Case StudiesCase
Case Study: AI for a Building Materials Wholesaler - Requests and Specifications in Bitrix24
A building materials wholesale company already running Bitrix24 as its CRM added AI on top of an established process - and shifted managers from manual data entry to actual selling. This case covers every AI scenario, what worked, and where the limits were.
Every channel ends up in one portal
Company profile and starting point
This is a B2B building materials wholesaler with a sales team handling high-volume, multi-line specification requests from site managers and procurement officers - a context where response speed and data accuracy matter more than price alone.
The company had already built out its Bitrix24 deal funnel, counterparty cards, product catalog, and invoicing before AI was introduced. The full logic of how a distributor's workflow sits inside Bitrix24 is covered in Bitrix24 for Distributors and Dealer Networks. This article picks up from where that setup ends.
The core sales cycle is stable: a procurement officer or site manager sends a request - by email, as an attached Excel bill of materials, a PDF specification, or sometimes a photo of a handwritten list. A manager breaks down the line items, prices them, and sends a commercial proposal. Once agreed, the goods ship and closing documents follow. On large construction projects, shipments repeat over months as the site consumes materials in batches.
Before AI was added, six friction points kept coming up:
| Pain point | Practical impact |
|---|---|
| Manual re-entry of line items from attachments | An hour or more per large request before any actual selling |
| Slow first response | Competitor who replied first won the order |
| Transcription errors at the re-entry stage | Wrong grade, wrong unit - disputes arose at delivery |
| Empty deal cards after calls | Project details, timelines, and agreements stayed in the manager's head |
| Reactive repeat orders | Manager waited for the client to call; no proactive outreach |
| Script oversight at ~10% of calls | Most conversations went unreviewed |
AI scenarios we introduced and how they work in Bitrix24
Six AI workflows were layered onto the existing Bitrix24 setup: three run on built-in CoPilot and three are apps built on Alaio Vibecode - together they cover the full request-to-proposal cycle.
An email with an attachment enters CRM through the connected mailbox, CoPilot processes the call recording, a Vibecode app handles document parsing and repeat-order prediction, and the manager receives a structured deal card ready to work with.
An email with an attachment and an inbound call both feed into the Bitrix24 deal card, where CoPilot and Vibecode apps process them into structured outputs for the manager.
Parsing incoming requests from emails and attachments
Requests arrive in every format imaginable: Excel bills of materials, PDF specs, scanned documents, and phone photos of handwritten lists taken on a construction site. Previously a manager would open each attachment and retype every line into the deal card or a separate spreadsheet. On a request with several dozen line items, that alone consumed an hour or more before any pricing work could begin.
A Vibecode app - built on the Alaio Vibecode platform with a vision-capable AI model - now reads the attachment. For tabular files it extracts rows directly; for images and scans, a vision model reads the text including photos taken at an angle or in poor light. The result lands in the deal card as structured rows: item name, quantity, unit of measure, and any comment. The manager opens the card to a ready list and moves straight to pricing and sourcing.
This scenario is a custom Vibecode app rather than a built-in CoPilot feature, because document and image recognition from email attachments goes beyond what CoPilot's native CRM field-fill handles (CoPilot in CRM fills fields from the email itself; extracting line items from attachments is not documented as a CoPilot feature).
If your team works with the AI Call Analysis in Bitrix24 article alongside this one, note that document parsing from attachments is a separate app from call transcription.
Autofilling deal fields from calls with CoPilot
When a procurement officer calls to discuss a project, the conversation covers the site address, delivery timeline, required grades, and agreed conditions. Without AI, a manager might add a brief comment to the CRM card - the details stayed in memory and were inaccessible to colleagues and management.
Bitrix24 CoPilot transcribes the call recording and creates a summary. It then fills empty CRM fields - project name, delivery deadline, source, conditions - directly from the transcript. If a field already has a value, CoPilot suggests an update rather than overwriting it, so no existing data is lost. The result appears in the deal timeline marked "Processed by CoPilot," with the full transcript and a short summary available in the same view.
What managers in this case observed is that the card was filled by the time the call ended, rather than remaining blank until the manager had a free moment.
CoPilot works with calls from 10 seconds to 1 hour and transcribes in the language of the conversation (official help, 25 Aug 2026).
Script-compliance scoring on every call
A sales script existed before AI was introduced. The practical problem: a single manager realistically listens to about 10% of calls manually. The other 90% went unreviewed.
CoPilot's script-analysis feature scores every recorded call against the loaded script and places the result in the deal timeline. Scores appear as numeric values - a call might score 83%, 42%, or 100% on script compliance. The manager sees the score next to the recording and transcript without listening to the call first.
This shifts the team leader's workflow from random spot-checks to exception handling: low scores flag the specific calls worth reviewing, and feedback becomes concrete ("you skipped the delivery timeline discussion at minute 4") rather than general coaching. All calls are visible, not just a sample.
The ready-made sales scripts in Bitrix24 CRM (new customer, complaint, product presentation) were used as the starting template, then adjusted to match the wholesale context (official help on script analysis, 24 Apr 2026).
Predicting repeat orders with a Vibecode app
Construction projects consume materials in batches over months. A client who bought roofing materials in April will need fasteners and underlays in June and finishing materials in August. Without a system prompt, managers waited for the client to call.
A Vibecode app analyzes each client's order history - typical intervals between shipments and volumes of previous orders - and generates a prompt for the assigned manager when the next batch is likely due. The manager reaches out proactively rather than reactively.
This is built as a custom app on Alaio Vibecode (similar in logic to the Lead Scoring and AI Lead Classifier scenarios in the Vibecode catalog) because it requires reading historical order data from the CRM and applying interval logic. The prompt appears inside Bitrix24 so the manager acts on it without leaving the system.
The prediction is a suggestion, not a certainty. If the client's project scope changed or the order history in CRM is incomplete, the timing will be approximate. The manager confirms actual need in the conversation. For more on lead and deal scoring mechanics, see Lead Management in Bitrix24: Capture, Distribution and Scoring.
Drafting replies and commercial proposals with CoPilot
Once the spec line items are extracted and the manager has priced them, CoPilot drafts a covering email or the text structure of a commercial proposal. The manager reviews, adjusts the language for the specific client relationship, and sends.
Prices, availability, and substitution decisions always go through the manager. CoPilot's draft handles the formatting and standard language; the specialist handles the commercial judgment. This fits how CoPilot's text-generation features work in the Feed and CRM context (Bitrix24 pricing page).
For teams that generate formal quotes and contracts inside Bitrix24, the Auto-Generating Quotes, Invoices and Contracts in Bitrix24 article covers the document template layer.
Manager digest built on Alaio Vibecode
The department head wanted a daily summary inside Bitrix24: how many new requests came in, average first-response time, and which clients had no repeat shipment for an unusual period. Rather than pulling these figures manually from separate reports, a small app was built on Alaio Vibecode.
The department head described what he wanted in plain language. A partner used Vibecode to assemble the app without bringing in a separate development team. The summary appears inside the Bitrix24 portal each morning. This matches the "Daily CRM Digest" scenario type in the Vibecode solutions catalog (vibecode.bitrix24.com/solutions).
Essentials plans have AI usage limits; unlimited AI starts from Standard Vibe+ upward (bitrix24.com/prices, 5 Oct 2026). The attachment-parsing, repeat-order and digest apps run on Alaio Vibecode, which requires a Vibe+ plan; their server time is charged in Vibe credits.
What changed in daily work
The most visible shift was that managers began the pricing and sourcing conversation - the work they are paid for - from the first minute of handling a request, rather than spending that time on data entry.
Specific qualitative changes observed:
- Faster first response. With line items already in the deal card, the manager moves to pricing and sourcing immediately. The client receives a response sooner, which in wholesale often decides who gets the order.
- Fewer transcription errors at delivery. When a model extracts the line items rather than a tired manager retyping them at the end of the day, missed rows and unit-of-measure mistakes become less frequent. Disputes at the delivery stage linked to specification mismatches decreased.
- Deal cards stopped being empty. Project name, delivery timeline, conditions, and source are captured from the call automatically. Colleagues and the team leader can see them without calling the manager.
- Repeat shipments became proactive. Managers reach out with a system prompt rather than waiting for the client to remember the next batch.
- Full call visibility for the team leader. Script reviews became targeted: a low score identifies the exact call and the exact point where the script was not followed. Coaching conversations became specific rather than general.
This case is a composite account. No revenue percentages or cost-saving figures are stated.
Limits: where AI did not help
AI in this workflow is a preparation and flagging tool, not a decision-maker - every scenario has a boundary where the manager's judgment is required and cannot be removed.
| Scenario | What AI does | What AI cannot do |
|---|---|---|
| Spec parsing | Extracts line items from attachments | Confirm current stock or apply client-specific pricing |
| Call autofill | Fills empty CRM fields from transcript | Replace the manager's review of agreed terms |
| Script scoring | Scores every call numerically | Decide whether a low score matters in context |
| Repeat-order prediction | Suggests timing based on order history | Adjust for changed project scope or partial history |
| Proposal drafting | Writes structure and standard language | Make substitution decisions or confirm availability |
Poor-quality scans produce incomplete results. A blurred photo or a document photographed at a sharp angle will yield partial extraction - some rows recognized, some not. Human review of extracted line items is mandatory and should not be removed from the process. This is the correct workflow, not a workaround.
Analog substitution is always a specialist decision. The wrong substitute material on a construction project costs more than the time spent checking. AI can surface options by specification parameters; the engineer or experienced sales manager confirms or rejects.
Repeat-order prediction needs a complete history. If the order history in CRM is sparse or fragmented, the model has little to work with and the suggestions will be rough approximations.
Data routing through CoPilot. CoPilot uses third-party AI providers including OpenAI; Bitrix24 stores data for up to 14 days for technical purposes and then deletes it (official FAQ, 7 May 2026). Providers may use content to improve their models under their own terms. For teams with strict data-handling requirements, review the Where Bitrix24 AI Processes Your Data article before enabling CoPilot on sensitive call content.
How a similar company can start
A practical checklist for a wholesale or distribution company that has Bitrix24 configured and wants to add AI on top of its existing process.
- Run a real-data test first. Take 20 requests from last month and run them through the document-extraction tool before any wider rollout. This shows concretely how the model handles your specific attachment formats - Excel, PDF, scan, photo.
- Define the deal card fields before enabling autofill. CoPilot fills empty fields from calls, but the fields must exist and be named. Agree upfront: which fields are mandatory, what they are called, and what valid values look like - at minimum project name, delivery timeline, conditions, and lead source.
- Load the sales script into CoPilot and observe for a week without drawing conclusions. Get used to the scoring format first. After a week, check whether the script reflects real conversations - adjust it if it does not, then use scores for coaching.
- Build order history in CRM before activating repeat-order prediction. The Vibecode app works from historical data. If that history is incomplete, migrate or reconstruct at least 12 months of shipment records first.
- Note AI plan limits. CoPilot call processing has AI usage limits on Essentials plans; "Unlimited AI" starts from Standard Vibe+. The attachment-parsing app runs on Alaio Vibecode, which needs a Vibe+ plan and charges server time in Vibe credits.
- Start with one manager, then scale to the team. A pilot on one person lets you refine the line-item review process, agree on the commercial proposal format, and tune the script scoring before it goes live across the whole department.
For the broader implementation approach, the Bitrix24 Onboarding Questionnaire covers the questions a partner needs answered before configuring any of these scenarios.
ACP Group is a Bitrix24 Gold partner and can configure CoPilot call analysis, build custom Vibecode apps for document parsing and digest reporting, and run team training on AI workflows. Contact us at acp-24.com to discuss your specific setup.
Questions we get asked
FAQ: AI for Building Materials Wholesale
Can AI send the client a price quote automatically?
No. The Vibecode parsing app extracts line items from the request and CoPilot can draft the email structure, but prices, stock availability, and client-specific conditions are always confirmed by the manager. The final commercial proposal always passes through a human.
What happens if the specification arrives as a low-quality photo?
The vision model will extract some rows and miss others, especially if the image is blurred or taken at a sharp angle. Human review of extracted line items is a required step and should not be removed. This is the normal working process, not an edge case.
Will AI suggest a substitute if an item is out of stock?
CoPilot or the Vibecode app can surface options by specification parameters, but the substitution decision stays with the manager or technical specialist. An incorrect substitute in building materials carries real project risk, so the human confirms or rejects any proposed alternative.
Where does the repeat-shipment prompt come from?
A Vibecode app reads the client's order history in Bitrix24 - typical shipment intervals and previous volumes - and generates a prompt for the assigned manager when the next batch is likely due. It is a suggestion based on history, not a confirmed order; the manager verifies actual need in the conversation.
Will script scoring feel like surveillance to the sales team?
The score appears alongside the call recording and is intended primarily for coaching: identifying which stage of the conversation diverges from the script and why. Explaining the purpose to the team before launch - improvement tool, not monitoring - determines how it is received.
What Bitrix24 plan is needed to run all these scenarios at volume?
Built-in CoPilot features (call transcription, field autofill, script scoring, proposal drafts) are available on cloud and on-premise plans; Essentials plans have AI usage limits. Unlimited AI starts from Standard Vibe+. The Vibecode platform for custom apps (document parsing, repeat-order prediction, digest) requires any Vibe+ plan.
What we hear in the first ten minutes
Send one message and skip the discovery call
Write to +971 55 780 1481 on WhatsApp. Describe your setup and the headcount, and you get a written scope and price back in one working day.