Most conversations about AI start with a model. The useful ones start with a piece of work: a request that waits too long for an answer, a report someone rebuilds every month, a decision that depends on information nobody can find quickly.
That is the difference between experimentation and value. Our position is simple: practical value, not experimentation. AI earns its place when it sits inside a real business process, works from trusted data and acts within the same roles and approvals as everyone else.
Start from the process, not the model
We think about AI as a chain, and every link has to hold:
- Business process: the work that should get easier, faster or better informed.
- Trusted data: the records that process depends on, in a state people already rely on.
- AI and agents: the capability that reads, drafts, answers or triggers.
- Roles and controls: who may do what, who approves, and what is recorded.
- Business outcomes: the result the organisation actually cares about.
Skip the first link and the project becomes a demonstration looking for a use. Skip the second and the answers cannot be trusted. Skip the fourth and nobody is accountable for what the AI does. When an AI effort stalls, look for the missing link before looking at the model.
Automate, augment, accelerate
Once the chain is in place, AI tends to help in three ways.
- Automate the routine: the steps that follow the same pattern every time and do not need judgement.
- Augment people: a first draft, a summary or a suggestion that a person reviews, edits and owns.
- Accelerate the flow: fewer handovers and less waiting between one step and the next.
In business terms, that shows up as four kinds of impact: finding insight faster, automating routine work, supporting better decisions and triggering connected actions across the systems a process touches.
AI with control
The principle we apply is AI with control: AI acts within roles and approvals, and people stay accountable.
That sounds like a constraint. In practice, it is what allows AI into processes that matter. A draft that a named person approves can be used in front of a customer. An answer that shows where it came from can be checked. An action that requires approval leaves a record. Control is not the brake on adoption; it is the condition for it.
What it looks like in practice
DONOR 360, the charity CRM ConsultTik builds and implements, shows the pattern in everyday work:
- AI drafts, people approve. In the omnichannel inbox, AI prepares replies across channels such as email, SMS and WhatsApp Business; a person approves before anything is sent.
- Answers with references. Ask AI answers questions from live data and shows the references behind the answer, so a manager can check it rather than take it on trust.
- Suggestions with reasons. On the donor record, AI proposes a next best ask and explains why, leaving the decision with the fundraiser.
- An honest voice. The AI voice agent speaks Arabic and English, discloses that it is virtual, and hands over to a person on request.
- Nothing irreversible without a person. An emergency appeal goes live only after approval and a fundraising-permit reference; there is no payout without approval; and approvals sit in an inbox with a full audit trail.
None of these patterns is specific to charities. Replace donors with customers, suppliers or employees and the same design holds: AI does the preparation, people make the commitments, and the system records who decided what.
The Microsoft stack behind it
Because the chain runs through business systems, the technology has to as well. The Microsoft platforms we work with map onto it naturally: Dynamics 365 and Power Platform where processes and records live, Microsoft Fabric for data people can trust, Copilot Studio and Microsoft Foundry for copilots and agents, and Azure for the cloud and the governance underneath.
Building on the platforms an organisation already owns also lets AI work with the security model, roles and audit capabilities already in place, instead of creating a parallel world with its own rules.
Where to start
For organisations across the GCC, a practical first step is small and specific:
- Choose one process where waiting, re-keying or searching is visible every week.
- Check the data that process depends on, and fix what people do not trust.
- Decide the roles: what the AI may draft, what a person must approve, and what is recorded.
- Agree how you will recognise the outcome before you build anything.
Then build the smallest version that works inside the real process, with the controls switched on from the first day. It may sound slower, but nothing has to be rebuilt later to be trusted.