Skip to content
Data, AI & AutomationConsultancy2026

An assistant that answers, listens and hands over the lead: Akdan.

An on-site assistant that knows Akdan's six practices, understands each enquiry and hands it to the team the moment the visitor chooses to leave an email.

We wanted an assistant that answers like the firm would, reads the room like a good associate, and hands the conversation over the moment it becomes a lead.

The Context

Overview

Akdan Consultancy's services are the kind people research before they call: company valuation, project finance, restructuring, asset and wealth management. A visitor arriving at eleven at night with a question about a share transfer used to find a services page and a contact form. This case study covers what answers instead.

The assistant lives in two places on akdan.com and is one conversation: a panel that opens inside the hero of the front page, and a button in the corner of every other page. Move between pages and the chat continues. Ready-made questions carry the service they belong to, so a visitor who starts with "how is a company valued" and ends with "let's talk" reaches a contact form already filled with the right topic.

What it knows is the site itself. The knowledge base is generated from the site's own texts: the six practices, the founder's profile, the way the firm works, the site-wide FAQ. Live listings are added at request time, so the assistant can speak about the assets currently for sale without anyone editing a script. It answers in Turkish or English according to the page, and it does not invent a fee, a timeline or a guarantee the site does not state.

Each answer also produces a reading for the team. The model scores the conversation from 0 to 100 against a rubric agreed with the firm: how squarely the topic sits on one of Akdan's practices, how concrete and urgent the need is, and whether a next step is being asked about. The score maps to cold, warm or hot, so the team can see in broad strokes which topics draw interest, including conversations where no contact details were shared.

When a visitor writes an email address into the chat, the platform itself opens the lead: a CRM record with the transcript, the score and the intent in one line, a notification to Akdan's team, and a receipt to the visitor. The assistant simply says the request has been passed on. No form appears; the contact card the site used to open is now only a fallback.

The assistant runs through the PumpGrowthOS AI gateway on a key of its own, separate from every other client, with the layered protection a public endpoint needs: origin and session checks, limits per visitor and per day, a daily budget that stops the assistant before it costs the firm, and a human-verification step that appears only when the network asks for it.

The Problem

The Challenge

A consultancy sells judgement, and judgement is what visitors want to test before they pick up the phone. A contact form cannot answer a question, and a generic chatbot cannot be trusted with a firm's reputation.

Service questions arrived by email and waited for office hours; the visitor with the question had often moved on by morning.

A generic chatbot would guess at fees, timelines and guarantees the firm has never stated.

Every enquiry reached the team looking the same; nobody could tell idle curiosity from a shareholder's pressing deadline.

A public AI endpoint invites abuse, and every abusive call costs money.

What We Built

One conversation, two surfaces, a second output nobody sees.

The assistant is generated from the site's own texts and live listings, scores every turn against the firm's rubric and opens the CRM lead itself when a visitor leaves an email.

Visitor asks
Answer from the site's own knowledge
Score and stage on every turn
Email appears in the chat
Lead opened in the CRM
Team notified, visitor thanked
Automations
  • Knowledge base regenerated from the site's texts
  • Live listings added to the prompt on every request
  • Lead score and session record updated on every turn
  • Lead, activity and notification created when an email appears
  • Receipt to the visitor, alert to the team
  • Daily call and token circuit breaker
The Impact

Answers at night, leads by morning.

0 to 100
Every conversation scored against the firm's own rubric: fit, concreteness, urgency, authority, intent.
2 surfaces
Front-page panel and corner button, one continuing conversation across the site.
1 email
A valid address in the chat is enough: the platform opens the lead, notifies the team and thanks the visitor.
6 practices
Knowledge generated from the site's own texts, with live listings added on every request.
The Outcome

The bottom line.

Visitors now get an answer in the firm's own words at any hour, in either language, about the practice they are actually asking about. The conversation follows them across the site and ends, when it should, in a lead.

Akdan's team receives enquiries with the reading already attached: score, stage, the service in question, a one-line summary of the intent and the transcript. And because interest is read across the whole site, the picture of demand is wider than the official inbox.

The same engine, rubric and hand-over now run as a service inside PumpGrowthOS, so the next client site starts from a working pattern rather than a blank page. The site it lives on has its own story, and so does the CRM and email foundation the leads land in.

Growth Journey

The Akdan Danışmanlık growth journey

Every layer we built for this brand, in one place.

Client Signal

What partners say

Hear from the teams we build and grow with, once the work ships and the numbers move.

Pumpmedya is a highly skilled team that masters every lever critical to business development. Open to genuine dialogue and focused on the goal, they add the human touch that makes a partnership truly productive. We gladly recommend them.

Antonio Di Bacco
Marketing Manager · Eutelsat
Next case study

The catalogue that audits itself: Bağlıca Bisiklet.

Read case study
Retail

Let's build your growth engine.

Tell us the goal and the constraints. We'll come back with a plan, not a pitch.

back to top · back to top · back to top ·