We launched Gigi, the AI media manager for the Amazon DSP, in July 2025. In a little over six months, we’ve increased our $ of managed ad spend and # of advertisers by +700%. In November, we were thrilled to win Amazon’s Global Technology Innovation partner award. While it is still early, we are showing signs of early product market fit. Most importantly, we’ve learned a ton and thought it would be helpful to share some of our GTM learnings more broadly for those building vertical AI agents across similar customers segments (services business—our customers are primarily media agencies) and similar business models (usage based pricing—we charge a % of managed advertising spend).

First, what do we do? Gigi is a vertical AI agent for enterprise media management and measurement, our first product is for the Amazon DSP. Our small team of 14 has +25 years of retail media experience. We have leveraged this deep domain expertise to allow our customers to automate workflows in their day with agentic AI.

To do this, we’ve created a robust repository of RAG documents to train Gigi, have exposed a collection of Amazon’s API and metrics as tools that Gigi can call, and allow our customers to customize their Gigi agent unique to their own bespoke ways of working. Our promise to our customers is that Gigi will improve advertising performance, positively impact their operating model, and enhance the way people work by offloading mundane and low level tasks to Gigi.

Here’s what we learned:

Sales

People want to buy AI. Gigi was founded in 2023 as a legacy SaaS ad tech company for CTV ads. We were a point solution in a mature market. Few of our prospects wanted to buy a new point solution and (post COVID) were content to consolidate tech vendors. AI has fundamentally altered the desire to onboard new software at enterprises. There is a clear appetite to “buy AI” if the AI can provide demonstrable value. When selling AI, it’s easier to get the meeting, and sales cycles are faster. Doors are opening for new entrants and mature categories are up for grabs.

Making sense of “all the AI” is part of the sales process. On a sales call, one of our prospects, asked “can you help us make sense of all these AI agents?” This agency advertising leader had been inundated with pitches for AI agents and genuinely needed help interpreting where certain AI agents would be competitive or complementary.

My guidance to her was to think of a new AI agent as a job at your agency (i.e. analyst or media manager). Can that AI agent perform many of the lower level functions of that specific job? If so, then it can likely add value. If not, then the value is unclear. Additionally, like team members, there will likely be overlap between the skills of vertical AI agents. This isn’t bad, and is similar to how humans overlap and divide skills. In fact, this overlap could likely lead to future agent to agent collaboration.

Job postings are ideal intelligence for lead gen. If you are able to successfully align your vertical AI agent with a specific job in the work force then anytime a prospect in your ICP is hiring for that job they are a prime candidate for outbound outreach. Bonus points if there are several job postings for that role. To date, our strongest cold outbound motion has started with job boards. We look at ICP companies’ open roles, map each role’s day-to-day tasks to what our agent automates, and then counduct outbound outreach to the person most likely to own that hire. This tactic can be broadly applied across any vertical AI agent.

More than ever, conviction and alignment from executive leadership and operational leadership is required. Whether there is a top down directive to “buy AI” or bottoms up initiative to implement AI across a function, the forthcoming change management required to successfully scale and implement AI needs harmony between executive and operational leadership. Executives at agencies are now starting to adopt a Tobi-style directive to encourage operational leaders to “demonstrate why they cannot get what they want done using AI” before asking for headcount. Similarly, operational leaders are lobbying executives to introduce new AI products to dramatically enhance the way their teams work. Successfully selling a vertical AI agent into an enterprise requires both executive sponsorship and operational commitment. A pilot can sometimes start with only one. But, enterprise wide adoption during the pilot and approval of ongoing spend afterward requires both.

Pricing and Terms

To refresh: As a AI agent for advertising, we operate on a usage based pricing model tied to managed ad spend. For most of our customers, we are not unseating an incumbent and we are bringing on a new cost that needs to be rationalized by the value that Gigi provides.

We encourage longer pilot terms to accommodate the change management of rearchitecting org structures in an AI-first manner. We begin working with new customers on pilots for fixed terms. Typical pilots in legacy SaaS tend to be 1-2 months. We’ve found this is not enough time to successfully scale a vertical AI agent and now actively encourage our customers to engage in longer pilot terms, usually 4 months.

Once Gigi begins to demonstrate value, the pilot no longer becomes about testing the technology. The internal work of our customers flips to undergoing the change management required to transform manual workflows and rearchitect the function to be AI-first. Change management takes time and longer pilots allow our customers the time required to profitability implement Gigi by making personnel decisions like not backfilling for attrition or repurposes team members to other roles in the company.

We eliminate as much pricing friction as we can in our pilots to gain maximum adoption and impact. Like most vertical AI agents, our product becomes increasingly valuable with scale and context. If an agency manages 30 clients, it’s impossible for Gigi to prove demonstrable value if that agency only onboards 1-2 of the 30 clients to Gigi.

In fact, with limited scope, a vertical AI agent can provide negative value by simply introducing unhelpful operational cost. Usage base pricing may pose initial friction if you’re trying to optimize for short term impact and value creation. To remove this friction, we’ve been offering discounted pilots on flat monthly fees with unlimited usage. Since doing this, we’ve seen an increase in our customers’ desire to test Gigi at scale and has led to more immediate wins in our pilots.

Vertical AI agent implementation only works if there’s a near term path to $100k ACV (minimum). Selling and deploying a vertical AI agent requires significant short term account managements resources (more on this below). It requires managing multiple stakeholders: execs, operational leadership, and hands on practitioners through the sales process. The resources required pre and post sale for a successful vertical AI agent implementation only make sense if there is a near term path to a minimum $100k ACV with a customer. You need not start with $100k ACV, there can be land and expand potential. But, it is simply not worth your time to try and acquire SMBs at $10k-$25k ACV at full expansion because it will be incredibly challenging to staff your team profitably given the amount of work required to successfully implement the AI agent.

Everyone underestimates the 2nd order cost savings of a smaller team. When making value assessments on the price of a vertical AI agent, customers only price the role and job you’re AI agent performs. No customers value the cost savings across talent acquisition, HR, enablement, ops, and middle management. Even though they clearly exist, it is incredibly challenging to quantify the cost savings of these 2nd order benefits and I wouldn’t even try. You need to demonstrate value beyond the immediate opex efficiencies your AI agent brings, such as potential increases in revenue by new workflows with AI and/or being a force multiplier in productivity for the best team members.

Implementations and Account Management

Since Gigi’s launch, we’ve onboarded dozens of new customers. Perhaps our greatest learning over the past six months is the typical onboarding playbook of a legacy SaaS product does not work with vertical AI agents. Here’s how we’re adapting that playbook.

Enterprise AI agents don’t require forward-deployed engineers as a rule, but they do require hands-on account management. If you’re building a vertical AI agent you do not need to hire a team of forward deployed engineers to successfully scale your agent in the enterprise. While this may be both fruitful and profitable on customers with >$1m ACV potential, they are not required.

It is likely, though, that your AI agent is the first one your customers have deployed. On account of that, it’s unlikely they know the ways to operate your AI agent to quickly demonstrate maximum value. Your account management team needs to take a hands on role at building sets of prompts and workflows in your product for your customers. This requires a heavy up front lift for every new customer onboarded.

We’ve found it invaluable to sit side-by-side with our customers and monitor how they’d manually manage advertising campaigns and then replicate building those workflows agentically with Gigi. In our pilots with customers, we strongly encourage on-site workshops for maximum and immediate impact. Once we took the position that we needed to do the upfront work—and not our customers—we began to see much higher customer engagement and value demonstrated.

Prompting is a skill. It’s your problem if your customers are bad at prompting. Note: most will be. Most customers approach AI agent deployment with the wrong mentality. They’re used to general-purpose tools like ChatGPT, where vague questions still produce acceptable answers. AI Agents in the enterprise do not work that way. Most vertical AI agents are powered with robust tool calling capabilities and access to proprietary customer data. The tradeoff to this power is the agent requires clear intent and specificity to deliver value.

Anthropic’s latest research makes this empirically true. They found a near-perfect correlation (r>0.92) between the sophistication of human prompts and the sophistication of AI outputs. Put simply, models calibrate to the user. Ambiguous inputs will produce unhelpful results.

We encourage customers to think of their AI agent like a highly skilled but junior team member. This team member may be highly capable, but without clear instructions, the output will suffer. This is not a model failure; it’s an input problem.

Because prompting doesn’t come naturally to most customers, we’ve made it a core part of our onboarding. We run prompting workshops using Duolingo-style language learning principles and build custom prompt libraries tailored to customer’s workflows. Early on, we also built saved prompts into our product so customers can reuse prompts that produce high-quality outputs. Treating prompting as a skill to be taught, not a prerequisite, has been essential to enable our customers to extract value from Gigi.

Every customer underestimates the time required to train and manage a vertical AI agent. When we say Gigi is your next media manager, we mean it literally.

Implementing AI Agents in the enterprise is closer to hiring an employee than buying software. Like any new hire, it doesn’t show up on day one knowing how you work, and most customers underestimate the time required to get it there.

Every successful pilot we’ve seen has had a clearly defined owner and that owner allocates an allotment of their week dedicated to training Gigi. The same principles apply that a functional leader applies to new hires: onboarding, training, validation of work, and ongoing management. If you have a new hire without a manager singularly responsible for their success, that new hire is going to fail. Without a clear owner singularly responsible for the success of your AI agent, your AI implementation will also fail.

While implementing an AI agent will lead to increased efficiency and productivity for an enterprises’ best team members, the ongoing training and management of that agent will not lead to a reduction in work for functional leaders. In fact, at the outset, they might have more work on their plate.

Deploying agents is hard. There is still a relatively narrow universe of people with this skill. Rather than perceiving this as a burden, we’ve started reframing this to our customers: training and managing an AI agent will be one of the most profoundly valuable skills of the next decade. Being the person that helps their enterprise understand and deploy AI agents is one of the biggest opportunities available to anyone looking to rise within an organization. And early adopters have a material opportunity to build that muscle now. To acknowledge this and generate excitement with our Gigi owners, we’re now sending each owner a hand written card congratulating them on the opportunity to build these new skills.

Adapting with these lessons has allowed us to stack small wins. We know these lessons may not apply to every vertical, but many of them—in principle—can be broadly applied to most verticals. We’re excited to share more learnings along our journey.

Note: this blog first appeared as part of our weekly Cherry Picked newsletter. You can subscribe here 🍒