What Decides What AI Integration Actually Costs
The single biggest factor in what AI integration costs is not the AI at all — it is what the system you are connecting to actually exposes. A modern platform with a documented API is a fast, predictable build. A legacy system with no API, or one that only lets you read data and never write to it, adds real time and cost regardless of how simple the AI part looks. Everything else — how many systems are involved, whether the AI only answers questions or also updates records, and what compliance the sector brings — moves the number from there.
"AI integration" gets used loosely, so it is worth being precise about what it actually covers before pricing it. It is not the same project as building a chatbot from scratch or automating a workflow end to end, though the three overlap constantly in practice — a booking chatbot for a dental practice, for example, is a chatbot build until the moment it needs to write into the practice's calendar software, at which point it becomes an integration problem too.
What AI Integration Actually Looks Like
Adding AI search or a chat widget to an existing website. The AI answers from your pages and documents without you rebuilding the site around it. Outcome: visitors get answers your existing content already contains, without a new platform migration.
A knowledge system for your own team. Staff ask questions and get answers drawn from your policies, product data or past cases, with the same access permissions they already have. Outcome: fewer repeated questions reaching a manager or a shared inbox, and a faster answer than searching a drive.
Connecting AI to the software you already run — a CRM, a practice management system, an ERP or a helpdesk. The AI reads live records rather than a static export, and where the platform allows it, writes back: booking a slot, updating a field, flagging an account. Outcome: one connected system instead of a chatbot that cannot actually see or touch your data.
A multi-system rollout. Several of the above at once, sharing one identity and audit log so a CRM, an ERP and a support desk agree with each other rather than drifting apart. Outcome: no re-keying between systems, and one place to see what the AI did and why.
See AI Integration for how BigBerri scopes each of these, and AI CRM Integration UK: How to Add AI to HubSpot, Salesforce, Zoho and More (2026) for a CRM-specific version of this same question if your integration is mainly about Salesforce, HubSpot, Zoho or Pipedrive.
The Factor That Moves Price Most: What the Target System Exposes
Three situations, three very different projects:
- •A modern, documented API. Most current CRM, helpdesk, calendar and e-commerce platforms fall here. A clean-API connection is the fastest and least involved integration, because the platform has already done the work of exposing its data safely.
- •An older or heavily customised system with a limited or undocumented API. Common in practice management software, some ERPs and anything a business has run for a decade or more. The AI side of the build barely changes; the integration side takes longer because someone has to work out, test and stabilise a connection the vendor never fully documented.
- •No API at all. Rare today but not extinct, particularly in some legacy line-of-business software. The realistic options are a vendor-side export the AI reads on a schedule, a browser-automation layer that is more fragile and needs more maintenance, or asking the vendor whether an API is on their roadmap before committing to a workaround.
Low-code platforms such as Zapier, Make and n8n handle a large share of straightforward, well-documented connections and are a reasonable route for a single, simple integration; ask any provider whether that is what they are actually building on, since it changes both the cost and what happens if you outgrow it. Enterprise integration platforms such as Boomi and MuleSoft exist for the multi-system rollout end of the scale, where the number of connections and the governance around them, not the AI, is what is being priced. For a single business system, a custom, purpose-built connection is usually the middle ground between the two.
Read Versus Write: The Second Biggest Cost Driver
An integration that only reads data — answering questions from records it can see — is simpler to build and safer to launch than one that writes back, because writing means the AI can change something in a system your business depends on. A write-back integration needs guardrails a read-only one does not: validation before anything is saved, an approval step for anything customer-facing, and a clear log of what the AI changed and when. Ask any provider, plainly, whether your quote includes those guardrails or just the connection itself — a connection with no guardrails is not a smaller version of the same project, it is a different, riskier one.
What Decides Which Tier Fits Your Project
| Tier | What it covers | What moves the price | Timeline |
|---|---|---|---|
| Single Integration | Adding AI to one tool you already run, without replacing it | Whether the target system exposes a clean, documented API | 1–3 weeks |
| Knowledge System | A retrieval system over your own documents, respecting existing permissions | Document and data volume, and how many access levels it needs to respect | 3–6 weeks |
| Multi-System Rollout | AI connected across several systems — CRM, ERP and support together | Number of systems, how much mapping sits between them, and shared audit logging | 6–12 weeks |
Most single-site UK businesses with one clear integration need sit in the first tier; a team wanting staff to query its own documents sits in the second; a business running several connected systems that all need to agree sits in the third. Book a free discovery call at our contact page once you know roughly which shape your project is, and we will confirm the tier and give you a fixed price.
Sector Systems That Commonly Complicate an Integration
Practice and case management software varies enormously by sector, and the platform your business already runs is usually the first thing worth naming on a discovery call, not the AI feature itself. Dental practices typically run Dentally, SOE Exact or Software of Excellence; veterinary practices run ezyVet, Provet Cloud, RxWorks or Animana; recruitment agencies run Bullhorn, Vincere, JobAdder or Firefish; accountants and law firms typically run practice management software alongside Xero, QuickBooks or Sage for the accounts side. Some of these expose a clean API; others need more care. See AI for Veterinary Practices, AI for Recruitment Agencies and AI for Accountancy Firms for how this plays out in each sector specifically.
UK GDPR and Data Residency When You Connect a Model to Live Records
Once an AI model can read live business records, the records it touches are in scope for UK GDPR the moment they contain anyone's personal data — a customer, a candidate, a patient, a client. Three points come up on every integration project, and none of them are legal advice.
Where the data actually goes. Know which model provider processes the request, in which region, and whether it is used to train future models by default. Prefer a UK- or EU-hosted endpoint where the option exists, and keep the setting on file rather than assuming a default.
Sending a restricted transfer outside the UK. If an integration sends personal data to a provider or server outside the UK, that is a restricted transfer under UK GDPR Chapter V and needs a safeguard, typically the UK International Data Transfer Addendum. The ICO's own guidance, updated in January 2026, replaced what used to be called a Transfer Risk Assessment with what it now calls a data protection test — the substance is broadly the same duty to check the destination offers comparable protection, under a new name.
Automated decisions. Where an integration goes beyond answering questions into scoring or flagging a person — a lead score, a risk flag, a shortlisting decision — different rules apply since the Data (Use and Access) Act 2025 changed how UK GDPR treats automated decision-making. AI CRM Integration UK: How to Add AI to HubSpot, Salesforce, Zoho and More (2026) covers what changed and what it requires in practice; the short version is that a human still needs to be able to review or override the outcome.
None of this should stop a well-scoped project. It needs designing in from the start with your own data protection lead, not treated as a detail to fix after launch.
Questions to Ask Before You Sign a Quote
- 1. Have you actually connected to this specific system before, or would it be new to you?
- 2. Is the quote for a read-only connection, a write-back one, or both — and does the price reflect that difference?
- 3. What happens if the vendor changes or deprecates the API this integration depends on?
- 4. Where is the data processed, and is that compliant with UK GDPR for our sector?
- 5. What guardrails and approval steps are included for anything the AI can write or send?
- 6. What does support cost once the integration is live, and what does it actually cover?
- 7. Who owns the integration code and the configuration once the project is delivered?
How BigBerri Approaches AI Integration
We scope every integration the same way, whatever system it touches.
- •Discovery call first. We ask what the target system actually exposes before quoting anything, because that is what decides the real cost.
- •Fixed price, written scope. One number for the connection, the guardrails, testing and go-live, agreed before build work starts.
- •Guardrails built in, not bolted on. Read-only access by default; write access goes through validation and, where it touches a customer or client, a human approval step until you choose otherwise.
- •UK-based delivery. Built and supported from Manchester, on UK GDPR-compliant infrastructure.
- •You own the build. The integration code and configuration are yours once the project is delivered.
For the broader picture of what drives AI project cost generally, across chatbots, voice agents, automation and integration together, see AI Development Cost UK: The Real 2026 Pricing Guide. If your integration need is really about connecting AI to a sales CRM specifically, AI CRM Integration UK: How to Add AI to HubSpot, Salesforce, Zoho and More (2026) goes deeper on that one case, including native AI features already built into HubSpot, Salesforce and Zoho versus a custom connection.
Book a free discovery call at our contact page and tell us what system you are trying to connect. We will tell you plainly what it would take and give you a fixed price, not a vague estimate.
Frequently Asked Questions
What does AI integration cost in the UK?
It is quoted as a fixed price after a free discovery call rather than a day rate, because the number depends entirely on your specific systems. What moves it is whether the target platform has a modern API, whether the AI only reads data or also writes to it, and how many systems are involved. Book a free discovery call at /contact for a fixed figure against your own setup.
Can you add AI to a system that has no API?
Sometimes, but it changes the project. Options include a scheduled export the AI reads periodically or a browser-automation layer, both of which need more ongoing maintenance than a direct API connection. It is worth asking the vendor directly whether an API is on their roadmap before committing to a workaround.
Is AI integration different from building a chatbot or an automation?
They overlap. A chatbot or an automation becomes an integration project the moment it needs to read from or write to a system you already run, such as a calendar, CRM or practice management platform. Many projects are genuinely both at once, and the discovery call is where that gets untangled.
What happens to our data once it is connected to an AI model?
That depends on where the model provider processes the request and whether the data leaves the UK. A transfer outside the UK needs a safeguard such as the UK International Data Transfer Addendum. Ask any provider which model runs the integration, in which region, and whether your data is used for training by default.
How long does an AI integration project take?
A single integration with a well-documented API typically takes one to three weeks. A knowledge system built over your own documents runs three to six weeks. A multi-system rollout connecting several platforms together typically takes six to twelve weeks.