Where should a B2B company start with AI: sales, quotes or customer service?
Start in the process that hurts the most and that can be measured. It should recur often, take a lot of time every week and clearly affect the business. The input should be available, the regulatory risk manageable and the team willing to work in a new way. For many B2B companies that means quotes, leads or follow-up.
Why does the first process matter so much?
Because the first solution proves more than its own impact. It shows the organisation what it feels like when AI support works day to day.
Choose well and trust grows internally. Colleagues see something that works, instead of hearing about something being planned. Then the next process is easier to get going.
Choose something too big, too sensitive or too infrequent and it becomes a pilot nobody remembers. Read more in the guide How do you go from AI pilot to AI in production?
Which six questions decide where you should start?
These are the same questions our Process Prioritiser asks. Answer them for each process you are considering.
| Question | What you are looking for | Signs to wait |
|---|---|---|
| 1. How often? | Work that recurs several times a week or daily | Something that happens once a month |
| 2. How much time? | Hours every week, ideally spread across several people | A few minutes here and there |
| 3. What impact? | A clear effect on revenue, response time or customer experience | Hard to say what gets better |
| 4. Is the input available? | Data and documents you can get at without a big project | The information only exists in someone's head |
| 5. How sensitive? | Manageable risk, where a person can review before anything goes out | Decisions about individuals, sensitive personal data |
| 6. Is the team willing? | People who say “finally” | A team already busy with another change |
The process with the most strong answers is your first candidate. One with high impact but also high sensitivity is not wrong, it just needs more preparation: clear ownership, control over where data is stored and a human in the loop.
How do the most common processes compare?
No process is right for everyone. But the patterns are familiar.
| Process | Typical AI support | Why it is often a good start | What needs solving |
|---|---|---|---|
| Quotes | Reads the request, finds similar deals, writes a first draft | Recurs often, takes a lot of time, impact shows directly in lead time | Prices and templates need to be in place. Read the guide to quote automation. |
| Leads and prospecting | Watches buying signals, prioritises which companies are worth a call right now | Affects revenue directly, easy to measure in meetings and deals | Salespeople need to follow up on what the AI support finds |
| Follow-up after meetings | Structures notes, suggests next steps, drafts emails | The least change to how people work, a quick win | Small each time, but it adds up |
| Customer service | Suggests answers to recurring questions, sorts cases | High volume, clear patterns | The customer meets the AI support directly: higher demands on transparency and quality |
| Reporting | Compiles input and writes a first version | Hours every week, often the same structure | Data often sits in several systems |
| Internal administration | Sorts, summarises, fills in | Relieves many people | The impact is harder to link to the business |
Which way in is the least daunting?
Follow-up after meetings. It asks the least of everyone.
I do it myself after trade fairs. As soon as a conversation ends, I photograph the business card and record thirty seconds on my phone. Not polished. Just what I need: what they are struggling with, what we agreed, how they sounded.
The AI support does the rest the same evening. A structured note, a clear next step with a date and a first draft of a follow-up email that sounds like me. I read it through, change a sentence or two and send.
The effect is that the follow-up arrives within a day, while the meeting is still fresh in the other person's mind too.
Nobody changes how they sell. Nobody has to learn a system. You talk into a phone, just as you already do with voice messages.
How do you make the first step easy?
- Pick one process, not twelve. Small enough to test in a few weeks, real enough to show in the numbers.
- Give it a named owner. A person who has the process on their list at eight on Monday morning.
- Measure the baseline first. Time, volume or response time, before anything changes.
- Start where a person already reviews. That keeps the regulatory risk manageable from the start.
- Expand when it sticks. The next process is chosen in order of what costs the most.
That is the method we call Prove & expand. Read more in the guide What is AI transformation?, in our service ladder and among solutions we have built.
FAQ
Common questions about where to start
Should we start in sales or in customer service?
Where the answers to the six questions are strongest. Sales and quotes often give the fastest measurable business impact. Customer service has high volume, but since the customer meets the AI support directly, quality and transparency need to be in place from the start.
Do we need an AI strategy before choosing a process?
You need a prioritisation. A 90-minute use-case workshop with leadership gives you one, with a concrete action plan.
Can we start in several processes at once?
You can, but the first process should prove the impact. One at a time makes it clear what produced the result, and gives the organisation time to learn the way of working.
What if the process that hurts most is also the most sensitive?
Then it is still the right one, but it needs preparation: clear ownership, mapped data flows and a person who reviews. Sometimes it is wise to start in a simpler process first and build trust.
How do we get an answer for our company?
Not sure where to start?
Bring your processes to a conversation. Together we will find the one that pays off first.
Book 30 minutes