01
Inbox and communications
Classify emails, summarise threads, draft replies, detect urgency and route messages from Gmail, Outlook, Slack or WhatsApp.
Useful intelligence inside the systems that already run the business.
Custom AI integration services for UK businesses, built in London.
WiseSolutions connects AI into CRM, Gmail, WhatsApp, Slack, Drive, databases and internal tools so information moves, gets understood and reaches the right person.
Section 01 · What it means
AI integration means adding judgement to the flow of data.
Most businesses already have the data. It sits in inboxes, CRMs, chat tools, PDFs, spreadsheets, call notes, Drive folders and databases. The gap is that no one system understands what the data means or what should happen next.
A custom AI integration adds that layer. It can classify an enquiry, summarise a thread, extract fields from a document, route a request, draft a response, check a policy or flag the cases that need a human. The point is not to add a shiny AI box. The point is to remove the manual reading and forwarding that slows the team down.
WiseSolutions builds these integrations for practical teams that need production systems, not generic demos. We care about authentication, data boundaries, retries, monitoring and handover because that is where real integrations succeed or fail.
A good integration should feel almost boring after launch. The right fields appear in the CRM. The right person gets the right summary. The risky case gets escalated. The logs show what happened. Nobody has to ask which inbox has the truth.
Common AI actions
- Classification: decide what type of request arrived.
- Summarisation: turn long threads into useful context.
- Extraction: pull structured fields from messy documents.
- Routing: send the item to the right person or system.
- Review: escalate low-confidence or sensitive cases.
The output is only useful if the next step is wired properly.
Section 02 · Where it matters
The best use cases are usually hiding in repeated handoffs.
02
CRM and pipeline work
Score enquiries, enrich contact records, generate follow-up tasks and keep sales notes consistent across a team.
03
Files and storage
Read PDFs, scanned files and Drive folders, then extract the fields a human would normally copy by hand.
04
Databases and operations
Match records, update statuses, generate internal summaries and flag exceptions before they become missed work.
05
Specialty workflows
Property, healthcare, legal, finance and local services often need domain-specific routing with a clear audit trail.
Section 03 · Build model
How we build a custom AI integration.
Step 01
Source-of-truth audit
We identify which system owns each field, which records are reliable and where staff currently make manual judgement calls.
Step 02
Authentication and permissions
We use scoped credentials, OAuth where appropriate, least privilege and clear ownership so the integration does not depend on a shared personal login.
Step 03
Error handling
Every important route needs retries, fallbacks, validation and a place for exceptions. Quiet failure is not acceptable in production.
Step 04
Monitoring
We track run status, cost, failed records, latency and the business events that show whether the integration is saving time.
Section 04 · Real cases
Integrations we have built or patterned for real operations.
Clinic operations team, UK, anonymised
Inbox triage for supplier and patient operations.
A clinic team needed a calmer way to handle recurring emails. We designed an integration that reads the inbox, classifies message type, extracts useful details, stores attachments and sends exceptions to a human reviewer.
- Systems: Gmail, Drive, audit table and staff notification route.
- Controls: duplicate checks, exception queue and review notes.
- Outcome: less manual sorting and a clearer operational record.
Dental clinic, UK, anonymised
WhatsApp and website chatbot routing into one backend.
The clinic had enquiries arriving from different channels with inconsistent handoff. We designed routing that standardises the message, identifies intent, records the conversation and sends the right summary to the right team.
- Systems: website widget, WhatsApp provider and contact records.
- Controls: consent marker, staff fallback and conversation logging.
- Outcome: one operational view across two front doors.
Property management firm, UK, anonymised
Renters Rights Act assistant for landlord and tenant questions.
A property team needed help interpreting repeat questions around new regulation without giving unsupported answers. We built a grounded question flow using approved text, email capture, logging and escalation for out-of-scope questions.
- Systems: public web page, document source and enquiry database.
- Controls: source-only answers, refusal path and weekly review.
- Outcome: faster triage for repeat regulatory questions.
Multi-site CRM operator, UK, anonymised
WhatsApp, SMS and email routing for a multi-site CRM.
A growing operator needed messages from several sites to reach the right pipeline without mixing client data. We designed routing rules, templates, contact matching and activity logs across WhatsApp, SMS and email.
- Systems: CRM, messaging providers and pipeline board.
- Controls: per-site configuration, permission boundaries and audit trail.
- Outcome: faster follow-up with fewer copied messages.
Names redacted out of respect for our clients. We can share full details on a call.
Section 05 · Build route
AI integrations vs in-house build vs no-code.
WiseSolutions
Best when you need a production result quickly.
You get discovery, build, documentation, monitoring and a practical handover without hiring a full internal team first.
In-house build
Best when AI integration is a core product capability.
A strong internal team is the right answer when the integration needs constant product development and deep ownership.
No-code
Best for low-risk internal automations.
Good for quick tasks, but weak when auth, data protection, custom logic and observability become important.
Do nothing
Best when the volume is still tiny.
If the manual work takes 20 minutes a week, we will say so. The first useful threshold is usually repeated work with visible cost.
Section 06 · Pricing
Starting ranges for scoped integration work.
Proof of concept
from £4,000
- One integration path.
- Real test data and visible output.
- Two to four weeks.
- Recommendation for production scope.
Production integration
from £10,000
- Multiple systems and business rules.
- Auth, error handling and monitoring.
- Documentation and staff handover.
- Six to ten weeks for most builds.
Managed support
from £300/month
- Monitoring and minor fixes.
- Monthly review of failures.
- Dependency and API-change support.
- One named point of contact.
Usage fees, hosting and third-party platform costs are quoted separately before launch.
Section 07 · Security and handover
The integration has to be safe enough for someone else to operate.
We do not start by asking which model to use. We start by asking where the source of truth lives, who can read it, who can write to it and what happens if the output is wrong. That keeps the work grounded in the business process instead of the demo.
For regulated teams, this often means mapping personal data, checking whether a record contains sensitive information and deciding what should be logged. Some integrations can store full prompts and outputs. Others should store only status, source record ID and a short audit note.
We also design the monitoring path before launch. A failed integration should show which system failed, which records are affected, whether a retry is safe and who owns the next action. That is the difference between managed automation and a hidden script.
What we document
- System owner: who owns the source, the destination and the integration.
- Credential scope: what each token can read, write and delete.
- Data retention: how long prompts, outputs and logs are kept.
- Failure route: retries, alerts and manual review path.
- Change process: how edits are tested before production.
The handover is part of the build, not a PDF written after everyone is tired.
Section 08 · Integration detail
What changes when the integration touches real customers and staff.
01
The source system stays in charge.
We do not let an AI step become the secret source of truth. The CRM, database, inbox or document store still owns the record. The integration reads, classifies, suggests and routes around that source.
02
The human review path is explicit.
If confidence is low, the output is sensitive or the action could affect a customer, the workflow should route the case to a person with enough context to decide quickly.
03
The cost model is visible.
We estimate volume before launch and track usage after launch. A useful integration should not surprise the business with quiet per-message or per-document costs.
04
The failure state is designed.
Every integration eventually meets a timeout, malformed record, revoked token or changed API response. We decide what happens before that day arrives.
This is where WiseSolutions is useful for teams with real operational pressure: we are not only wiring tools together, we are deciding what happens when the data is messy, the user is impatient, and the business still needs a clear answer. We normally find the best first integration by watching the team work for a few days. The strongest candidates are not abstract AI ideas. They are repeatable judgement calls, slow handoffs, and records that get copied from one system into another because no one trusts the current flow. That observation step usually saves more money than any clever prompt because it shows where the process is actually breaking.
Section 09 · Platform decisions
The right integration route depends on how open the platform is.
Some systems are straightforward. They offer a clean API, OAuth, webhooks, stable documentation and sensible rate limits. Those are the integrations we can usually move through quickly.
Other systems make the work slower. They may have partial APIs, awkward permissions, file exports instead of events, or fields that are visible in the interface but not available through the API. We find that early and tell you before the budget is committed.
When a platform has no reliable integration path, we do not pretend otherwise. Browser automation and screen scraping can be useful for internal rescue work, but they are rarely the right foundation for a customer-facing process.
The best integrations are simple to explain. This system receives the event, this step understands it, this record gets updated, this person sees the exception, and this log proves what happened.
Decision factors
- API coverage: can we read and write the fields the business actually needs?
- Auth model: can we use scoped tokens or OAuth instead of shared logins?
- Event model: are webhooks available or do we need controlled polling?
- Rate limits: will the workflow survive normal daily volume?
- Recovery path: can failed records be replayed safely without duplicates?
These checks save money because they catch bad integration ideas before build week three.
Section 10 · FAQ
Questions about AI integrations.
What counts as a custom AI integration?
Any system where intelligence is added to an existing workflow: classification, summarisation, extraction, routing, drafting, enrichment or decision support across the tools your team already uses.
How do you handle data security?
We start with a source-of-truth and permissions audit, then design the integration so credentials stay protected, data movement is logged and sensitive content only reaches the systems that need it.
Can this be GDPR-compliant?
Yes, if it is designed with data minimisation, lawful basis, retention, access control and deletion routes from the start. For higher-risk use cases we recommend a DPIA before production.
Do you work with OAuth?
Yes. OAuth is the right route for many CRM, email, Drive and workspace integrations. We avoid shared passwords and prefer scoped tokens with clear owner and renewal rules.
What platforms do you integrate with?
Common systems include CRM platforms, Gmail, WhatsApp providers, Slack, Drive, databases, spreadsheets, website forms, booking tools and custom APIs. We will say no if the platform is too closed or unstable.
What do you not integrate?
We avoid systems where the only route is brittle screen scraping, shared personal logins, unsupported private endpoints or workflows that would create a compliance problem for the client.
How quickly can you build a first version?
A focused proof of concept usually takes two to four weeks. Production integrations with monitoring, error handling and documentation usually take six to ten weeks.
Do you use webhooks or polling?
Webhooks are better when the source system supports them reliably. Polling is useful when events are not available or when we need controlled batching. The right choice depends on rate limits and failure modes.
How do you monitor integrations?
We track trigger volume, failed runs, retry count, processing time, cost and important business events. Alerts go to the people who can actually fix the problem.
Can you maintain the integration after launch?
Yes. Managed support starts from GBP 300 per month and covers monitoring, minor fixes, dependency changes and monthly review of failures or missed cases.
Can you connect multiple client sites or teams?
Yes. We design multi-site routing where each business, inbox or location has the right configuration, permissions and logs without mixing data between clients.
⬡ Node · 07 · ai-integrations-discovery
Have tools that should already be talking to each other?
Tell us which systems and what should flow between them. We will reply within one working day with first questions or a discovery call slot.