Skip to main content
BigBerri

Manchester software house. We build websites, web and mobile apps, MVPs and AI for UK businesses.

Book a Free Discovery Call
info@bigberri.com
@bigberri_official
BigBerri
AI Service

AI Integration

AI integration adds a specific AI capability to a system you already run — search, a knowledge assistant over your own documents, a smart feature inside your CRM — connected through the system's own API rather than a rebuild. BigBerri builds it for UK businesses on accounts you own, with a fixed price quoted after a free discovery call.

  • Single feature live in 1–3 weeks
  • Fixed price, never a day rate
  • Runs on your accounts, not ours
  • Your data grounds answers, never trains someone else’s model
Book a Free Discovery CallWhatsApp us

Or call 07305 694001.

What We Integrate

AI Capabilities for Every System

We add AI where it delivers the most value — without touching what already works.

AI Search & Discovery

Add semantic search, AI-powered filtering, and natural language queries to any website or application.

RAG Knowledge Systems

Connect AI to your documents, policies, and product data. Staff or customers get instant, accurate answers from your specific content.

CRM AI Features

Lead scoring, sentiment analysis, auto-summarisation of calls and emails, smart follow-up suggestions — layered onto your existing CRM.

AI Content Generation

Automatically generate product descriptions, email drafts, reports, and other content within your existing workflows.

Website AI Layer

Add AI chatbots, smart forms, personalisation, and AI-powered recommendations to your existing website without rebuilding it.

API & Webhook Connections

Connect an AI model provider to your existing applications via REST APIs, GraphQL, or webhook triggers.

Integrations

The systems it connects to

The model itself is rarely the hard part of an AI integration. What decides the scope is what the business system on the other end permits: some expose everything through a documented API, some allow reading but not writing, and some need the vendor to approve the connection first. We check yours before quoting.

Model providers

  • OpenAI
  • Anthropic
  • Google
  • Azure OpenAI Service
  • Self-hosted open-source models

CRM & sales

  • Salesforce
  • HubSpot
  • Zoho
  • Pipedrive
  • Microsoft Dynamics

Website & commerce

  • WordPress
  • Shopify
  • WooCommerce
  • Custom web apps

Documents & knowledge

  • Google Drive
  • SharePoint
  • Notion
  • Confluence

Helpdesk & support

  • Zendesk
  • Intercom
  • Freshdesk

Everything else

  • Any system with a documented API
  • Internal databases
  • Bespoke line-of-business tools
  • ERP and finance systems
Scope

What a Project Looks Like

Every project is scoped and quoted after a free discovery call, so you get a fixed price for your situation rather than a number that assumes someone else’s. These are the shapes most engagements take.

Single Integration

Adding AI to one tool you already run, without replacing it.

Typical timeline: 1–3 weeks

  • AI added to one tool (CRM, site, helpdesk)
  • API or webhook connection
  • Prompt and guardrail design
  • Usage monitoring
Book a Free Discovery Call about Single Integration
Most common

Knowledge System

Teams who need answers drawn from their own documents, with permissions respected.

Typical timeline: 3–6 weeks

  • Retrieval over your documents
  • Permissions and data residency
  • Web, Slack or Teams access
  • Quality evaluation set
Book a Free Discovery Call about Knowledge System

Multi-System Rollout

Businesses connecting AI across several systems that must agree with each other.

Typical timeline: 6–12 weeks

  • AI across CRM, ERP and support
  • Shared identity and audit logging
  • Model routing for cost control
  • Ongoing optimisation
Book a Free Discovery Call about Multi-System Rollout

Timelines are typical and depend on how many systems are involved and how ready your data is. Tell us what you need and we will give you a fixed price and a date.

The Build

What the weeks actually look like

This is a typical single-integration build. A knowledge system spanning several sources or a rollout across multiple systems runs longer, on the same weekly structure. The column on the right is the part most proposals leave out: an integration stalls on API access and on what the business actually wants the AI to see, far more often than on the model itself.

Week by week build plan and what we need from you
WhenWhat we doWhat we need from you
Week 1Audit and designWe review the system in scope, check what its API actually permits before promising anything, agree what the AI should and should not be able to see, and design the security model and the failure path alongside the happy path.Access to the system being connected, a decision on what data the AI should and should not see, and any documents or knowledge base it needs to ground its answers in.
Week 2Build and connectWe build the connection on your own accounts — an API call, a webhook, or a retrieval pipeline over your documents — write the guardrails, and test it against real inputs from your system, not sample data.Read or write access on your own logins, and someone to review the AI’s draft outputs before they reach a customer or a live record.
Week 3Validate and go liveWe test outputs against a quality evaluation set built from real questions, tune the guardrails until the answers are reliably right, deploy to production with monitoring, and watch the first runs closely.Review of real outputs during the settling-in period, and a decision on where the human-review step sits for the long term.
The Boundary

What it must never do

An AI layer sitting inside your systems can reach further than most software, which is exactly why we agree its limits before we build the part that reaches them. An integration that respects a boundary is worth more than one that can do everything, and the most expensive failures we are asked to fix are the ones that quietly went beyond what anyone had agreed to.

The list is yours, not ours. We draft it from your systems and your data obligations, you correct it, and it becomes the specification the integration is built against and tested on.

  • Send your data to a model provider for training rather than grounding a live answer, without your explicit agreement
  • Write to a system beyond the access scope you granted, or before you have approved what write access means here
  • Store an API key or credential anywhere but your own environment or secrets manager
  • Return an answer with no way to tell a grounded response from a fabricated one — retrieval-based answers cite what they drew from
  • Run without monitoring — a broken connection or a silent quality drop raises an alert to a named person, not a support ticket weeks later
  • Delete or overwrite a record without an audit trail you can read and reverse
What to Check

Three things to ask any integration supplier

Not legal advice, and your own advisers should confirm how each applies to you. But these are the questions that separate a supplier who has thought about putting AI inside your systems from one selling a demo.

Is our data used to train someone else’s model?

Not by default with the providers we build on — OpenAI’s own enterprise privacy documentation confirms API inputs and outputs are not used to train or improve its models by default. We confirm the equivalent commitment in writing for whichever provider a project uses, in the contract rather than as a verbal assurance.

Where does the data go, and who is the controller?

You stay the data controller. The integration processes only the data the task needs, under a lawful basis you have recorded, on infrastructure in the UK or EU under a data processing agreement where the data is sensitive. Any model provider it relies on is named as a sub-processor rather than left implicit.

What happens if the AI gives a wrong or fabricated answer?

A retrieval-based system should cite what it drew its answer from, so a wrong answer is traceable rather than a mystery, and a human-review step sits in front of anything customer-facing or high-stakes until the evaluation set shows it is reliable. Ask any supplier how they would know if the AI started quietly getting things wrong.

After Launch

How handover and support work

An integration is something you come to depend on quietly, so it has to keep working whether or not you are still talking to us. The arrangement is built to survive the end of the engagement rather than depend on it.

Built on your accounts

Every integration runs on your own logins and API keys, with us as an invited collaborator where a platform allows it. Ending the relationship means removing a user, not migrating an integration off our infrastructure — because there is none of ours to migrate off.

What you receive

The integration written up as a document you can read, the evaluation set it was tested against, the monitoring dashboard, and the logs. Enough for another developer to pick it up without ringing us.

The first fortnight

We watch every run while it beds in, because the queries and edge cases an integration trips over are the ones nobody thought to mention. Fixes to the guardrails ship while you watch rather than at a quarterly review.

After that

Optional ongoing monitoring and improvement as your systems and content change. Agreed before launch rather than during an incident: what is covered, how quickly we respond, and what counts as new work. Every build has a named person to escalate to.

Who It's For

Built for the systems UK businesses already run

Each sector page sets out where AI features tend to earn their place there, and the systems they usually run on.

Accountancy firmsRecruitment agenciesEstate agentsLaw firmsInsurance brokersE-commerce & retailVeterinary practicesLogistics & haulage

Not listed? If you already run a system that could do more with an AI feature added to it, it is worth a conversation. Tell us about yours.

FAQ

The questions people ask before hiring us

Do we need to rebuild our existing systems?

No — that is the point of AI integration. We add AI capabilities to what you already run, whether that is your website, CRM, ERP or custom software, connecting via APIs, webhooks or a retrieval pipeline without disrupting existing functionality. If a system genuinely needs rebuilding first, we say so before quoting rather than after.

What is RAG and do we need it?

Retrieval-Augmented Generation (RAG) connects an AI model to your specific business data — documents, policies, product catalogues, SOPs — so it answers from your content instead of a generic response. You need it if you want an assistant that knows your business specifically rather than the internet in general; you do not need it for a narrower feature like AI search or a single automated draft, which reads live data directly instead.

How long does an AI integration actually take?

A single feature added to one tool — AI search on a website, or one automated draft in your CRM — typically runs 1–3 weeks. A knowledge system that answers from your own documents, with permissions respected, typically runs 3–6 weeks. Connecting AI across several systems that all have to agree with each other runs 6–12 weeks. What moves it between tiers is how many systems are in scope and what each one’s API actually permits, which we confirm before quoting rather than guessing.

Is our data used to train someone else’s AI model?

Not by default with the providers we build on. OpenAI’s own enterprise privacy documentation states it does not use API inputs or outputs to train or improve its models by default, for the API platform and its business tiers alike — a policy in place since March 2023. We confirm the same commitment in writing for whichever provider a project uses, and your data is used to ground the AI’s answers, not to retrain it.

How secure is connecting AI to our business data?

Credentials live in your own environment or secrets manager, never hardcoded into the integration. Access is scoped to what the task needs — read-only where nothing needs to be written, and a defined write scope where it does — under a data processing agreement, with UK or EU hosting for anything sensitive. We follow this as a default, not an upgrade.

What happens when an integration breaks?

Integrations break for reasons outside anyone’s control — a vendor changes an interface, a credential expires, a rate limit moves. We build monitoring in from the start so a failure raises an alert to a named person rather than going unnoticed until someone spots missing data weeks later. Under a support arrangement, fixing that is our job rather than a new quote.

Who owns the integration if we stop working together?

You do. It runs on your accounts and against your credentials, the code is yours on completion and full payment, and the documentation is written so another developer can take it over. We deliberately avoid putting anything of ours in the middle of your systems, because a dependency on us is a risk to you rather than a feature.

How do we know you will still be here in two years?

The design answers this better than any promise could. Nothing depends on infrastructure we own, so an integration keeps working whether or not we do. Big Berri Limited is registered in England and Wales, company number 16562429, and ICO registered under ZC119995. Both are checkable, which is more than a reassurance is worth.

Guides

Read before you buy ai integration

Before You Get In Touch

What you can hold us to

Review our own live products, check our company registrations and read the commitments that shape a project. These give you concrete work and clear terms to discuss before booking a call.

Our own product

Writford

A workspace for UK legal firms connecting cited research, matter records, time recording and billing. The research workflow keeps source checking and professional review with the solicitor.

Explore Writford

Our own product

Let AI Browse

Tools for connecting shopping assistants with storefront product, cart and checkout actions. Its WordPress and storefront integrations show the kind of connected web workflows we build.

Explore Let AI BrowseRead the LetAIBrowse case study

What the discovery call actually is

Thirty minutes, free, no obligation and no salesperson. You describe what is costing you time or customers, we ask about the systems involved, and you leave with an honest view of whether this is worth doing. We regularly end that call by saying the tool you already pay for covers it.

You own the result

The code, the repository and the accounts are set up in your company name rather than ours, and they are yours on completion and full payment. Nothing runs on infrastructure we control, so ending the relationship does not end the thing we built.

A fixed price, never a day rate

Every project is quoted as a fixed price with a written scope after the call, so an over-run is our problem rather than yours. If the scope changes, we say so and re-quote before doing the work, not after.

Where your data sits

In the UK or EU, under a data processing agreement, kept only as long as you specify. Any model provider we rely on is named as a sub-processor rather than left implicit, and we will tell you plainly where something you have asked for would put a commitment of yours at risk.

What happens when it goes wrong

Agreed before launch rather than negotiated during an incident: what is covered, how quickly we respond, and what counts as new work. Every build has a defined escalation to a person, because the failure that costs you is the one nobody noticed.

Two things you can check

Big Berri Limited is registered in England and Wales, company number 16562429, and registered with the Information Commissioner's Office under reference ZC119995. Both are public registers. Look us up rather than taking our word for it.

Book a Free Discovery CallAsk a question firstor call 07305 694001

Add AI Without Starting Over

Book a free discovery call — we'll review your existing systems and show you where AI can make the biggest impact.

Book a Free Discovery Call
WhatsAppBook a Free Call