Bonusly’s senior revenue architect on AI in sales

Revenue Brew spoke with Bonusly’s senior revenue architect Amani Phipps via email about using AI in the sales org ahead of our latest IRL event.

These responses have been edited for length and clarity.

What’s the biggest misconception sellers have about AI?

The biggest misconception is that AI is already better than it actually is, and that it’s good at basically everything. That shows up as two failure modes I see constantly in change management, when I’m trying to get people across an organization to actually adopt a tool instead of just poking at it.

The dangerous one is trusting it too much and turning your brain off. AI often has poor taste and judgment. It’s weak on strategy, because strategy means weighing a pile of variables at once and knowing which ones matter right now, in your specific situation, while that situation keeps shifting under you. It’s strong on deterministic work: clear input, clear output, one right way to get there.

The other failure mode is treating it like a finished, all-knowing product instead of what it actually is: a general-purpose tool. Ask Claude, GPT, Perplexity, whatever model you want, to draft an email, and it’ll draft you an email. Ask it to draft that email with real taste, in your voice, with your read on what actually matters in that specific deal, and it can’t, not until you’ve taught it who you are and how your organization thinks. Swap the model and nothing changes. The context is what changes the output.

Where do you see the most resistance when teams try to adopt AI into their revenue workflows, and how do you move past it?

The resistance breaks into a few areas, but two have come up consistently: trust and change management. Trust is the bigger one day to day. Almost everyone I talk to has felt AI’s presence at work by now, whether that shows up as excitement, dread, amplification, or confusion in work they’re already doing.

But trust breaks down for specific reasons. If your data isn’t clean or consistent across teams, people start pulling different numbers for the same question, and that alone kills confidence in anything built on top of it. If the model you’re running on has a bad day and an analysis comes back degraded, that erodes trust too, and it doesn’t take much for that to stick.

And if people don’t actually know what their own real numbers look like, unfamiliar figures feel suspicious rather than accurate. We’ve had people inside our own company who didn’t know our real reward transaction volume or revenue, so when a number came back that didn’t match what they knew, the instinct in some cases was to distrust the tool. That’s not really a technology problem. That’s an education problem.

The other resistance is the fear of “different,” and I don’t think that fear is irrational. I’ve talked to plenty of operators genuinely wondering what happens to their role, their team, their department. Some organizations have already started replacing people with agents. Others are chasing the 10x or 50x employee and think they see a real path there. Others hand down “figure out AI” as a mandate with no clarity on what job it’s actually meant to solve, and people end up doing worse work because they’re feeding everything into a model without knowing what problem they’re solving.

Moving past both comes down to getting specific. Start with the real job to be done, not the technology. Pick something small in your individual life/workstream: a daily briefing agent that reads your meetings, your emails, and your Slack, and hands you a clean readout each morning; it’s an easy first step that gives someone a sense of control instead of a sense of threat.

If you had to pick one AI use case that’s delivered the most measurable impact on your team’s performance, what would it be and why?

The single highest-impact use case we’ve built is our internal business intelligence tools, BonuslyGPT and SignalForge.

It started as BonuslyGPT: a Slack app wired through MCP into OpenAI’s models, able to query Snowflake and search our own codebase. It worked well enough that we rebuilt it into something bigger with a broader orchestration platform running through Claude, that we now call SignalForge.

The reason it matters isn’t just the model underneath; it’s the philosophy behind it. Our goal was to build an AI-adopted culture, and these tools helped solidify that. We never wanted a handful of AI specialists supporting the rest of the company. We wanted an organization where every employee is good at using AI in their own job.

That range is the point. Someone building a board deck can pull everything relevant to it in minutes. An engineer chasing a bug can trace who’s affected, how many others might be hit, and what code needs to change. A seller can see how a prospect stacks up against similar existing customers, how long they’ve stayed, how they engage, how they’ve adopted the product. We’re at over 14,000 uses from about 70 people over the past year, and that number almost certainly undercounts it.

The downstream effects are what continues to convince me we made a good bet. Designers push their own [pull requests] now and get them reviewed. Engineers ship noticeably more code, and they’re spending more of their judgment on what should get built and why, not less, because the tool bought them the time to ask more questions. Our ops team functions like a product team shipping code ourselves, deploying agents and other workflows to help accelerate teams.

How do you think about the balance between AI-driven automation and maintaining authentic, human-led relationships with prospects and customers?

A human in the loop is still incredibly valuable. People do business with people, and that doesn’t change because the tooling got better. We’re a group of people building a product together, and that product happens to be our company. Every deal we do is with another group of people at another company. That human element isn’t going anywhere; as a matter of fact, I believe it gets even more important.

We build a recognition and rewards platform, so this matters twice over for us. Our internal line is AI first, human always. We use AI to amplify the human moments in the work, not replace them.

The actual discipline isn’t a blanket rule. It’s figuring out, workflow by workflow, where a model can run and where a person has to stay in the loop, and that call comes down to risk. I’ve run well over 1000 analyses through AI in the past year, and I still don’t fully trust a model to run one unsupervised.

What I do trust is AI getting me most of the way there. Sometimes that’s 90% of the work, sometimes 80, sometimes less, but there’s always a piece left that’s mine. I treat trust in these tools as something you build in stages, not something you grant on day one.

What does a revenue team that’s truly AI ready look like?

A revenue team that’s truly AI ready is grounded in four things: data, process, people, and strategy.

Data comes first because bad data in is bad out, every time. If your numbers aren’t clean, if they’re not consistent across every workstream, if two people can’t ask the same question and get the same answer, nothing you build on top of that will hold up.

Process is second. Do you actually understand your funnel and your business, and have you cut out the complexity that doesn’t need to be there? That clarity does two things: It lets you train a model properly, and it tells you exactly where specific work should happen in the first place.

People are third, and this is less about specialist versus generalist than people assume. Both still win in different ways in the AI era. What actually matters is whether you’re hiring people who think well and whether they have the agency to want to use the tool at all. Building that agency takes deliberate space…People need to learn from each other and see the context behind decisions, not just receive a mandate.

Strategy is fourth…If you’re not clear on where the business is going, your people will use AI to go in a dozen different directions, simply because they can. Give a doer access to a powerful tool and they will produce something. Whether it’s the right something depends entirely on whether you’ve oriented them toward an outcome you already know matters. That alignment is the whole game.