What problem does it solve?
In many business to business and project sales, the quote is the bottleneck. A distributor receives a customer's list of thousands of items in the customer's own descriptions and has to match each one to its own catalog. A solar installer has to measure a roof from a satellite image before it can say how many panels fit. A steel or building products supplier has to read architectural drawings to know what to price. The work is skilled, slow and repetitive, so customers can wait days for an answer.
Manual quoting also produces errors: a miscounted roof, a wrong substitute product, a price rule applied inconsistently. dida, which built the roof assessment step of Enpal's quotes, describes Enpal's old process as taking a salesperson 120 minutes per quote and as error prone, leading to inaccurate projections of cost and energy production.
How does it work?
- Read the request. The AI reads what the customer sent: an email, a spreadsheet of items, a PDF drawing or an image of a roof or site, and extracts items, dimensions and quantities.
- Match to the catalog. Each item is matched to the organization's own products or configurable options, with a confidence score and alternatives where no exact match exists.
- Calculate. Quantities, dimensions and technical constraints are calculated with deterministic rules or models (for example the usable roof area and the number of panels), not left to a language model's arithmetic.
- Price. List prices, customer contracts, discounts and margin rules are applied from the pricing system, with anything outside the seller's authority flagged for approval.
- Review and send. The seller or estimator checks the draft, adjusts it, and sends the quote from the CRM or quoting system; accepted and rejected quotes feed back into matching.
- Audience
- Employee facing
- Autonomy
- Copilot
- Adoption
- Early adopters
- Channels
- Internal tools, Microsoft Teams, Email
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Handling time reduction | Too few to pool | 87.5% | 1 | 1 vendor |
| Users served | Not pooled | 150 | 1 | 1 vendor |
Value drivers: Speed and cycle time, Revenue growth, Employee productivity, Risk and loss reduction.
Indicative value
A distributor or installer that issues 10,000 custom quotes a year
USD 160,000 to USD 1.1 million
Quoting time released per year
How this is calculated
Formula: quotes * minutesPerQuote / 60 * reduction * hourlyCost. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Custom quotes per year quotes, quotes per year | 10,000 | 10,000 | The reference organization. |
| Staff time per quote today minutesPerQuote, minutes per quote | 60 | 120 | Editorial assumption; dida reports 120 minutes per solar quote at Enpal before automation. Replace with a time study of your own quotes. |
| Share of quoting time saved reduction, fraction of time | 0.4 | 0.8 | Conservative against the benchmark on this page (dida reports an 87.5% reduction at Enpal), because most catalogs are less uniform than solar roofs. |
| Fully loaded cost of a seller or estimator hourlyCost, USD per hour | 40 | 70 | Editorial assumption, replace with your own cost. |
What it leaves out: Time value only. It leaves out the build and integration cost, and the revenue effect of faster quotes, which Microsoft reports as 20% more sales opportunities a quarter at ODP but which depends on the market.
Who already uses it?
3 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
DeAcero
Mexico · Manufacturing · 2026
DeAcero, a Mexican steel producer, built agents on Google's Gemini Enterprise Agent Platform with multimodal models that analyze architectural blueprints in PDF format and turn them into detailed cost proposals for customers. Google Cloud reports that this sharply cut DeAcero's response time to customers but gives no figure.
No outcome disclosed.
The ODP Corporation
United States · Retail and ecommerce · 2025
The ODP Corporation, parent of Office Depot and ODP Business Solutions, built a sales assistant on Azure OpenAI and Azure AI Search that lets representatives generate quotes in natural language. Its SKU matching cross references large product lists in minutes instead of days and produces quote ready summaries and tables; in one anecdote a representative cross referenced 4,000 competitor items. Microsoft reports that pricing bids now take hours instead of one to two days, that tailored quotes save representatives five to eight hours a week, and that the assistant drives 20% more sales opportunities a quarter.
No outcome disclosed.
Enpal
Germany · Energy and utilities · 2024
Enpal, a German solar energy company, worked with the AI firm dida to automate the roof assessment and panel sizing step of its solar quotes, which a salesperson had done by hand. A model trained on rooftop images from the Google Maps Platform detects the usable roof area and obstacles, projective geometry estimates the roof angle, and further steps calculate the number of panels and visualize their placement. dida says that during the six month build Enpal was able to manually adjust details such as the dimensions of a roof. The tool sizes the installation rather than setting prices. dida reports that the process now takes 15 minutes instead of 120 and that 150 Enpal employees use the tool.
- Handling time reduction: 87.5%, staff time per solar quote, from 120 to 15 minutes
"Thanks to our solution, and the efficiency of building it with Google Cloud, what was once a manual process taking an Enpal salesperson 120 minutes to complete is now an automated process of just 15 minutes: a reduction of 87.5%."
Claimed by: vendor - Users served: 150, Enpal employees using the tool, four years after launch
"Four years on, this has risen to 150 Enpal employees, each saving 87.5% of their time, which they can now dedicate to other, more specialized tasks."
Claimed by: vendor
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- A product catalog with structured attributes and approved substitutes
- Price lists, customer contract prices and discount rules in a system of record
- A set of past requests with the quotes that were sent, to measure matching accuracy
- Technical rules for configuration and quantity calculation
Systems to integrate
- CRM or configure, price, quote (CPQ) system
- ERP or pricing engine for prices, stock and margins
- Document and email intake for requests
- Mapping or imaging services where the quote depends on a site
Complexity: Medium
Reading the input is usually the easier part; the difficulty is a clean product catalog with attributes good enough to match against, and a pricing system the AI can call instead of guessing prices.
- 1
Pick one quote type with volume
Start with a quote type that is frequent and fairly standard, such as a list of catalog items or a single product family, and measure the minutes and errors per quote today.
- 2
Keep prices out of the model
Let the AI match and count, and let the pricing system price. Every price in the draft should come from a call to the system of record, never from generated text.
- 3
Measure matching accuracy
Use past requests and the quotes actually sent as a test set, and report precision per product family before sellers rely on it.
- 4
Show confidence and alternatives
Mark low confidence matches and unusual quantities so the seller checks those lines first, and keep the seller able to adjust anything.
- 5
Learn from sent quotes
Feed the seller's corrections and the quote outcome back into matching and into the test set, and review the lines sellers change most often.
Guardrails
- Prices, discounts and availability only from the pricing and ERP systems, never generated
- Discounts or margins outside a seller's authority routed for approval
- Human review of every quote before it is sent to a customer
- Technical calculations done by deterministic rules or validated models, with the inputs shown
KPIs to instrument
- Minutes of staff time per quote and elapsed time from request to quote
- Line level matching accuracy and the share of lines sellers change
- Quote to order conversion, before and after
- Quotes with pricing or quantity errors found after sending
Human in the loop
The seller or estimator reviews and sends every quote and owns the price. Pricing managers own the rules and approve exceptions, and product specialists maintain the catalog attributes and approved substitutes the matching relies on.
Common failure modes
- Plausible but wrong matches
- The AI picks a similar product that does not meet the customer's specification. Show confidence per line and test matching on past quotes.
- Invented prices
- A language model fills in a price or discount it was never given. Keep pricing in the system of record and block generated prices.
- Garbage catalog in, garbage quote out
- Matching fails because product attributes are incomplete. Fix the catalog data first; product content enrichment helps here.
- Speed without margin control
- Faster quotes with inconsistent discounts erode margin. Enforce discount authority in the workflow, not in the prompt.
What are the risks and rules?
EU AI Act
Minimal risk
Drafting business quotes for a seller to review is not an Annex III use and does not interact with the customer as an AI system. It would need a fresh assessment if the system set individual consumer prices or terms in areas such as credit or insurance, where Annex III point 5 can apply.
Guidance
- Algorithms: how they can reduce competition and harm consumers (UK Competition and Markets Authority, Europe). Describes how pricing algorithms and personalized pricing can harm competition and consumers, relevant when quote tools also set prices.
Controls to put in place
- Pricing rules and discount authority enforced in the quoting system, with an audit trail
- A record of the AI draft and the seller's changes for every quote sent
- Periodic review of matching accuracy and of quote errors reported by customers
- Data protection review when quotes use images or data about a customer's home or site
Frequently asked questions
- Can AI produce a customer quote without a salesperson?
- For simple, standard requests it can draft the whole quote. At Enpal the tool is used by Enpal staff and at ODP by sales representatives; DeAcero's public description does not say how its proposals are reviewed. The safer design keeps a seller reviewing every quote and takes prices from the pricing system, never from a language model.
- How much faster does AI make quoting?
- It depends on the input. dida reports that Enpal's solar quote went from 120 minutes to 15, a reduction of 87.5%, and Microsoft reports that ODP Business Solutions now returns pricing bids in hours instead of one to two days.
- Is this the same as CPQ software?
- It sits in front of it. Configure, price, quote systems hold the rules and prices; the AI reads unstructured customer requests and fills the quote, so sellers spend less time on data entry.
How to cite this page
Blits.ai AI Use Case Library, "AI quote and estimate generation from customer requirements", last verified 27 September 2026, https://www.blits.ai/ai-use-cases/sales-quote-and-estimate-generation. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 27 September 2026: First published