From building to buying: how Merchant Growth got an AI agent that works

The challenge: building an AI agent in-house

Merchant Growth, an alternative small business lender based in Canada, partnered with Superpanel to build Sam, a digital teammate responsible for re-engaging with past clients by phone, text, and email. But Merchant Growth didn't start with Superpanel. They first set out to build their own AI agent in-house.

When Hashem "Hash" Aboulhosn joined to lead growth, he inherited a company-wide priority: an outbound AI sales agent, built in-house on Vapi. Its job was to reconnect with past clients who might need financing again, a group the sales team rarely had time for while working new leads.

The prototype came together quickly, but live calls exposed what the demo hid: lag, robotic pacing, and an agent that drifted off script. Making it useful meant connecting it to Salesforce, scheduling follow-ups, and building a decision engine that could hand client conversations between text, email, and voice without losing the thread. 

The pivot: focus on what you do best, partner for the rest

After one or two test cohorts, it was clear that the in-house build was moving too slowly on work that would never be a core competency. Changing course meant walking away from three to four months of work on a company-wide priority, but Hash's reasoning was simple: Merchant Growth's edge is lending, not voice infrastructure. No company should build its entire software ecosystem in-house. The faster path is to focus on what you do best and partner with teams who are the best at the rest.

Why Superpanel?

1. Performance: an AI agent that sounds human

Hash had seen what out-of-the-box voice platforms could do, and where they fell short.

Superpanel's calls sounded different, and Hash heard it right away. He credits the investment behind them. Superpanel built Voco, its own proprietary voice orchestration layer, after hitting the same limits on off-the-shelf platforms like Vapi and Bland. It knows when a caller is still thinking, handles interruptions without talking over anyone, and keeps latency low enough that the call feels like a conversation.

2. Control: configuring the experience instead of prompting it

What Hash valued most was control over the client experience. With the in-house build, the prompt was the only lever, and every fix meant another warning stacked on top of the last one.

Superpanel gave him a different kind of control. Instead of steering the AI with prompts, his team was able to configure the rules in code, using Superpanel's platform: which questions get asked, when a client goes to a rep, and what's written to Salesforce. The AI handles the conversation and the judgment calls within those rules. That made a multi-step lending flow predictable, even when it ran over days or weeks.

3. Speed: a fast test with aligned incentives

Superpanel gave Merchant Growth an easy way to test a new approach to serving its clients. From the start, Superpanel's implementation team was clear about the work involved: the configuration Merchant Growth's sales flows would need, how the team would build and test it, and a cohort-based rollout with a clear decision point along the way.

The pricing matched that approach. A meaningful share of what Merchant Growth pays is tied to loans funded, so Superpanel succeeds only when Merchant Growth does. A fast timeline and an outcomes-based pricing model made the decision easy internally.

What changed for Merchant Growth

Merchant Growth started with one use case: re-engaging past clients the sales team rarely had time for. Along the way, it found a better way to serve their client: 60% of clients who reengaged responded by text only, a channel Merchant Growth had used before but not consistently, since reps defaulted to calls. Now clients engage when and how they prefer, from sending bank statements to checking on their loan via text or call.

The first cohorts also showed how much demand was already sitting in the client base.

Cohort size grew 400% from the first cohort to the third, and Merchant Growth plans to scale more than 16-fold next quarter. Merchant Growth is now deploying Superpanel across more of the client journey. The payoff: Superpanel handles outreach, follow-up, and document collection, so reps can focus on building rapport, matching clients to the right financing, and winning their commitment.

Advice for growth leaders weighing build vs. buy

Hash's view: for most businesses, buying makes more sense. Prototypes are cheap now, but a prototype isn't the system you'll run your sales on. Here's his advice for leaders making the call.

1. Spend your team's time on what you do best. Building a voice agent is easy to start and hard to finish. The hard part is the plumbing, and that's rarely a lender's or a law firm's edge.

2. Judge a pivot by speed of progress, not sunk cost. Walking away from months of work feels expensive. Staying on a slow path usually costs more. A short timeline and pricing tied to your results make it cheap to find out.

3. Set up a weekly working group. Put sales, the team that owns your CRM and operations, and your vendor in the same room every week. Each group does its part, and progress comes fast.

4. Give the AI the work your reps shouldn't be doing. Let the agent handle prospecting, follow-up, and document collection. Keep your best people on rapport, matching clients to the right product, and winning commitment. 

Two next steps

Discuss your use case: Thinking about building your own AI agent?
Talk to our team before you commit the engineering time.

Read the AI agent evaluation guide: Evaluating AI agents for your sales team? Our evaluation guide covers the questions that separate a good demo from a system you can run in production: decision logic, voice quality, integration, deployment, and escalation.