AI Agents & Workflow Automation: Agents that do the legwork. People who make the call.
Intelligence
What are AI agents and workflow automation?
An AI agent is software that uses an AI model to work through a multi-step task — gathering information, using tools such as your CRM or inbox, and producing a result — within limits you set. Workflow automation connects the steps between your systems so information moves without re-typing. Witty builds both, with human approval wherever a mistake would matter.
We run AI agents inside our own business: our internal website-audit agent captures pages in a real browser, runs accessibility checks and drafts findings for a person to review before anything reaches a client. We bring the same approach to yours — useful autonomy, clear boundaries, nothing irreversible without sign-off.
Sound familiar?
Problems we solve
Leads go cold before anyone replies
An agent can capture, enrich and score a lead in seconds, draft a tailored reply and alert the right person.
Research eats your best people’s time
Company research, competitor scans and document review can be prepared by an agent and checked by a person.
Reports take a day every month
Automated pipelines collect the numbers, write the summary and send it for review.
Tools do not talk to each other
Forms, inboxes, CRM, spreadsheets and chat can be orchestrated into one flow with logging.
What’s included
What we deliver
Pick what you need, or let us recommend a scope after a short discovery call.
Lead research & qualification agents
Enrich new leads from public sources, score them against your criteria and route them with a summary.
Customer support assistants
Triage, suggested replies and answers from your knowledge base, escalating to people for anything sensitive.
Internal knowledge assistants
Ask questions of your policies, proposals and documentation, with sources shown.
Research & information-processing agents
Read, extract and summarise documents, emails, websites and spreadsheets into structured data.
Cross-platform orchestration
Automations across forms, email, CRM, project tools and messaging using n8n, Make, Zapier or custom code.
Human-in-the-loop controls
We design agents with approval queues, role-based access, audit logs and a way to pause them at any time.
Platforms & tools we typically use
- Claude
- OpenAI
- n8n
- Make
- Zapier
- Python
- Node.js
- Playwright
- HubSpot / CRM APIs
- Slack & email
How it works
From first call to results
Step 1: Pick one workflow
We start where volume is high, rules are clear and errors are recoverable.
Step 2: Map the steps & decisions
Which steps the agent may take alone, which need approval, and what it must never do.
Step 3: Build & shadow
The agent runs alongside your team first, so we can compare its work with theirs.
Step 4: Go live with guardrails
Monitoring, logs, cost limits and an owner on your side.
Explore a workflow
Where the agent works — and where a person decides
From a new enquiry to a qualified, followed-up lead — in minutes, with a person approving the reply.
- TriggerNew enquiry arrives
- SystemDetails captured
- AI stepAI qualifies the lead
- SystemCRM updated
- AI stepFollow-up drafted
- Human checkTeam member approves
- OutcomePerformance tracked
New enquiry arrives
A message comes in through your website form, email or WhatsApp.
Example“We need a new website for our three clinics.”
Details captured
Name, company, service and budget are pulled into a structured record.
Exampleservice: website · sites: 3 · budget: stated
AI qualifies the lead
The lead is scored against criteria you define, with a short reason you can read.
Examplefit 82/100 — clear scope, timeline this quarter
CRM updated
Contact and deal are created, tagged by source and assigned to the right person.
Exampledeal created → owner: sales lead
Follow-up drafted
A personalised reply is prepared, referencing what the prospect actually asked.
Exampledraft ready · 3 questions to confirm scope
Team member approves
Your colleague edits or approves the draft. Nothing is sent without them.
Exampleapproved with one edit · sent
Performance tracked
Time-to-reply and conversion by source appear on your dashboard.
Examplereply time: 6 min (example)
Common questions answered from your own content; anything sensitive goes straight to a person.
- TriggerCustomer message
- AI stepIntent and urgency
- SystemOrder looked up
- AI stepAnswer drafted with sources
- Human checkAgent reviews
- OutcomeResolved and learned from
Customer message
A question arrives by email, chat or social DM.
Example“Where is my order? It was due yesterday.”
Intent and urgency
The message is classified — order status, complaint, product question — and urgency is flagged.
Exampleintent: order status · tone: frustrated
Order looked up
The assistant checks the order in your e-commerce platform through its API.
Exampleorder #—— · courier delay reported
Answer drafted with sources
A reply is drafted from your policies and the order data, citing the source used.
Exampledraft cites: delivery policy §2
Agent reviews
Low-risk FAQs can be auto-sent above a confidence threshold you set; everything else is reviewed.
Examplerouted to agent (frustrated customer)
Resolved and learned from
The ticket is tagged and gaps in your knowledge base are flagged for an update.
Examplegap flagged: courier delay FAQ
Proposal prep without the blank page — pricing stays a human decision.
- TriggerDeal reaches “proposal”
- SystemContext gathered
- AI stepOutline drafted
- Human checkLead sets scope and price
- SystemSent and scheduled
- OutcomePipeline updated
Deal reaches “proposal”
A deal moves stage in your CRM.
Examplestage → proposal
Context gathered
Call notes, emails and requirements are collected from your tools.
Example4 notes · 1 call summary · 2 emails
Outline drafted
A proposal outline and a list of open assumptions are drafted in your template.
Exampleoutline + 5 assumptions to confirm
Lead sets scope and price
Scope, pricing and commitments are always decided by a person.
Examplepricing added by sales lead
Sent and scheduled
The proposal is sent and follow-up tasks are created automatically.
Examplefollow-up task: in 3 days
Pipeline updated
Forecasts and pipeline reports refresh without manual entry.
Examplepipeline report refreshed
One good piece of content becomes a week of posts — edited by people who know your brand.
- TriggerNew article published
- AI stepRepurposed
- Human checkEditor reviews
- SystemScheduled
- OutcomeResults fed back
New article published
A blog post, case study or video goes live.
Examplecase study published
Repurposed
Drafts for social posts, captions, a newsletter blurb and short-video hooks are generated in your tone of voice.
Example6 post drafts · 1 email blurb
Editor reviews
An editor checks facts, tone and claims, and picks what to publish.
Example4 approved · 2 rewritten
Scheduled
Approved posts are queued in your scheduling and email tools.
Examplequeued across 3 channels
Results fed back
Engagement is tracked and the best-performing formats inform the next batch.
Examplecarousel outperformed (example)
The monthly report writes its first draft; your manager adds the judgement.
- TriggerMonth ends
- SystemNumbers pulled
- AI stepSummary written
- Human checkManager adds context
- OutcomeReport delivered
Month ends
A schedule kicks off the reporting run.
Example1st of the month, 07:00
Numbers pulled
Data is collected from analytics, ads, CRM and accounting tools.
Example4 sources connected
Summary written
A plain-language summary explains what changed, with anomalies highlighted.
Example“Enquiries up; paid search cost rose.”
Manager adds context
A person checks the numbers and adds what the data cannot know.
Examplecontext added: campaign paused mid-month
Report delivered
The report goes to stakeholders on time, every month.
Examplesent to 5 stakeholders
FAQ
Questions we’re often asked
Straight answers about ai agents & workflow automation. Can’t see your question? Ask us directly.
What is the difference between an automation and an AI agent?
An automation follows fixed rules: when X happens, do Y. An AI agent can interpret unstructured information and choose between steps — for example reading an email, deciding it is a sales enquiry and drafting a reply. Most good systems combine both: predictable automation for the plumbing, AI only where interpretation is needed.
Can agents send emails or update records on their own?
They can, but we only allow that for low-risk actions you have approved. Anything customer-facing, financial or irreversible goes to an approval queue by default.
Which tools can you connect?
Anything with an API or a reliable integration platform connector: CRMs, helpdesks, email, calendars, spreadsheets, project tools, e-commerce platforms and databases. We confirm specifics during the workflow audit.
How do we know it is working?
We agree measures before building — time per task, response times, accuracy against human review, cost per run — and report against them.
Works well with
Related services
- IntelligenceAI SolutionsPractical AI built into your workflows, website and systems — with people in control.
- BuildCustom Software & Business SystemsERP, CRM, dashboards and internal tools that replace spreadsheets and copy-paste work.
- GrowthDigital Marketing & Lead GenerationPaid search, social campaigns, funnels and email that turn attention into qualified enquiries.
Further readingWhere AI automation pays off in a small business — and where it does not
Your project is the missing tile.
Tell us what you’re trying to achieve. We’ll come back with honest advice, a recommended approach and next steps — whether or not we’re the right fit.