routing between email, CRM and quoting systems
• Elyment: Automation economics and measuring business value
• Office of the Australian Information Commissioner: Privacy and commercially available AI products
• NSW Government: NSW AI Assessment Framework
• Elyment: AI readiness and operational data assessment