A sales team's reputation rests on what each rep says and sends. If an AI tool puts a wrong figure in a customer email, the rep wears the cost. That makes caution a reasonable starting point.
Understand what worries the team
Ask the team what concerns them before you choose a tool. The answers usually fall into a few groups.
- Mistakes sent under their name. A wrong price or a made-up detail in a customer email damages the rep's relationship.
- Being watched. People worry that call summaries and activity data will be used to rate them.
- Customer data going somewhere unknown. Reps know customers share private details in calls and emails.
- Job security. Some people will wonder whether the tool is a first step towards smaller teams.
- Extra work. Checking a poor draft can take longer than writing it from scratch.
Write down each concern and the answer you give. Share the list with the team, and update it as the pilot runs.
Start with tasks that stay inside the business
Choose first tasks where a mistake is easy to catch and no customer sees the output directly. Good first tasks save time on admin that reps already dislike.
- Summarising call notes into the CRM, as a draft the rep checks.
- Drafting a first version of a follow-up email for the rep to edit.
- Preparing account research from public sources before a meeting.
- Turning a long email thread into a list of open actions.
Leave higher-risk tasks until the team has experience and agreed rules. These include sending messages without review, setting prices or discounts, and scoring individual reps.
Set rules for data and privacy
Decide what may be sent to an AI tool before anyone uses one on real work. Keep legal requirements and your own policy apart, so people know which rules can change.
In Australia, the Privacy Act 1988 and the Australian Privacy Principles apply to many businesses. Ask your privacy officer or legal adviser which obligations apply to yours. The Office of the Australian Information Commissioner (OAIC) publishes guidance on privacy and the use of commercially available AI products.
As best practice, the OAIC recommends not entering personal information into publicly available AI tools, particularly sensitive information. It also says due diligence should not be "set and forget".
A signed agreement with the vendor is a starting point for your privacy assessment. Before you approve a tool, check four things with the vendor. Find out where data is stored and processed. Ask whether your inputs are used to train the vendor's models. Check how long data is kept. Confirm who in your business can see other users' prompts and outputs.
Repeat the checks when the vendor changes its terms or you use the tool for a new purpose.
Then agree a data table with your privacy officer or legal adviser. The table below is an example of organisational policy. Where a row reflects a legal requirement, note the requirement beside it.
| Type of information | May it go into an AI tool? | Conditions |
|---|---|---|
| Public information, and your own approved product and pricing material | Yes | A tool approved for this use, with the permitted information and purpose documented |
| Customer names and business contact details | Yes, with limits | Approved business tools, after a privacy assessment for this use |
| Call recordings and transcripts | Yes, with limits | Customer told, consent obtained where the law requires it, and state and territory laws checked |
| Sensitive information, such as health details | Not in this pilot | Exclude it for this initial pilot. Any later use needs a separate privacy and security assessment. |
| Passwords, access keys and other credentials | No | Never enter them into any AI tool |
| Confidential contract terms | Under their own rules | Follow the contract's confidentiality terms and your policy for that material. If unsure, leave it out. |
| Any customer information in a personal AI account | No | Use business accounts the company controls |
Recording laws differ between Australian states and territories. Check the rules where your team and your customers are before you record calls.
Check research and summaries before you rely on them
- Account research: ask the tool to list the sources behind each fact. Open each source and confirm the company, the date and the detail before you use it.
- CRM summaries: save each summary as a draft until the rep who held the call verifies it. A wrong date or action in the CRM becomes the version colleagues rely on.
- Instructions inside content: web pages and emails can contain text written as instructions to an AI tool. That text must not change the task you set, or make the tool send, delete or update anything.
Choose tools that ask before taking any action. Stop the task and report it if a tool does something you did not ask for.
Keep a person approving what customers see
For this initial pilot, make one rule clear from the start. Anything a customer sees is read, approved and sent by a person. Any later automation needs its own review before this rule changes.
The approver checks the output against the source, such as the call notes or the price list. A short checklist helps reviewers work quickly.
- Are all figures, dates and prices correct?
- Are names, companies and job titles correct?
- Does it promise anything the business has not agreed?
- Does it include anything the customer did not tell you?
- Does it sound like the rep who will send it?
Example (illustrative): A team of eight account managers tried AI drafts for meeting follow-up emails. Before the trial, each follow-up took about 15 minutes to write. With a draft built from the call notes, review and editing took about 6 minutes.
Each account manager held about 10 customer meetings a week. That saved each person about 90 minutes a week.
In the first two weeks, reviewers caught 4 drafts that listed an action the customer had not agreed to. The team added a line about actions to the review checklist.
Run a pilot with volunteers
Start small, with people who want to try it. Here is an example four-week schedule.
- Week 1: Pick four to six volunteers and one task. Get the data rules signed off. Measure how long the task takes today.
- Week 2: Use the tool on real work. For each output, log whether it was used as is, edited or thrown away.
- Week 3: Review the log together. Update the rules and the review checklist. Share what went wrong as openly as what went well.
- Week 4: Decide whether to extend, change or stop. Share the results and the numbers with the wider team.
Use the template below to write the usage rules for the pilot. Keep it to one page, so people read it.
| Task | Allowed? | Who approves the output | Notes |
|---|---|---|---|
| Summarise call notes into the CRM | Yes | The rep who held the call | Saved as a draft until the rep checks actions and dates against their own notes |
| Draft a follow-up email | Yes | The rep who sends it | Use the review checklist before sending |
| Research an account before a meeting | Yes | The rep | Public sources only. Open each source and confirm facts before you rely on them. |
| Send messages to customers without review | No | Not applicable | Not allowed in this pilot. Any later change needs its own review. |
| Rate or rank individual reps | No | Not applicable | Written into the rules for the whole team |
Talk openly about jobs and monitoring
Say plainly what the AI tool is for. If it is there to cut admin, say so and show the hours saved. If you have not decided how the time will be used, say that too.
Put the limits on monitoring in writing. For example, state that AI call summaries will not be used to rate individual performance. Keep to what you write, and tell the team before any rule changes.