Kortical

Kortical is a UK-based enterprise AI company, founded in 2016, that sells 2 products. The first is AI Agents, which covers bespoke agents built with its data scientists and developers plus KorticalChat, a self-serve RAG chatbot builder for sales and support. The second is AI Video Personalisation. For this directory, KorticalChat is the part that matters: a no-code way for a support, sales or operations team to build a chatbot trained on its own data, with a services team behind it when the job outgrows a self-serve tool. The footer gives London as its base and names the company as Kortical Ltd.

What you build and how

KorticalChat is described on the company's site as "our self-serve RAG chatbot builder", and the company's own guide to it is titled "Build a custom AI chatbot trained on your data, no coding required". RAG here means the chatbot answers from a knowledge base you supply rather than from the model's general memory. The guide, written by the company's COO and co-founder, walks through 5 steps, and they double as a tour of the product.

Step 1 is defining the purpose. The guide asks you to decide who will use the bot (customers who know little about your business, or employees doing a specific task), what its primary function is (customer support, lead generation, answering user queries, writing sales emails, drafting LinkedIn posts), and what personality it should have. It notes that most people choose friendly and professional, but you can go further: the company says it instructed one bot for an antiques client to be eloquent like Stephen Fry and another to use Scottish flair for a Scottish tourist site. You can also paste examples, such as past emails to customers, so the bot copies your tone and style.

Step 2 is the model. The guide says you pick and test a large language model inside the platform, with no machine learning hire and no Python chatbot to code yourself. Its advice is that ChatGPT is "the chattiest" of the models, GPT4 omni is best at remembering the context of a conversation, Claude is often better for marketing content and articles, GPT4 turbo suits large context, and the cheaper GPT4 omni mini works well for tightly constrained tasks.

Step 3 is the knowledge base. You upload website links, documents, PDFs or other data, then test the answers. The guide warns that out-of-date web pages, forgotten processes or old stock information will surface here, and that you can exclude that content or change the answers. An overwrite function lets you replace a correct but blunt answer with your own wording. The guide's worked example is an online shop that charges for returns: instead of simply quoting the fee, the bot explains the environmental cost of returns and points shoppers to the sizing information the shop has invested in, so they choose the right size the first time.

Step 4 is the interface. Chatbots can be a chat widget on a webpage or a full-screen chatbot for larger copy tasks, and KorticalChat provides a built-in chat UI rather than leaving you to code one. In a few clicks you match it to your branding, including colors, fonts and logo, and set a greeting that starts the conversation, such as an invitation to ask about products and installation, or an instruction to paste a news article for the bot to turn into commentary for your website. You test the bot inside the platform before release.

Step 5 is deployment and monitoring. You deploy on the platform you want, such as a website or a mobile app, and on a website it takes "a few lines of code snippet on your website header". The platform's insights tools and analytics show what users ask and where they get stuck, and the guide treats those conversations as a source for improving the website and social posts. It says to start with a basic chatbot, add integrations with a back-end ordering system or a CRM later, and retrain the bot as often as you update the product catalog or website information.

Beyond the self-serve builder, the company sells bespoke AI agents "built with our expert data scientists and developers, joined up to your data and systems". These automate whole business workflows end to end, and the company says they typically reach production in weeks rather than months. An AI consulting service sits beside them.

The second product, AI Video Personalisation, generates a unique video for every customer, with the customer's name, their car and their next step, plus personalized thumbnails and landing pages. The site claims over 1,000% ROI for it. The login menu also lists the Kortical ML Platform for AutoML, ML Ops and deployment, and the footer adds AI Cloud and Data Prep, so Kortical is a wider AI vendor than a chatbot specialist.

What the company's site states

The site publishes no prices for any product. KorticalChat is "free to try" and "free to start", and the company says it can be live the same day. Bespoke agents and video campaigns run through a booked demo, where the company says it will show "the quickest path to production".

Security and compliance claims are specific. The company says it has been ISO 27001 certified since 2019, independently audited every year, with zero security breaches. It says it is GDPR and EU AI Act compliant, "built for European regulation from day one", that client data is never used to train models, is encrypted at rest and in transit, and that EU and US hosting options exist. It also says it is trusted with NHS patient-level data in the cloud.

The homepage says it is trusted by the NHS, Deloitte, Santander and Reed, and its About text adds Hyundai and Volvo. It attaches figures to several of them, all as the company's own claims. For Deloitte, AI agents took an R&D tax compliance workflow from 5 hours to 6 minutes. For Santander and Hyundai, models lifted customer retention by 8%. For Reed, AI agents delivered a return of over 600% and more than 5,000 hours saved, and Jon Campbell of Reed is quoted saying "Kortical has automated over 5,000 hours, a huge productivity boost." For NHS Blood and Transplant, a demand planning platform backed by a £750,000 Innovate UK grant cut blood supply chain wastage by 54%. These are bespoke engagements, not KorticalChat deployments.

The company also claims a 92% success rate going from proof of concept to production, says in its FAQ that 92% of its deployments have delivered positive ROI, holds a granted US patent on its platform, and publishes peer-reviewed research including "Transformers Can Learn Rules They've Never Seen: Proof of Computation Beyond Interpolation". The site says each of these claims links to an independent source.

Who it suits and who it does not

KorticalChat suits a support, sales or operations team that wants a chatbot grounded in its own documents and product catalog, and wants it on the website the same day without writing code. The guide's examples are customer support, lead generation and internal writing tasks, and the choice of models lets a team trade cost against quality. A team that expects its needs to grow into integrations with ordering systems or a CRM has a stated path for that.

The bespoke agent side suits larger organizations with workflows complicated enough to justify a services engagement, such as tax compliance or demand planning. The named clients are large enterprises, and the results quoted come from that side of the business.

The site gives no prices or message limits for KorticalChat, so the free trial is where a team learns what it will cost. The builder is set up by instructions, examples and uploaded content rather than a branching flow canvas, which suits answer-from-knowledge bots better than scripted decision trees.