Freddy AI ships in three flavours: Copilot for agents, AI Agent for customers and employees, and Insights for leaders. Turning everything on at once produces an impressive demo and a disappointing quarter.
Each capability depends on data the organisation may or may not have. A bot with no knowledge base deflects nothing. A copilot that summarises miscategorised tickets summarises the mess faithfully. This is the order we recommend to Freshworks clients in the GCC, and what to fix before each step.
Step 0: Fix categorisation and the knowledge base
Before any AI feature, make sure tickets are reliably categorised on arrival and that the knowledge base covers your top fifty repeat questions in the language customers or employees actually use. Freddy learns from this data; if it is poor, every feature downstream inherits the problem.
Step 1: Freddy Copilot for agents
Start here because it has the lowest risk and the fastest adoption. Ticket summarisation, suggested responses, tone rewriting and similar-ticket suggestions make agents faster without changing what customers see. Measure handling time and first-contact resolution before and after, and collect agent feedback for the first month.
Step 2: Freddy AI Agent on a narrow scope
Deploy the customer- or employee-facing agent on a defined set of intents - password resets, order status, leave balance, access requests - where the knowledge base and integrations are solid. Review every conversation for the first two weeks, tune the articles it draws on, and expand the scope intent by intent. A bot that answers ten questions well beats one that answers a hundred badly.
Step 3: Freddy for service requests and approvals
In Freshservice, let the AI agent raise catalogue items and route approvals through Teams or Slack. This is where employee deflection turns into fulfilment automation, but it only works once the service catalogue is well designed and approvals are configured.
Step 4: Freddy Insights for leaders
Once categorisation is clean and a few months of data exist, Insights' natural-language reporting and anomaly detection become genuinely useful. Before that, it will simply surface the data-quality problems you already suspected.
Governance from day one
- Define which data Freddy can access and confirm it meets your data residency and privacy requirements.
- Decide when the AI agent must hand off to a human, and make the handoff seamless.
- Review bot conversations weekly in the first quarter and keep a tuning backlog.
- Report deflection, containment and customer satisfaction, not just conversations handled.
Freddy AI rewards organisations that did the unglamorous work first: categorisation, knowledge and catalogue design. Do that, roll out in order, and the AI delivers. Skip it, and you have bought a very articulate way to expose your data problems.
Workdeft configures Freddy Copilot, AI Agent and Insights as part of Freshservice and Freshdesk implementations across the GCC. If you have licences switched on and results switched off, we can help you sequence it.
