AI Workflow Implementation: Redesigning the work, not just adding a tool

Executive summary

McKinsey found that redesigning workflows has the biggest effect on whether an organisation sees EBIT impact from generative AI. Yet only 21% of AI-using organisations had fundamentally redesigned any workflows.¹
Speed matters commercially. Research published in Harvard Business Review found that firms contacting a lead within an hour were nearly seven times more likely to qualify it than those that waited longer.²
Most delays in service businesses sit in hand-offs: between inbox and CRM, between sales and operations, between a request and an approval.
Implementation means mapping one workflow end to end, deciding what AI, automation and people each do, and building it into existing tools.
Success is measured against a baseline: response time, hours saved, errors and conversion.

01Market context

In McKinsey’s March 2025 study, the single factor most associated with bottom-line impact from generative AI was the redesign of workflows. Only about one in five organisations using AI (21%) had fundamentally redesigned at least some of them.¹

Exhibit 1 Few organisations have redesigned their workflows

AI-using organisations that have fundamentally redesigned at least some workflows21%

Organisations tracking KPIs for generative AI solutions (fewer than one in five)20%

Source: McKinsey & Company, March 2025. The second figure is shown at the upper bound of “less than one in five”.

The commercial cost of slow workflows is well documented. An analysis of US companies published in Harvard Business Review found that those which tried to contact a lead within an hour of receiving it were nearly seven times as likely to qualify it as those which waited even an hour longer.²

02The challenge

AI bolted on. A chatbot is added to the website, but enquiries still land in an inbox that someone checks twice a day.
Hand-offs nobody owns. Information is copied between email, spreadsheets and the CRM by whoever has time.
Rules that live in people’s heads. Who gets which lead, what needs approval and when to follow up are unwritten.
No test before launch. Workflows go live without being tried on real cases, and trust is lost in the first week.

03Our point of view

Implementation is a design job before it is a technology job. We take one workflow, write down every step and hand-off, and decide what each step should be: done by AI (reading, sorting, drafting), handled by simple automation (moving data, creating tasks, reminders), or kept with a person (pricing, commitments, judgement). Only then do we build.

We build into the tools you already use wherever possible. A new platform is a last resort, not a starting point.

04How implementation works

Design. The future workflow is drawn step by step, with owners, rules and approval points agreed.
Build. AI and automation are configured inside your existing email, WhatsApp, CRM and documents.
Test. The workflow runs on real past cases, then live with a small group, before full rollout.
Launch and hand over. The team is trained, the workflow is documented, and results are tracked from day one.

Exhibit 2 From design to launch

1. Design
Steps, owners, rules

2. Build
Inside existing tools

3. Test
Real cases, small group

4. Launch
Train, document, measure

Our implementation method.

05What you receive

One or more tested workflows, running inside the tools your team already uses.
Workflow documentation: steps, rules, owners and approval points.
A results dashboard showing the agreed measures against the baseline.
Team training for everyone who touches the workflow.

06Measures of success

Time to first response and time to resolution
Hours of manual work removed each week
Errors, duplicates and missed follow-ups
Conversion rate at each stage of the workflow
Share of cases handled without rework

07Risks and governance

Human approval stays in place for pricing, commitments and sensitive cases.
Data protection: personal data is handled in line with the UAE PDPL.³
Fallbacks: if an AI step fails, the case goes to a person, not into a void.
Change management: the people who do the work help design it.

This paper is general guidance and not legal advice.

08Illustrative scenario

ILLUSTRATIVE — NOT A CLIENT CASE STUDY

A three-branch clinic group takes appointment requests by phone, website form and WhatsApp. Requests are copied into the booking system by front-desk staff, and confirmations go out late or not at all.

The implemented workflow reads each request, identifies the service and preferred branch, checks availability, and drafts a booking for front-desk approval. Confirmations and reminders go out automatically. Medical questions are always routed to a person. Response time and no-show rates are tracked against the baseline.

09Frequently asked questions

Do we need new software?: Usually not. We build into the tools you already use, and recommend new tools only where there is a clear gap.
How many workflows should we start with?: One. A single workflow done well builds trust and evidence for the next.
Who owns the workflow after launch?: A named person in your team, with documentation and training. We can also support it under Managed AI Operations.

AI Workflow Implementation