Skip to content
Iván Pintor
Leer en español

Iván Pintor · Product, data and automation

I bring order to the way a company works.

For three years, at an accounting and advisory firm with four offices and more than 500 clients, onboarding each client went from seven programs to a single form.

See projectsDownload CVPDF

At a glance

building the technology function
3 years
in Artificial Intelligence
Certified
in applied AI for companies
Trainer
clients in the CRM I managed
500+

The journey

7 projects, from start to finish

Scroll down to go through them. At each stop you will see the problem, the solution and the result. Where there is a demo, you can take a detour to try it and then return to the route.

  1. Stop 01 · Automation

    Sales cycle automation: from seven systems to one

    End-to-end automation from quotation to client onboarding, with client data entered only once.

    • Bitrix24
    • CRM automations
    • API integrations

    How it works

    1. Client data

      Captured once, at the start of the process.

    2. Quotation

      With the agreed services and fees.

    3. Sent to the client

    4. Signed quotation

      The signed return triggers every set-up.

    5. Automatic set-up in seven systems

      CRM, accounting, payroll and billing ERPs, document manager, Mailchimp and control spreadsheet.

    6. Active client

      One point of data entry instead of seven.

    1. Situation

      At an accounting and advisory firm with four offices and more than 500 clients, onboarding a new client meant entering the same data into seven different systems. Every manual copy was a chance to make a mistake, and the inconsistencies between systems only surfaced later, in billing or in case management.

    2. Solution

      I designed and put into production an automated CRM process that chains quotation, sending and onboarding. The client details and agreed services are captured once; when the signed quotation comes back, the process sets the client up in the CRM, the accounting, payroll and billing ERPs, the document manager, Mailchimp and the control spreadsheet, without anyone having to retype anything.

    3. Result

      Data entry went from seven systems down to one. Onboarding time dropped dramatically and the inconsistencies between systems disappeared.

    Try the demo See the full case

  2. Stop 02 · Processes

    Subsidy processing with conditional logic

    A form with conditional questions, automatic calculation of the grant amount and follow-up of every case until it is closed, so none is ever forgotten.

    • Bitrix24
    • Conditional-logic forms
    • QR code
    • Rules engine

    How it works

    1. QR access

      The form opens by scanning a QR code, on a phone or a computer.

    2. Conditional form

      Each answer decides the next question: contract type, eligible group, degree of adaptation and percentage.

    3. Rules engine

      Exact grant amount according to the scheme criteria, plus the contract end date.

    4. Automatic notifications

      Billing and the account manager receive the data they need.

    5. Decision point

      Case closed on time?

      • No

        Escalating reminder: consultant, then team lead, then management.

      • Yes

        Closed as won or lost; records are updated.

    6. Recorded in the dashboard

      Every subsidy is visible to management.

    1. Situation

      Regional employment subsidies were handled manually: interpreting the scheme criteria case by case, calculating the amount by hand, and coordinating consultants, billing and management by email. There was a risk of calculation errors and of cases falling through the cracks.

    2. Solution

      A QR-accessible web form with conditional logic: each question appears depending on the previous answers (contract type, eligible group, degree of adaptation and percentage). It is connected to a CRM process that moves through its stages on its own. It includes a rules engine that calculates the exact grant amount against every criterion of the scheme, automatic calculation of the contract end date, and scheduled reminders that escalate to the consultant, then the team lead, then management if the case is not closed. Billing and the account manager are notified automatically with the data they need, and closing a case as won or lost updates the records and notifies the changes.

    3. Result

      Consistent calculations with no human error, no case left without follow-up, and a complete record of the subsidies obtained, feeding into the management dashboard.

    Try the demo See the full case

  3. Stop 03 · Applied AI

    Anonymised AI-assisted candidate assessment

    A web application for online tests with AI-assisted scoring, designed so that the model never knows who the candidate is.

    • In-house web app
    • LLM APIs
    • PDF reports
    • GDPR
    • EU AI Act

    How it works

    1. Online test

      The candidate answers in the browser.

    2. Anonymisation

      The answer is separated from any data that identifies the candidate.

    3. AI-assisted scoring

      The model only receives the answer and the reference model answer.

    4. Result linked back

      The score is matched to the candidate inside the application, never outside it.

    5. PDF report

      Comparable across candidates, for the hiring team to decide.

    1. Situation

      Recruitment relied on paper-based tests and subjective judgement, with no consistent criteria across candidates and no structured record of the results.

    2. Solution

      I developed an in-house web application with online tests, automatically generated PDF reports and AI-assisted scoring against model answers defined in advance. The architecture keeps the candidate anonymous to the model: it only receives the answer it has to assess, in line with the GDPR and the EU AI Act.

    3. Result

      Consistent, traceable assessment criteria for every candidate, with comparable reports and no personal data exposed to third parties.

    See the full case

  4. Stop 04 · BI & data

    Management dashboard

    A business intelligence system on top of the CRM: panels for each area and lists for consultants, with real-time data and comparisons with previous years.

    • SQL (Trino)
    • Apache Superset
    • Bitrix24

    How it works

    1. CRM data

      Companies, work-group tasks and deals.

    2. SQL datasets

      Trino queries that clean, join and aggregate the data in real time.

    3. Panels per area

      Overview, accounting and tax, payroll and administration, comparable with three previous years.

    4. Lists for consultants

      Filters to prepare the monthly allocation in seconds.

    5. Management decides with instant data

      No more report requests, and the same report format as before.

    1. Situation

      Management had no consolidated view of the client portfolio, workload or profitability. Everything came from a tracking spreadsheet and from lists prepared by hand every month: per consultant, new and lost clients, extras per department and hours spent on shared tasks.

    2. Solution

      I built a business intelligence system on top of the CRM, with three SQL datasets read in real time (companies, tasks and deals) and panels for each area: overview, accounting and tax, payroll and administration. I added filtered lists so each consultant can prepare their monthly allocation, and CSS templates keep the reports in the format management already used.

    3. Result

      Management gets instantly what it used to request as reports, and can compare with up to three previous years, which was not possible before. Both departments save several hours a month on preparing lists.

    Try the demo See the full case

  5. Stop 05 · CRM/ERP

    Minutes workflow

    Generation, sending, follow-up and filing of minutes automated in the CRM, with automatic reminders and the status of every case at a glance.

    • Bitrix24
    • Document generation
    • Digital certificate signing

    How it works

    1. Minutes generated

      The document is created from the case, with its data.

    2. Sent to the company

    3. Decision point

      How does the client respond?

      • Returns them signed

        The system detects it and moves straight to filing.

      • Asks for certificate signing

        The workflow branches to digital certificate signing.

      • No reply

        Two automatic reminders and, if there is still no answer, an alert for the consultant to call.

    4. Automatic filing

      The signed minutes are stored in the right place with no manual step.

    5. Status visible in the case

    1. Situation

      Minutes were drafted, sent and tracked by hand. Finding out which ones had come back signed meant searching through email.

    2. Solution

      A complete workflow in the CRM: the minutes are generated from the case and sent to the company. From there, the client's response decides the path: if they return them signed, the system detects it and files them automatically; if they ask us to sign with a digital certificate, the workflow branches to that signature; and if they do not reply, the automation sends two reminders and, if there is still no answer, alerts the lead consultant to call them.

    3. Result

      The status of every case visible at a glance, and no more manual follow-up by email.

    Try the demo See the full case

  6. Stop 06 · CRM/ERP

    Segmented campaigns to the client base

    Client communications centralised and segmented by profile, with open and click tracking used to improve every send.

    • Mailchimp
    • Bitrix24
    • Segmentation
    • Open and click metrics

    How it works

    1. Client data in the CRM

    2. Segmentation

      By profile, type of service and client status.

    3. Campaign sent

    4. Measurement

      Open and click rate of every send.

    5. Next send fine-tuned

      Subject line, content and timing.

    1. Situation

      Client communications (income tax return campaigns, new subsidy schemes, service changes) were sent in a scattered way, with no segmentation criteria and no measurement afterwards.

    2. Solution

      I centralised campaign management, first in Mailchimp and later inside the CRM itself, to make use of the data already held on each client. Sends are segmented by profile, type of service and client status.

    3. Result

      The 2024 and 2025 income tax return campaigns and successive subsidy schemes were communicated in a segmented way, with open and click rates tracked to fine-tune the subject line, content and timing of the next ones.

    See the full case

  7. Stop 07 · Applied AI · Own prototype

    AI waiter agent and kitchen display

    Two-agent prototype for restaurants: an AI waiter that understands orders in plain language and an AI-free kitchen, organised by section, that queues and times every order and analyses each service.

    • Claude Haiku 4.5 (Anthropic API)
    • Tool use
    • Netlify Functions
    • Preact
    • Web Speech API
    • Deterministic simulation

    How it works

    1. The guest orders in plain language

      “One margherita without basil and two beers.”

    2. The AI waiter checks the menu and the kitchen

      Through tools: it never answers from memory. It knows the wait in each section and warns guests when the kitchen is busy.

    3. Decision point

      Is everything on the menu?

      • Yes

        It summarises the order and asks for confirmation.

      • No

        It says so clearly and offers what is available.

    4. Order validated against the menu

      Dishes, quantities and allowed changes. Only menu data reaches the kitchen, never free text.

    5. AI-free kitchen: sections, queue, timers and voice

      Each dish goes to its section, gluten-free dishes to their own area, drinks to the bar. At peak time, priority by dish; when the wait goes up, an automatic notice to the online platforms. Guests see what stage their order is at.

    6. Order served and recorded

      Each dish records when it arrives, starts and is ready, its section and the crew size: the basis for the service analysis.

    1. Situation

      In a busy restaurant, an order passes through several hands: the waiter jots it down in a hurry, allergen questions depend on what they remember, and the kitchen receives orders with no clear sequence and no idea how long each has been waiting. At peak times, every mistake ends up on the table. And the crew is allocated to sections from memory, with no data on where the kitchen gets stuck.

    2. Solution

      Two agents with distinct roles. In the dining room, an AI waiter understands the order in plain language and uses tools to check the menu and the state of the kitchen and to create the order: everything it says about dishes, prices, allergens and waiting times comes from data and rules, and the order is validated against the menu before reaching the kitchen. The kitchen uses no AI: queue, timers and priorities are deterministic logic, predictable, auditable and with no per-use cost. Each dish goes to its section (pizza oven, hot line, salads and cold starters, desserts and wash-up), and gluten-free dishes go to a separate area run only by the designated person. At the start of service each employee is assigned a section, and capacity and waiting times are worked out per section. At peak time the kitchen prioritises by dish; when the wait goes up, a rule automatically notifies the online ordering platforms (via API or webhook) of the estimated wait, and the waiter tells guests in the dining room. Guests see what stage their order is at (in the oven, nearly ready, ready), and the waiter knows it too if they ask. Every dish is recorded, and the service analysis shows demand, bottlenecks and the best crew allocation, always by section and never by person. AI is used only where it adds value: understanding people and talking to them. Rules make the decisions.

    3. Result

      A working prototype you can try in the browser: from an order in plain language to a ticket in the kitchen, read aloud, with allergen information always taken from the menu and the usage and cost limits a public-facing service needs. In a simulated peak with fictitious data, automatic priority gets 8 of 10 tables served on time, against 7 of 10 in order of arrival, without lengthening the longest wait (19 minutes in both cases). No online order is turned away, and all 4 are ready within the time quoted to the customer; with the platform's usual time, only one would have been. Over eight weeks of simulated services, the analysis points to the pizza oven as the bottleneck: with a single person, 56% of pizzas run late, against 7% with two. And the crew planner finds that on a Friday night with 7 people it pays to put 3 on pizzas and 2 on the hot line (83% of orders on time) rather than 2 and 3 (81%).

    Try the demo See the full case

End of the journey

These are the featured projects. There are more in the full list.

See all 10 projects

How I work

  1. Identify

    Where time is lost or errors are made, and what it costs.

  2. Design

    The process and the solution, prioritised by cost and benefit.

  3. Build

    The automation, dashboard or application, all the way to production.

  4. Train

    The people who will use it, and measure whether it really works.

Let’s talk

I’m looking for a role with real responsibility in product, data or digital transformation. If you have one, or a process that could work better, get in touch.

Get in touch LinkedIn