Implemented AI business solutions
AI agent for searching an internal candidate database
Semantic CV database search with candidate ranking and explanations.30–60 min → 5–10 secfinding relevant candidates
- finding relevant candidates
- 30–60 min → 5–10 sec
- relevant profiles through precise search
- +30-50%
- reduction in initial screening time
- up to 90%
Business challenge
Recruiters manually chose keywords and checked profiles. Traditional search struggled with meaning: suitable profiles were omitted because CVs used different wording.
Solution
The AI agent searches the internal CV database by meaning as well as exact keywords.
- Recruiter
- enters a free-form request: role, stack, experience, industry, seniority and location.
- Assistant
- finds relevant profiles, ranks candidates and explains why each appears in the results.
Architecture and integrations
Deployed in a private cloud with a local LLM inside a protected environment, with no CVs sent to external services. Connected to Huntflow through APIs and webhooks, synchronising approximately 200,000 CVs. Search combines semantic similarity, full-text matching and filters for experience, location and skills.
- SourcesCV database in Huntflow
- ProcessingSemantic search and filters
- OutcomeRanked profiles
Applications in other functions
- Sales
- similar customers, leads, deals, segments and cases
- Marketing
- campaigns, materials, research, audiences and insights
- Customer service
- similar enquiries, tickets, solutions and history
- Procurement
- suppliers, categories, terms, contracts and cases
Applications in other industries
- Retail and e-commerce
- products, categories, suppliers, alternatives and segments
- Financial services
- customers, applications, products and risk profiles
- Manufacturing
- procedures, process sheets, batches, defects and shifts
- Logistics and transport
- parts, batches, procedures and equipment
Restaurant analytics AI agent
Diagnosing revenue, occupancy and conversion variances from CRM and operational system data.250 hours/monthteam time freed up
- team time freed up
- 250 hours/month
- diagnosing variance causes
- 2–3 hours → 15–20 sec
- response to a deviation
- per day
Business challenge
Dashboards showed changes in bookings, occupancy, channels and cancellations but did not explain why. The team manually exported data and built breakdowns; diagnostics took 2–3 hours and depended on an analyst.
Solution
The AI agent accepts natural-language questions, clarifies parameters and runs step-by-step diagnostics using CRM and operational data.
- User
- asks why occupancy fell, channel conversion dropped or cancellations increased.
- Assistant
- selects metrics, segments and periods, identifies causes and produces management insights with visualisations.
Architecture and integrations
Cloud deployment connected to CRM and operational data through secure APIs. The Analyst Agent clarifies parameters and interprets results; the SQL Executor produces safe SQL queries. The process uses LangGraph with an explicit state graph and a dictionary of 100+ restaurant terms.
- SourcesCRM and operational data
- ProcessingLangGraph and SQL Executor
- OutcomeInsights and visualisation
Applications in other functions
- Finance
- revenue, margin, OPEX and cash flow variances by period
- Sales
- declines in funnel performance, conversion, average order value and channels
- Marketing
- campaign, promotion and traffic source effects on outcomes
- Operations
- deviations in KPIs, utilisation, SLAs and process quality
Applications in other industries
- Retail and e-commerce
- sales declines by store, category, channel and promotion
- Financial services
- customer churn, product profitability, rejections and channels
- Manufacturing
- output, defects, downtime, equipment utilisation and shifts
- Logistics
- delivery times, routes, carriers and delay causes
AI assistant for the sales department
Answers on rates, routes and delivery terms from corporate knowledge, with source references.up to 80%routine questions resolved without an expert
- routine questions resolved without an expert
- up to 80%
- faster customer responses
- 3–5×
- rates and wording
- consistent
Business challenge
Sales representatives spent substantial time finding current rates, routes, customs commodity codes, delivery terms and document requirements. Data was scattered across spreadsheets, instructions, documents and expert correspondence; answers depended on individual experience and carried error risks.
Solution
An internal AI assistant answers sales representatives' natural-language questions and searches the corporate knowledge base with source references.
- Sales representative
- asks about a rate, route, customs commodity code, cargo restrictions, timing or transport terms.
- Assistant
- finds relevant materials, produces a structured answer and shows the source.
Architecture and integrations
Cloud deployment with AI inside a protected environment. The RAG assistant connects to the knowledge base, rate tables and reference materials. Requests pass through vector retrieval and ranking before the LLM produces a source-linked answer.
- SourcesRates and reference materials
- ProcessingRetrieval and ranking
- OutcomeSource-linked answer
Applications in other functions
- Finance
- payment terms, limits, rates and invoicing rules
- Marketing
- product descriptions, unique selling points and industry arguments
- Customer service
- service terms, SLAs, statuses and escalation rules
- Procurement
- supplier terms, rates, timelines and procurement categories
Applications in other industries
- Retail and e-commerce
- products, shipments, returns, SLAs and delivery terms
- Financial services
- products, rates, procedures and service rules
- Manufacturing
- specifications, order statuses, production processes and dispatch
- Healthcare
- protocols, procedures, supplies and internal reference materials
AI sales quality assurance
Automatic assessment of completed calls against a checklist and evaluation of sales quality.95%agreement with expert QA assessment
- agreement with expert QA assessment
- 95%
- of calls, replacing selective monitoring
- 100%
- automatic processing
- up to 50 calls/hour
Business challenge
Managers and QA specialists manually reviewed only some calls, leaving an incomplete picture. Manual review delayed feedback to sales representatives and complicated transfer of results to AmoCRM.
Solution
The AI system automatically processes completed calls: transcribing speech, separating speakers and evaluating conversations against a sales checklist.
- Manager / QA
- defines assessment rules, consultation criteria and required conversation elements.
- System
- checks each item, identifies violations, shows timestamped fragments and produces a final score.
Architecture and integrations
Deployed in a private cloud with dedicated compute. Audio enters object storage and a task queue; the system transcribes speech, separates speakers and evaluates the checklist. Integrations include AmoCRM, a database, task queue and assessment interface.
- SourcesCalls from AmoCRM
- ProcessingASR and checklist evaluation
- OutcomeQA assessment interface
Applications in other functions
- Marketing
- lead quality, enquiry sources and rejection reasons
- Customer service
- answer completeness, tone and SLA adherence
- Operations
- adherence to procedures, processing stages and statuses
- Procurement
- supplier negotiations: prices, timelines and terms
Applications in other industries
- Retail and e-commerce
- calls about orders, returns, delivery and upselling
- Financial services
- advice on products, risks and compliance
- HoReCa
- calls about bookings, events and service
- Real estate
- calls about leads, properties, viewings and deals
AI/ML platform for seasonal scenario planning
Comparing seasonal procurement, markup and discount scenarios with financial outcome assessment.+2-5%increase in seasonal margin
- increase in seasonal margin
- +2-5%
- excess stock
- −4%
- scenarios to compare
- up to 5
Business challenge
Seasonal planning required advance decisions on order volume, channel allocation, markups and discounts. Scenarios were compared manually in Excel, making it difficult to quickly identify the best balance of revenue, margin and stock.
Solution
The AI/ML platform models seasonal scenarios and compares their financial effects.
- User
- sets seasonal parameters: order volume, channel allocation, markup, discounts and e-commerce share.
- System
- calculates up to 5 scenarios; an LLM analyst summarises them for leadership and explains differences.
Architecture and integrations
Private cloud deployment with a local LLM and no commercial data sent externally. Excel, CSV and BI data are normalised into a weekly data mart. ML models forecast demand, revenue, stock and margins across week → product group → SKU group → channel → region.
- SourcesExcel, CSV and BI
- ProcessingWeekly data mart and ML forecast
- OutcomeSeasonal scenarios
Applications in other functions
- Finance
- modelling revenue, margin, costs and cash flow
- Sales
- sales forecasts, channel allocation and average order value
- Marketing
- effects of promotions, discounts and channels on demand and results
- Procurement
- purchase volumes, stock, shortages, surplus and budget
Applications in other industries
- FMCG
- sales planning by SKU, region, promotion and margin
- HoReCa
- demand, procurement, write-offs, utilisation and margins
- Manufacturing
- output, raw materials, batches, utilisation and unit costs
- Logistics and transport
- orders, routes, warehouse workload and SLAs
Checkpoint vehicle monitoring system
Checkpoint vehicle and number plate recognition, with confirmed events sent to ERP.−60%manual checkpoint monitoring costs
- manual checkpoint monitoring costs
- −60%
- risk of unauthorised vehicle entry
- −70%
- of vehicles and specialised machinery
- 20 classes
Business challenge
Checkpoint traffic was monitored manually: staff compared cameras, logs and number plates. This caused delays, risks of unauthorised entry and limited visibility into vehicle movements between sites.
Solution
The video analytics system automatically monitors vehicle entry and exit at checkpoints.
- Cameras
- capture vehicle traffic at entry and exit points.
- System
- recognises number plates, classifies vehicles and links events to checkpoints; violations trigger alerts.
Architecture and integrations
A local GPU server, checkpoint IP cameras, database, web dashboard and ERP API integration. The computer vision model detects vehicles, reads plates, assigns one of 20 classes and tracks movement. Confirmed events go to ERP through an API and export to Excel/JSON; disputed passages are reviewed using frames.
- SourcesCheckpoint IP cameras
- ProcessingClassification and ANPR
- OutcomeEvents in ERP
Applications in other functions
Applications in other industries
- Retail and e-commerce
- supplier entry, dispatch, warehouse gates and couriers
- Financial services
- office and data centre access, cash collection
- FMCG
- deliveries, dispatch and movement of raw materials and products
- Logistics and transport
- entry/exit, queues, gates and carriers
AI agent for order status enquiries
Answers about order readiness, dispatch and documents based on company data.up to 70%of enquiries processed automatically
- of enquiries processed automatically
- up to 70%
- average answer preparation time
- < 2 sec
- satisfaction ratings
- 90%+
Business challenge
Customers repeatedly asked where an order was, its stage, when it would be ready or dispatched and which documents were available. Sales representatives checked systems manually and relayed questions between departments, making answers slow and employee-dependent.
Solution
The AI agent answers natural-language order status questions using the knowledge base, FAQs, process descriptions and order data.
- Customer or sales representative
- asks about the stage, readiness, dispatch timing or documents.
- AI agent
- finds information, prepares an answer and delivers it through a convenient communication channel.
Architecture and integrations
Cloud deployment with AI in a protected environment. The RAG assistant connects to the knowledge base, FAQs, process descriptions and order data. Requests pass through vector retrieval and ranking before the LLM produces a structured answer with a source reference.
- SourcesFAQs and order data
- ProcessingVector search and LLM
- OutcomeStatus answer
Applications in other functions
- Sales
- order statuses, readiness dates and specifications
- Finance
- invoice, payment, acceptance certificate and closing document statuses
- Logistics
- dispatch, route, warehouse and delivery statuses
- Customer service
- requests, process stages, approvals and owners
Applications in other industries
- Retail and e-commerce
- order, delivery, return and availability statuses
- Logistics and transport
- transport, route, delivery and document statuses
- Financial services
- application, product, document and stage statuses
- Education
- application, programme, schedule and enquiry statuses
AI agent for initial interviews
Initial web interviews with criteria checks and structured candidate profiles.up to 80%of applications handled without a recruiter
- of applications handled without a recruiter
- up to 80%
- qualification immediately after application
- < 5 minutes
- agreement with recruiter assessment
- 85-90%
Business challenge
The HR team spent time processing initial applications: making contact, asking basic questions, checking criteria and screening out unsuitable candidates. High volumes delayed processing for some candidates.
Solution
The AI agent conducts an initial web interview using a predefined scenario.
- Candidate
- completes a short browser conversation about experience, skills, work format and expectations.
- AI agent
- analyses answers, checks vacancy fit and assigns a suitable / unsuitable / clarify status.
Architecture and integrations
Private cloud deployment with a local language model in an isolated environment; candidate data is not sent to external services. The interview service manages the scenario, stores answers and creates a structured profile. Integrated with HuntFlow.
- SourcesCandidate answers
- ProcessingInterview scenario and LLM
- OutcomeProfile in HuntFlow
Applications in other functions
- Sales
- lead prequalification: needs, budget and timeline
- Marketing
- qualification of webinar participants and lead forms
- Customer service
- enquiry classification and routing
- Operations
- request capture, parameter checks and owner handover
Applications in other industries
- Retail and e-commerce
- processing enquiries, returns and consultation requests
- Financial services
- prequalification for products, loans and insurance
- HoReCa
- booking, event and banquet requests
- Education
- student qualification: goals, level and format
AI guest microsegmentation
Combining bookings and receipts into guest histories for behavioural segmentation and CRM marketing.+11%increase in repeat visits
- increase in repeat visits
- +11%
- increase in CRM campaign revenue
- +9.6%
- increase in average spend
- +4.7%
Business challenge
Guest data was split between two systems: bookings showed who visited and when, while POS showed purchases and amounts. The team lacked a complete guest profile and could not move to microsegmentation.
Solution
The AI system links booking and POS data, builds each guest's visit history and identifies behavioural segments.
- User
- selects a period, restaurant, channel, segment or guest type.
- System
- matches visits to receipts, analyses average spend, frequency and preferences, and generates CRM recommendations.
Architecture and integrations
Cloud deployment connected securely to booking and POS systems. The system links guest, visit and receipt through phone, email, ID and table. ML analytics calculate frequency, average spend, recency and promotion response; an LLM layer turns results into recommendations.
- SourcesBookings and POS
- ProcessingData linking and ML segmentation
- OutcomeCRM recommendations
Applications in other functions
- Finance
- revenue, average spend, margins and segment contributions
- Sales
- segmentation by purchase history, frequency and spend
- Marketing
- microsegmentation, personalised campaigns and reactivation
- Operations
- customer behaviour, peak periods and workload
Applications in other industries
- Retail and e-commerce
- linking customers and purchases, segmentation by frequency and spend
- Financial services
- segmentation by product, profitability and response
- FMCG
- customers, SKUs, outlets and repeat purchases
- Education
- student segmentation by programme and activity
AI text and voice communication system
Text and voice agents in one environment with communication quality control.up to 90%of enquiries handled without reception staff
- of enquiries handled without reception staff
- up to 90%
- availability regardless of staff shifts
- 24/7
- communication operating costs
- −46%
Business challenge
Guest communications were spread across SMS, WhatsApp, telephony and manual handling. Some messages were delayed, context was lost between channels and quality depended on the employee.
Solution
The AI system combines a text agent, voice agent and communication quality control module.
- Text agent
- answers routine questions, sends confirmations and reminders, and helps reschedule bookings.
- Voice agent
- takes calls, identifies the purpose, checks availability and creates or modifies bookings.
Architecture and integrations
Cloud deployment with a protected environment for conversations, audio and personal data. Access is role-based, sensitive fields are masked and only anonymised fragments go to external services. Messages and calls enter one environment, are classified and routed to the appropriate agent.
- SourcesMessages and calls
- ProcessingClassification and AI agents
- OutcomeAnswers and quality control
Applications in other functions
- Sales
- lead handling, qualification and consultation booking
- Marketing
- campaigns, reminders and reactivation
- Customer service
- routine answers, complaints and escalation
- Operations
- enquiry routing and reduced first-line workload
Applications in other industries
- Retail and e-commerce
- messages and calls about orders, deliveries and returns
- Financial services
- applications, consultations and document statuses
- Real estate
- lead handling, viewing bookings and consultations
- Healthcare
- appointments, reminders and routine answers
AI post-visit feedback collection
Automatic post-visit surveys, response analysis and service issue detection.+43%more feedback through automated surveys
- more feedback through automated surveys
- +43%
- of negative feedback monitored
- up to 100%
- of visits with channel attribution
- up to 88%
Business challenge
Feedback was fragmented: some guests left none, some posted public criticism and others used private channels. The team lacked a unified view of visit quality, problem causes and channel effectiveness.
Solution
The AI system automatically collects post-visit feedback and classifies responses by service quality, dissatisfaction causes and urgency.
- Guest
- receives a message through an agreed channel after visiting and leaves a rating or comment.
- System
- identifies sentiment, problem type and links to the visit and source; negative feedback goes to responsible staff.
Architecture and integrations
Cloud deployment using APIs from communication services, booking systems, POS and an AI analysis module. After a visit, a survey agent identifies the guest, chooses a channel and starts a short survey. Answers are classified by sentiment, topic, urgency and problem type.
- SourcesVisit data and communication channel
- ProcessingSurvey and response classification
- OutcomeFeedback by topic
Applications in other functions
- Finance
- customer experience effects on repeat purchases and returns
- Sales
- feedback after a deal, visit or demo
- Marketing
- acquisition sources, NPS/CSAT and campaign effectiveness
- Customer service
- dissatisfaction, complaints and recurring problems
Applications in other industries
- Retail and e-commerce
- reviews after purchases, deliveries, returns and enquiries
- Financial services
- experience after consultations, applications and support
- Logistics
- delivery, timing, cargo condition and carrier performance
- Healthcare
- appointments, booking, consultations, waiting and service
AI management reporting agent for P&L / food cost
Management P&L compilation from multiple sources and variance analysis linked to source data.1–2 days → 15–30 minP&L preparation for the chain
- P&L preparation for the chain
- 1–2 days → 15–30 min
- less manual reconciliation
- up to 80%
Business challenge
Management reports were manually assembled from POS sales, procurement and warehouse costs, and payroll, rent and Excel expenses. Restaurant P&L preparation took time, and reconciliation required manual category checks.
Solution
The AI agent combines data from multiple sources, prepares restaurant management P&Ls and helps analyse variances.
- User
- asks why food cost rose, where margins fell or which restaurant caused a variance.
- Assistant
- consolidates categories, compares budget and actuals, and explains variances with data references.
Architecture and integrations
Private cloud deployment with local AI models and no financial data sent externally. Data is brought into one P&L structure, cost categories are mapped and metrics calculated automatically. Calculations include source data references.
- SourcesSales, procurement and expenses
- ProcessingP&L mapping and calculation
- OutcomeReporting and budget variance analysis
Applications in other functions
- Sales
- deal and renewal receipts, receivables
- Procurement
- procurement budget, payment timing and shortages
- Operations
- operating expenses, resource utilisation and contractors
- Logistics
- delivery, storage and route costs
Applications in other industries
- Retail and e-commerce
- revenue, returns, costs, cash flow and margins
- FMCG
- sales, promotion impact, receivables and stock
- Manufacturing
- cost of sales, raw materials, output and working capital
- Logistics
- orders, transport costs and route profitability
AI student feedback and reporting system
Reports prepared from notes and observations with mandatory teacher review.up to 90%less reporting time
- less reporting time
- up to 90%
- entering observations by text or voice
- up to 3 minutes
- finalisation instead of manual assembly
- up to 10 minutes
Business challenge
Teachers recorded observations in scattered notes, spoken comments and handwriting. Report preparation took time, quality varied by teacher and historical context was difficult to use systematically.
Solution
The AI system collects student observations and prepares structured reports for human review.
- Teacher
- adds observations as text, voice or uploaded notes after a lesson.
- System
- recognises speech and handwriting, gathers context and drafts a report; the teacher approves it before sending.
Architecture and integrations
An independent web application with its own database, roles and protected student data storage. The AI pipeline includes speech recognition, note processing, fact extraction, student history search and draft generation. The teacher remains in the loop and approves reports before sending.
- SourcesTeacher notes and voice
- ProcessingFact extraction and AI draft
- OutcomeReview and approval
Applications in other functions
- Sales
- recording meeting outcomes, statuses and communication summaries
- Customer service
- enquiry history, issue causes and experience reports
- Operations
- observations on processes, statuses and incidents
- HR
- interviews, onboarding, performance reviews and 1:1s
Applications in other industries
- Retail and e-commerce
- reports on enquiries, store visits and service quality
- FMCG
- reports on field team visits, outlets and promotions
- Manufacturing
- shift, quality, defect and supervisor observation reports
- Real estate
- reports on sites, contractors and project stages
ML forecasting of procurement, stock and revenue
Forecasting demand, procurement and revenue using sales history, stock and external factors.Demand forecastingprocurement, stock and revenue by restaurant
- procurement, stock and revenue by restaurant
- Demand forecasting
- identifying shortages and surplus
- Inventory risks
- planning in spreadsheets and dashboards
- Scenarios
Business challenge
Procurement, stock and revenue planning relied on managers' experience and manual calculations. Changes in demand, seasonality or weather outpaced recalculation, causing surplus and write-offs for some items and shortages for others.
Solution
The ML system forecasts demand, procurement, stock and revenue using historical and operational data.
- User
- selects a restaurant, category, product group, period or scenario.
- System
- analyses sales, receipts, procurement, stock and external factors, then forecasts demand and shortage risk.
Architecture and integrations
Cloud deployment connected to POS, ERP, inventory and BI. Data is cleaned, normalised, aggregated by day/week and converted into features. Time-series and gradient boosting models forecast across category → product → restaurant, accounting for demand history and external factors.
- SourcesPOS, ERP and inventory
- ProcessingTime series and boosting
- OutcomeForecasts by product and restaurant
Applications in other functions
- Finance
- revenue, costs, cash flow and budget variances
- Sales
- demand, sales volume, conversion and channel utilisation
- Procurement
- requirements, purchase volumes, stock, shortages and budget
- Logistics
- dispatch, routes, warehouse workload and peak periods
Applications in other industries
- Retail and e-commerce
- demand, sales, stock, returns and promotion impact
- FMCG
- sales by SKU, region, distributor and promotion
- Manufacturing
- demand, output, raw materials, line utilisation and defects
- Logistics and transport
- orders, routes, warehouse workload and SLAs
ML forecasting of costs and receipts
Forecasting receipts, expenses and cash shortfalls using financial and operational data.+10%more accurate annual budgeting
- more accurate annual budgeting
- +10%
- unplanned expenses
- −15%
- cash flow forecast horizon
- up to 12 months
Business challenge
Financial planning relied on manual calculations and expert estimates. Tuition and donor receipts arrived unevenly, while expenses required careful cash flow planning. Forecasting annual budget execution was difficult.
Solution
The ML system forecasts receipts, expenses and cash flows using historical financial and operational data.
- User
- selects a period, scenario, budget category, receipt source or cost category.
- System
- analyses payments, grants, contributions, payroll and seasonality, showing cash flow forecasts, shortfall risks and variances.
Architecture and integrations
Cloud deployment connected to financial and operational sources. Payment exports, CRM/fundraising, budget and payroll data are cleaned and grouped by period and source. Time-series and gradient boosting models forecast receipts, expenses and cash shortfalls.
- SourcesPayments, CRM and budgets
- ProcessingTime series and boosting
- OutcomeCash flow forecast
Applications in other functions
- Sales
- deal and renewal receipts, receivables
- Marketing
- campaign, channel and seasonal effects on receipts
- Operations
- operating expenses, resource utilisation and contractors
- Procurement
- procurement budget, payment timing and shortages
Applications in other industries
- Retail and e-commerce
- revenue, returns, costs, cash flow and stock
- FMCG
- sales, promotion impact, receivables and stock
- Manufacturing
- cost of sales, raw materials, output and working capital
- Logistics and transport
- orders, transport costs and cash flow
AI distributor discovery and contact extraction
Discovering and ranking B2B distributors by profile, product range and geography.3 days → 15 minutesinitial shortlist preparation
- initial shortlist preparation
- 3 days → 15 minutes
- manual search costs
- −90%
- irrelevant companies filtered out
- up to 70%
Business challenge
The engineering equipment distributor market was fragmented across countries and sources: directories, associations, local portals and PDFs. The team manually searched for partners and collected contacts, producing duplicates and irrelevant company records.
Solution
The AI system automates discovery and ranking of potential B2B distributors.
- User
- selects a country, region, industry, equipment type and relevance criteria.
- System
- creates localised queries, analyses websites and directories, assesses fit and collects contacts.
Architecture and integrations
An automated discovery, scoring and enrichment environment. The system generates localised queries, collects results and analyses websites, directories and PDFs. An LLM classifier assesses relevance by profile, products and geography. It extracts emails, phone numbers, contact forms and public profiles.
- SourcesWebsites, directories and PDFs
- ProcessingCollection and LLM classification
- OutcomePartner and contact list
Applications in other functions
- Sales
- finding leads, partners, distributors and dealers
- Marketing
- company databases, platforms, competitors and market intelligence
- Procurement
- suppliers, alternative manufacturers and importers
- Operations
- monitoring contractors and service organisations
Applications in other industries
- Retail and e-commerce
- suppliers, brands, marketplace sellers and distributors
- Financial services
- companies, partners, brokers and public registers
- FMCG
- distributors, sales representatives and chains
- Logistics and transport
- carriers, freight forwarders and warehouse operators
AI competitive intelligence platform
Regular competitor monitoring and comparison of website, pricing and communication changes.15 hours → < 1 hourmanual analysis per month
- manual analysis per month
- 15 hours → < 1 hour
- parameter extraction accuracy
- 85%+
- change table
- comparative
Business challenge
Competitor analysis was manual and irregular: the team separately checked competitors' websites, prices, advertising, social media and SEO. It took 10–15 hours per month, quickly became outdated and produced a fragmented picture.
Solution
The AI platform automatically monitors competitors and market changes using a defined company list and parameters.
- User
- specifies competitors, pages, products, markets and monitoring parameters.
- System
- scans websites, extracts offers, prices, positioning, CTAs and SEO, and produces a change report.
Architecture and integrations
An AI competitor monitoring platform with models accessed through APIs rather than local deployment. Sources include search results, websites, product pages, aggregators, social media and advertising. The system removes duplicates, scans pages and stores versions for comparison.
- SourcesWebsites and public sources
- ProcessingScanning and versioning
- OutcomeChange comparison
Applications in other functions
- Sales
- competitors, offers, prices, USPs and objection-handling arguments
- Marketing
- positioning, messaging, CTAs, SEO and go-to-market
- Finance
- prices, rates, discounts and market conditions
- Customer service
- reviews, public complaints and customer expectations
Applications in other industries
- Retail and e-commerce
- prices, range, promotions, product listings and advertising
- Financial services
- rates, products, terms and digital funnels
- FMCG
- SKUs, packaging, promotions, trading terms and channels
- Real estate
- properties, prices, promotions, terms and positioning
AI logistics and supplier management platform
Managing requests from briefs and supplier discovery to commercial proposal comparison.−60%manual operational workload
- manual operational workload
- −60%
- logistics and procurement costs
- −3-7%
- time to supplier selection
- −30-50%
Business challenge
Logistics requests were fragmented across messaging, email, files and spreadsheets. Managers manually gathered requirements, found suppliers, requested prices, compared terms and updated statuses.
Solution
The AI platform manages requests from the customer's initial enquiry to supplier proposal comparison.
- Customer
- provides requirements: product, quantity, packaging, country, timeline, Incoterms and budget.
- AI agent
- structures the brief, finds suppliers on Alibaba and B2B directories, requests prices, MOQs and lead times, and compiles proposals in a table.
Architecture and integrations
A cloud environment with a server application, database, proposal object storage, task queue and access management. AI agents gather briefs, find suppliers, communicate, extract proposal terms and prepare comparisons. Discovery uses Alibaba and B2B directories with listing extraction.
- SourcesBrief and B2B directories
- ProcessingAI agents and term extraction
- OutcomeProposal comparison table
Applications in other functions
- Sales
- structuring requests and monitoring statuses and commitments
- Procurement
- supplier discovery, tender tables, MOQs and prices
- Operations
- tasks, deadlines, statuses, owners and timelines
- Finance
- costs, payment documents and payment terms
Applications in other industries
- Retail and e-commerce
- suppliers, orders, deliveries, documents and statuses
- FMCG
- coordinating suppliers, orders, batches and logistics
- Manufacturing
- raw material and component supply, deadline monitoring
- Healthcare
- equipment and pharmaceutical supplies, certificates
eSIM provider price monitoring platform
Daily collection and comparison of eSIM provider rates by country, data allowance and duration.−90%price monitoring costs
- price monitoring costs
- −90%
- increase in margin
- +2-5%
- updated rate table
- daily
Business challenge
The team needed regular competitor eSIM price monitoring across countries, data allowances, package durations and currencies. Manual monitoring quickly became outdated, required manual currency conversion and made comparisons time-consuming.
Solution
The platform collects prices daily from a defined list of eSIM providers by country, data allowance and package duration.
- System
- visits competitor websites, extracts rates and terms, normalises prices to USD and stores history.
- Analytics layer
- compares rates and identifies price gaps, market deviations and margin risks.
Architecture and integrations
A cloud environment with scheduling, scrapers, a processing queue, server application, PostgreSQL and a dashboard. Scrapers visit provider sites daily. Data is deduplicated, standardised and converted to USD; price per GB and market deviation are calculated. External AI models accessed through APIs extract data from complex pages.
- SourceseSIM provider websites
- ProcessingExtraction and USD normalisation
- OutcomeRate dashboard
Applications in other functions
- Sales
- competitor offer comparison and term adjustments
- Marketing
- promotions, pricing offers, positioning and campaigns
- Finance
- prices, margin, currency variances and price gaps
- Procurement
- market terms, suppliers and rate structures
Applications in other industries
- Retail and e-commerce
- competitor prices, product listings, discounts and promotions
- Financial services
- rates, fees, interest rates and service packages
- FMCG
- prices by SKU, region, retail chain and promotion
- Logistics and transport
- carrier rates, warehouse services and surcharges
AI sponsor presentation generation
Preparing sponsor reports from event data, photographs and media placement results.−70-85%report preparation time
- report preparation time
- −70-85%
- presentation preparation costs
- −40-60%
- final .pptx
- editable
Business challenge
Sponsor presentations were assembled manually from customer databases, spreadsheets, contractor reports, media placements and photos. Managers checked metrics, selected photos, prepared summaries and verified logos.
Solution
The AI system creates a final sponsor presentation using event data, media effectiveness and visual materials.
- Sales representative
- selects a sponsor and event, checks report structure and makes edits.
- System
- collects data, creates slides, selects photos, checks logos and prepares a template-based presentation.
Architecture and integrations
A web application with a backend, database, photo/document storage, preview and .pptx generator. A language model writes slide text; a multimodal model selects photos and recognises logos. Data fills an approved template, producing an editable .pptx.
- SourcesData, documents and photos
- ProcessingText, photo selection and template
- OutcomeEditable presentation
Applications in other functions
- Finance
- management reports and board presentations
- Sales
- proposals, client reports and account presentations
- Marketing
- campaign, reach and channel effectiveness reports
- Operations
- operational reports on statuses, KPIs and incidents
Applications in other industries
- Retail and e-commerce
- reports for brands, suppliers and marketplaces
- Financial services
- reports for clients, investors and committees
- FMCG
- presentations on promotions, outlets and brands
- Real estate
- presentations on properties, project stages and investors
AI agent for marketing material creation
Creating marketing materials from a knowledge base, adapted to objectives, audiences and formats.−70-85%report preparation time
- report preparation time
- −70-85%
- presentation preparation costs
- −40-60%
- style and brand alignment
- consistent
Business challenge
Marketing materials were manually prepared from PDFs, DOCX, PPTX, websites and competitor analysis. Teams repeatedly rebuilt structures, formulated arguments and checked brand alignment, leaving presentations, emails and landing pages inconsistent in quality and style.
Solution
The AI agent creates marketing materials for a specified objective, audience and format.
- User
- uploads sources and chooses material type, audience, style and communication objective.
- AI agent
- extracts key ideas, builds a structure, considers competitors and turns text into finished template-based material.
Architecture and integrations
A web interface, backend, document storage, knowledge base, AI agent and file generator. The system extracts text, splits it into semantic blocks and retrieves relevant fragments. The agent structures and writes content, adapts tone and checks brand alignment. Content exports in the required format.
- SourcesDocuments and knowledge base
- ProcessingContext retrieval and generation
- OutcomeMaterial in the required format
Applications in other functions
- Sales
- proposals, client emails, solution presentations and follow-ups
- Marketing
- campaigns, landing pages, presentations and one-pagers
- Customer service
- instructions, customer answers and training materials
- Operations
- procedures, presentations and process descriptions
Applications in other industries
- Retail and e-commerce
- product descriptions, promotions, emails and landing pages
- Financial services
- product presentations, investment materials and emails
- FMCG
- brand, promotion and sales channel materials
- Education
- programme presentations, emails and course landing pages
AI bankruptcy petition preparation
Document recognition and template-based petition preparation, with disputed fields reviewed by a lawyer.−70-90%petition preparation time
- petition preparation time
- −70-90%
- manual data transfer errors
- −50-70%
- legal review
- human-in-the-loop
Business challenge
Preparing an individual's bankruptcy petition required manually processing a large package of scans, PDFs and photographs. Lawyers checked quality, located information and copied it into a template, consuming time and creating error risks.
Solution
The system automates bankruptcy petition preparation using the client's document package.
- Legal department
- uploads documents through a web interface and starts processing.
- System
- checks file quality, recognises data with OCR, extracts fields and fills the petition template.
Architecture and integrations
A protected private-cloud web platform with a server service, database, file storage and task queue. OCR recognises text and extracts full names, passport details, tax and pension insurance identifiers, debt amounts, creditors and other details. Data is mapped to template fields; uncertain fields are highlighted for a lawyer.
- SourcesScans, PDFs and photographs
- ProcessingOCR and field mapping
- OutcomeDraft for review
Applications in other functions
- Finance
- recognising invoices, acceptance certificates, statements and contracts
- Sales
- processing requests, forms and commercial documents
- Customer service
- document intake and package completeness checks
- Procurement
- recognising proposals, invoices, contracts and specifications
Applications in other industries
- Retail and e-commerce
- invoices, acceptance certificates, return documents and claims
- Financial services
- forms, applications, statements and KYC packages
- Logistics and transport
- invoices, bills of lading, delivery notes and certificates
- Healthcare
- medical documents, consents and prescriptions
Visual clothing search and recommendations
Finding similar clothing by photo and visual features in a product catalogue.+4%increase in catalogue revenue
- increase in catalogue revenue
- +4%
- unsuccessful search sessions
- −19%
- increase in product clicks
- +12%
Business challenge
Users struggled to find clothing through categories and text search because descriptions missed cut, silhouette, fit and style. Ready-made visual search models compared backgrounds and model poses rather than clothing, producing irrelevant recommendations.
Solution
The system shows similar products as soon as a product page or photograph is opened.
- User
- opens a clothing image and sees similar items by cut, colour, silhouette and style.
- System
- pre-analyses catalogue images, converts them to vectors and finds visually similar items.
Architecture and integrations
A server service, image storage, product database, vector index and catalogue API. Computer vision extracts clothing features: cut, colour, silhouette, fit, details and style. Images become vectors; new products are vectorised daily without rebuilding the entire index.
- SourcesProduct images
- ProcessingComputer vision and vector index
- OutcomeSimilar products
Applications in other functions
Applications in other industries
- FMCG
- comparing packaging, SKUs, promotional materials and displays
- HoReCa
- similar dishes and analysis of menu, plating and interior photos
- Manufacturing
- similar parts, samples, defects and reference checks
- Real estate
- similar properties, interiors, layouts and finishes
Computer vision packaging efficiency monitoring
Packed box counts by workstation, with disputed events verified through video.−5-12%variable pay costs
- variable pay costs
- −5-12%
- downtime losses
- −10-13%
- output tracking accuracy
- up to 98%
Business challenge
The packing workshop needed accurate box counts for each employee. Manual counting was inconvenient, caused pay disputes and lacked objectivity. Different camera angles, no markers and overlapping zones made tracking harder.
Solution
The video analytics system automatically counts packed boxes and records results for each workstation.
- Cameras
- film work areas on the packing line.
- System
- detects boxes, packing actions and workstations; managers see output by shift and hour.
Architecture and integrations
A local GPU server, IP cameras, database, video processing module and dashboard. The system handles multiple camera streams, divides frames into zones and links events to locations. The computer vision model detects boxes and actions and is fine-tuned on workshop data. Events count only when conditions are met; disputed cases are checked against video clips.
- SourcesIP camera video streams
- ProcessingComputer vision and workstation attribution
- OutcomeOutput tracking
Applications in other functions
Applications in other industries
- Retail and e-commerce
- order picking, packing, labelling and sorting
- FMCG
- filling, packing, labelling, defects and output
- HoReCa
- order assembly, delivery packaging and kitchen pace
- Logistics and transport
- sorting, picking, packing and loading
Specialised vehicle monitoring on site
Monitoring specialised vehicle movement on site against route plans.−30-50%incident search time
- incident search time
- −30-50%
- route delays
- −20-40%
- losses from schedule disruption
- −10-25%
Business challenge
The site needed real-time monitoring of specialised vehicles: location, checkpoints and route compliance. Manual monitoring lacked visibility and recorded violations late.
Solution
The computer vision system monitors specialised vehicle movement around the site and compares it with the route plan.
- Vehicles
- receive ARUCO markers for reliable identification of individual machines in video streams.
- System
- recognises vehicles, records checkpoints and compares them with planned routes and permitted travel times.
Architecture and integrations
Client-hosted servers, IP cameras, video analytics, an event database, web interface and TRASSIR integration. ARUCO markers provide identification, with additional detection when a marker is partially lost. Events are compared with route plans and permitted delays; clips are retrieved through TRASSIR.
- SourcesCameras and ARUCO markers
- ProcessingDetection and route verification
- OutcomeEvents and video in TRASSIR
Applications in other functions
Applications in other industries
- Retail and e-commerce
- warehouse vehicle and forklift movements, dispatch zones
- Financial services
- cash-in-transit vehicles and restricted area access
- FMCG
- raw material, product, forklift and vehicle movements
- Logistics and transport
- fleets, forklifts, gates, routes and parking
Industrial safety: helmet and zone monitoring
Helmet and hazardous-zone entry detection with violation notifications.up to 95%helmet monitoring accuracy
- helmet monitoring accuracy
- up to 95%
- manual monitoring costs
- −30%
- risk of delayed response
- −30-50%
Business challenge
The site needed to reduce injury risk through automatic helmet and hazardous-zone monitoring. Manual checks did not consistently capture violations; some went unnoticed and responses were delayed.
Solution
The video analytics system monitors industrial safety compliance in real time.
- Cameras
- cover production areas, walkways and hazardous locations.
- System
- detects people, checks helmet use and hazardous-zone entry, and sends violation notifications.
Architecture and integrations
A local GPU server, IP cameras, video analytics, an event database, web interface and notification system. The computer vision model detects people, recognises helmets, classifies violations and tracks movement. Rules are set per camera: helmet requirements, zone boundaries and violation types. Every violation is saved with an evidentiary frame.
- SourcesProduction area cameras
- ProcessingPPE detection and tracking
- OutcomeNotification with supporting frame
Applications in other functions
Applications in other industries
- Retail and e-commerce
- safety requirements in warehouses and loading areas
- Financial services
- restricted area, data centre and archive access
- FMCG
- PPE, safe zones, production lines and warehouses
- Logistics and transport
- warehouse areas, loading, hazardous zones and access
Student engagement monitoring
Analysing online lesson video streams and student engagement trends.student profileindividual engagement trends
- individual engagement trends
- student profile
- of distraction in real time
- signals
- for each lesson to improve delivery
- analytics
Business challenge
After some courses moved online, teachers found it harder to judge student engagement. Online classes lack familiar classroom feedback, and teachers cannot always see who is distracted or no longer following the material.
Solution
The system analyses online lesson video streams and helps teachers monitor engagement trends.
- System
- analyses student faces, gaze direction and head position to identify signs of attention.
- Teacher
- sees where group attention falls, which topics cause difficulty and who disengages.
Architecture and integrations
Integration with the online teaching platform, video stream processing, a metrics database, GPU server and teacher interface. Streams arrive through WebRTC/platform APIs and frames are processed in real time without manual annotation. Computer vision provides face detection, gaze and head position estimation, and basic emotional feature analysis.
- SourcesLesson video streams
- ProcessingComputer vision and gaze estimation
- OutcomeMetrics in the teacher interface
Applications in other functions
Applications in other industries
- Retail and e-commerce
- engagement in online consultations and video sales
- Financial services
- customer attention during consultations and onboarding
- Manufacturing
- attention during briefings and safety training
- Logistics and transport
- driver and dispatcher attention during briefings
Tyre wear assessment from photographs
Photo-based tyre wear assessment for initial warehouse sorting.−50-70%tyre assessment time
- tyre assessment time
- −50-70%
- manual inspection costs
- −20-40%
- sorting errors
- −10-25%
Business challenge
Tyre wear was assessed manually from photos and depended on visual inspection quality. Large warehouse volumes increased employee workload, subjectivity risks and sorting errors.
Solution
The system estimates tyre wear from a photograph for initial sorting.
- Employee
- uploads or takes a tyre photograph in the warehouse interface.
- System
- analyses the image, identifies wear features and estimates the percentage of remaining condition.
Architecture and integrations
An image processing and computer vision training pipeline. Photos are resized, normalised and augmented through flipping, brightness and contrast changes to reduce shooting-condition effects. EfficientNet B0 is trained for regression, producing a numerical wear estimate rather than just a category.
- SourcesTyre photographs
- ProcessingPreprocessing and EfficientNet B0
- OutcomeNumerical wear estimate
Applications in other functions
Applications in other industries
- Retail and e-commerce
- condition of products, returns, used items and defects
- FMCG
- packaging condition, damage, defects and expiry dates
- Manufacturing
- wear on parts, components, tools and surfaces
- Real estate
- wear on materials, machinery, facades and finishes
Showing 6 of 28
