INDUSTEC: measuring the ROI of an AI quotation assistant

INDUSTEC is an illustrative example that uses explicit assumptions about an industrial quotation process to explain a measurement method connecting lead time, quality, adoption, capacity, margin and full cost.

• The volumes, durations and costs below illustrate the calculation; they are not presented as observed outcomes.
• The protocol separates operational performance from its financial translation.
• The final decision may be to expand, correct or stop the system.

Define the process and its baseline

A quotation assistant merits a pilot when drafting consumes scarce capacity, slows the commercial response and relies on sources that are structured enough to be used under control.

• Business friction: active drafting and rework time that occupies application engineers.
• Client effect: end-to-end lead time from a received request to an approved quotation.
• Available assets: quotation history, product catalogue, pricing rules, clauses and authorised CRM data.
• Accountability: the engineer designs the solution and approves the document; AI prepares a traceable draft.

To make the method calculable, the central scenario uses coherent assumptions drawn from the published scoping material without treating them as authenticated client data.

• Illustrative volume: 700 quotations per year, the midpoint of a 600–800 range.
• Illustrative starting time: 3 active hours per quotation, or 2,100 hours per year.
• Target population: 12 application engineers using one approval process.
• Working budget: €44,000 over twelve months, comprising €32,000 for support, €8,000 for licences and €4,000 for internal preparation.

Phase 1: establish a comparable baseline

The baseline should describe the process before assistance on a representative sample and retain the variables needed to compare quotations of similar difficulty.

• Perimeter: start at an adequate brief and finish at commercial and technical approval.
• Segmentation: renewal, standard variant, complex opportunity and cases containing an exception.
• Time: active time by contributor, waiting time, number of loops and total lead time.
• Quality: corrections to references, prices, clauses, warranties and technical content before sending.
• Commercial outcome: accepted quotation, realised margin and reason for loss where known.

The comparison remains interpretable when the pilot uses the same definitions, covers a sufficient period and controls for changes in sales mix, price and workload.

• Timestamp each stage consistently before and during the pilot.
• Compare assisted cases with matched historical cases or a contemporary control group.
• Separate seasonal, complexity, new-product and team-change effects.
• Publish sample sizes and missing data alongside every measure.

Phase 2: frame the pilot and its decision gates

The scoping document turns a technology intention into a falsifiable test by fixing users, data, controls, budget, thresholds and the decision date before development begins.

• Included use: produce a draft from a technical brief and authorised sources.
• Excluded use: define the solution, override pricing rules or send without human approval.
• Population: named users, business champions, process owner and budget sponsor.
• Thresholds: separate targets for time, quality, adoption, incidents and cost per accepted quotation.
• Decision: continue, correct or stop on a fixed date without turning a partial result into overall success.

A drafting-time target alone is inadequate because acceleration that increases corrections, errors or approval time merely moves the cost elsewhere.

• Active time: drafting, research, correction and approval combined.
• Quality: share of quotations acceptable at first review and severity of detected errors.
• Adoption: share of eligible users and cases actually assisted.
• Risk: unauthorised data, invented references, pricing deviations and confidentiality incidents.

Phase 3: prepare the foundations

The pilot becomes operationally useful only when data, permissions, architecture, skills and manual fallback form a controlled environment rather than a demonstration assembly.

• Corpus: cleaned quotations, versioned product catalogue, approved clauses and documented exception rules.
• Access: named identities, least privilege, managed secrets and usage logging.
• Architecture: supplier and hosting selected for the data, contracts, continuity and ability to exit.
• Training: system limits, permitted data, verification method and incident procedure adapted to each role.
• Fallback: manual process available if the model, integration or a critical source becomes unavailable.

Phase 4: pilot on representative cases

The assistant can prepare quotation structure, retrieve authorised references and apply standard rules while technical and commercial decisions remain under explicit human accountability.

• Generate a draft from a structured brief rather than a few free-text lines.
• Cite the source and version of every proposed product reference or clause.
• Block or escalate cases containing an exception, a missing source or low confidence.
• Record human corrections to measure quality and improve the corpus.

A monthly dashboard connects six families of measures to one decision so that higher usage or lower drafting time does not obscure either quality or the system's economics.

• Use: active users, frequency and coverage of eligible cases.
• Performance: active time, total lead time, rework loops and throughput.
• Quality: first-review pass rate, correction types and reviewer satisfaction.
• Risk: incidents, exceptions, blocked data and fallback operation.
• Cost: committed TCO and cost per approved quotation.
• Impact: redeployed capacity, attributable margin and the decision to continue, correct or stop.

From time saved to economic benefit

Under the central assumptions alone, an illustrative 50% reduction would lower 2,100 annual hours to 1,050, releasing 1,050 hours per year or 87.5 hours per month before adoption and rework are considered.

• Formula: 700 quotations × 3 hours × 50% = 1,050 hours per year.
• Adoption adjustment: at 70% of eligible cases, theoretical capacity falls to 735 hours per year.
• Quality adjustment: additional correction time must still be deducted.
• Limit: this calculation measures operational potential rather than a cash saving.

An hourly value can inform an internal decision, but it remains an economic convention until the hours reduce expenditure or produce demonstrable additional margin.

• At €35 per hour, 1,050 hours represent €36,750 of annual capacity.
• At €45 per hour, they represent €47,250 of annual capacity.
• At €55 per hour, they represent €57,750 of annual capacity.
• These three amounts are sensitivity assumptions with no claim of cash saved.

Calculate attributable ROI

ROI for a period equals (attributable benefits − total costs) / total costs, with a conversion rule for each benefit and an explicit ban on double counting.

• Cash savings: expenditure actually avoided, net of displaced costs.
• Released capacity: hours tracked separately until observable redeployment.
• Growth: attributable incremental margin on comparable opportunities, rather than revenue or pipeline.
• Risk: documented change in expected loss, with probability, impact and horizon.
• Costs: licences, integration, data, security, training, support, operations and internal time.

With the illustrative €44,000 cost, financial break-even requires €44,000 of attributable benefit over the same period; capacity valued at €47,250 passes that threshold in cash terms only if its effect materialises and is not counted elsewhere.

• Prudent case: recognise no financial benefit until redeployment produces avoided expenditure or measured margin.
• Central case: include only margin from comparable, traceable additional business in the numerator.
• High case: publish any more favourable assumption separately with its volume, margin, cost and adoption rate.

Attribute conversion without inventing causality

A change in conversion after the pilot does not by itself prove that the assistant created a sale because opportunity mix, prices, season, sales staff and the market may change at the same time.

• Compare cohorts with similar product, size, sector, channel and complexity.
• Measure signed and realised margin rather than gross ordered revenue alone.
• Document other commercial changes introduced during the period.
• Present a range or several scenarios where attribution remains uncertain.

Phase 5: consolidate only what the evidence supports

Production entry rests on thresholds met repeatedly and starts an operating loop that maintains quality, cost, security and value after each change of model, corpus or process.

• Industrialisation: controlled CRM integration, versions, evaluations, monitoring and support.
• Governance: business owner, register, periodic review, incident log and budget arbitration.
• Reversibility: tested model change and return to the manual process.
• Expansion: frame each new use case with its own baseline rather than extrapolating the quotation result.

Read the case through the Junyr Method™ Scale

A quotation assistant advances one specific capability, while the company's level depends on cumulative evidence across the framework's nine dimensions and cannot be inferred from either one tool or an isolated ROI.

• From Spectateur to Artisan: inventoried uses, known rules, priority case, owner and baseline.
• Towards Orchestre: recurring production use, fallback and monthly review of adoption, quality, incidents, costs and impact.
• Towards Architecte: register, permissions, logs, integrations, continuity and exit become operable across the platform.
• Nine dimensions to document: sovereignty, resilience, augmented staff, AI literacy, governance, technical debt, security, workflows and agents, value and FinOps.

What management can decide

At the end of the pilot, the decision committee should connect each conclusion to evidence and choose a proportionate option rather than pursue a spectacular headline figure.

• Expand: quality, risk, adoption, cost and impact thresholds met on representative cases.
• Correct: potential value confirmed but corpus, interface, training or control remains insufficient.
• Stop: attributable benefit too low, residual risk too high or full cost unsustainable.
• Reassess: repeat measurement after a material change of model, supplier, process or population.

INDUSTEC is an illustrative example whose volume, duration, budget and valuation assumptions explain the calculation method and allow the reader to test how different choices affect the result.

• Each business should replace these assumptions with its own baseline and full costs.
• The example explains the reasoning; a real investment decision requires measured benefits and costs.