Having AI in your revenue stack no longer sets you apart. Almost every team has it. What separates them now is how far they have come, from simply recording conversations to having trusted systems act on them.
In a new piece for Forbes, our co-founder and CTO Devang Agrawal sets out the four stages of that journey. He also explains why the teams that try to skip ahead usually find themselves starting over.
Stage 1: Record
Everything starts with a complete, consistent base. That means capturing conversations across every channel and, just as importantly, agreeing as a team on what key terms mean. If two managers define “risk” differently, no amount of AI will reconcile them later. You have finished this stage when every deal leaves a useful record behind it.
Stage 2: Understand
Data only matters when it changes a decision. Dashboards alone will not get you there. The teams that move through this stage build a weekly rhythm in which insights turn into named actions with owners and deadlines, and reps read their own call analysis rather than waiting for it to be read to them. Devang’s test for this stage: can you name a decision the analysis changed this week?
Stage 3: Run
Now you automate, carefully. Pick one or two clearly defined plays, such as CRM updates or next step creation, and write the process out in plain language first. Reps have to be able to see where the system’s information came from before they will trust it. Managers stop chasing CRM hygiene and start handling exceptions.
Stage 4: Compound
At this stage, supervised agents take on whole chunks of work: account research, deal prep, follow-up. People still review and approve. Start with low-risk internal tasks, define where human approval is needed, and feed every correction back in. This is the stage where the system gets better over time, and it has no finish line.
The thread running through all four
Technology and people have to move together. A capable agent nobody trusts stalls. A motivated team with tools that do not fit stalls too. Maturity comes from moving both forward at the same pace.
The same pattern shows up in our 2026 GTM Maturity Report, which finds teams most often stall where operating a system turns into supervising one.
Where does your team stand?
Benchmark yourself with our GTM maturity model. The four-question self-assessment places your team on the curve in about a minute.
Read the full article in Forbes: You can’t skip your journey to AI maturity in sales. Here’s how to earn it, stage by stage.
Frequently asked questions
What are the four stages of AI maturity in sales?
Devang Agrawal names them Record, Understand, Run, and Compound. Record captures every conversation and agrees on shared definitions. Understand turns analysis into named actions with owners. Run automates one or two clearly defined plays. Compound hands whole chunks of work to supervised agents, with people reviewing and approving.
Why do AI rollouts in sales stall?
Most stall because a team tries to skip a stage, or moves technology and people at different speeds. A capable agent nobody trusts stalls, and so does a motivated team with tools that do not fit. Agrawal's argument is that each stage has to be finished before the next one can hold.
How do I know which stage of AI maturity my team is at?
Agrawal gives a test for each stage. You have finished Record when every deal leaves a useful record behind it. You are through Understand when you can name a decision the analysis changed this week. For a scored read, the GTM maturity self-assessment on goairspeed.com places your team in about a minute.
