See what we've automated
Real automations we've built, with the results measured afterwards. Every project is different, but they all share one thing: the manual work is gone.
This is what we've done
Every client has different needs. These are four real projects: the starting problem, what we built, and the results measured afterwards.
* Real cases. We omit client names and identifying details for confidentiality.
AI chatbot with a quoting engine built into the ERP
Each quote took 15 to 30 minutes of manual work: look the client up in the ERP, check pricing against their tier, verify stock by location, assemble the quote and generate the PDF. With three price lists, errors ran at 5-8%. Outside business hours, no enquiry was answered at all.
What we built
AI client portal
Clients check availability, pricing, spec sheets and stock in real time. The agent answers with data pulled straight from the ERP.
Guided 4-step quoting
Client, price tier, products with quantities, confirmation. A 3-5 minute conversation replaces half an hour of work.
Automatic ERP and CRM entry
Creates the CRM opportunity, generates the formal sale order, produces the PDF and leaves it ready to download.
RAG across the full catalogue
A knowledge base of 50,000+ products fed from the original sources and wired to the ERP over its API.
Measured results
| Metric | Before | After |
|---|---|---|
| Time per quote | 15-30 min | 3-5 min |
| Pricing errors | 5-8% | under 0.5% |
| Enquiries handled per day | around 30 | over 100 |
| Out-of-hours coverage | 0% | 35% of the total |
AI document management and data extraction
Companies receiving large volumes of documents by email —tax notices, supplier invoices, delivery notes— and processing them by hand: open the PDF, read it, pull out the key data, classify it, enter it into the accounting system and file it. Five to ten minutes per document, an 8-12% error rate, and documents lost along the way.
What we built
Email inbox sync
Automatic connection to Gmail, Outlook and IMAP, capturing every document the moment it lands, with no manual step.
OCR and structured extraction
An OCR-plus-AI pipeline that pulls out supplier, tax ID, line items, amounts, per-line taxes and the accounting code.
Automatic classification
AI categorisation into 16 document types, assigning urgency and response deadline based on the issuing body.
Entry into the accounting system
Creates the invoice in Holded with total verification, matches the supplier and attaches the original PDF for audit.
Measured results
| Metric | Before | After |
|---|---|---|
| Time per document | 5-10 min | 20-30 sec |
| Transcription errors | 8-12% | under 1% |
| Lost documents | 3-5% | 0% |
| Processing capacity | around 30 a day | unlimited |
Large-scale web data extraction with AI
Gathering information scattered across the web —spec sheets from dozens of suppliers, leads on professional platforms, competitor pricing— is unworkable by hand: copying and pasting hundreds of pages eats entire days, and the data is stale before you finish.
What we built
Supplier catalogue scraping
Extraction of specs, models, reference prices, availability and images from dynamic sites, handling pagination and authentication.
Qualified lead generation
Profiles and contact data from professional platforms, filtered by sector, size and location, and pushed into the CRM already segmented.
Competitor monitoring
Tracking of price, catalogue and availability changes, with alerts and history for trend analysis.
Scheduled, block-resistant pipelines
Daily or weekly runs with IP rotation and rate-limit handling, plus an alert if a source changes its structure.
Measured results
| Metric | Manual | Automated |
|---|---|---|
| Spec sheets processed per day | 20-30 | over 500 |
| Data freshness | weeks or months | daily or real time |
| Transcription errors | 5-10% | under 1% |
| Cost per record | hours of work | cents |
Sales funnel and CRM automation
A company running paid traffic on Meta, TikTok and YouTube generated hundreds of leads a month, but the sales process was chaotic: leads landed in the CRM unqualified, nobody knew which campaign they came from, they were assigned by hand, and there was no real visibility of any opportunity. The team wasted time calling leads that were never going to close.
What we built
Real-time automatic qualification
A webhook from the form triggers qualification against configurable business rules: available capital, location and product of interest.
Lead source recovery
Leads arriving without UTMs are enriched via API to recover campaign and ad. Of 255 leads with no source, 234 were recovered.
Automatic routing and deduplication
Qualified leads are distributed between SDRs and closers, with appointments booked by availability and dedup on email and phone.
Unified dashboard for management
The CRM syncs in real time with a secondary database: contacts, opportunities, appointments and team in a single view.
Measured results
| Metric | Before | After |
|---|---|---|
| Qualification per lead | 5-10 min by hand | instant |
| Leads with no identified source | around 35% | under 3% |
| Duplicate leads in the CRM | frequent | 0% |
| Pipeline visibility | partial | 100% in real time |
casosExito.cta.title
casosExito.cta.description