Agents Sign More Contracts, So Why Are Profits Getting Thinner?

Deep thoughts on AI and aspirations —— ByteDance Deep Thinking Circle

A wave of Agent companies is experiencing a strange kind of prosperity: more contracts signed, revenue curves steep enough to print directly on fundraising decks, then year-end accounting shows negative profit. What’s worse, most founders can’t pinpoint exactly where the losses are. This isn’t because they’re incompetent—this business, from pricing to costs, hasn’t yet developed a complete accounting framework.

First, See Which Tier You’re Charging At

Agent pricing models lay out on a four-tier staircase.

Most companies start at tier one. The problem is treating the starting point as the destination. Stop at per-action pricing, and your quote becomes a price list waiting to be copied: next month a competitor with similar features cuts prices in half, and clients have no reason not to switch.

The higher up the tiers, the harder to compare prices. Charge per outcome, and clients compare “what this outcome is actually worth”; charge per headcount, and they compare “what it costs to employ someone.” Neither number has public pricing to reference, and your pricing latitude opens right up.

Consider a real comparison. A fully equipped sales development representative costs seventy to ninety thousand dollars annually in salary plus overhead. Some companies simply charge per Agent—twenty thousand dollars each, doing the same work, with bonuses for hitting appointment targets. Clients calculate immediately: this is a headcount line item. The headcount budget pool is an order of magnitude larger than the software budget, yet many Agent companies squeeze into the software budget fighting for scraps when they’re actually standing at the wrong counter.

My take: these four tiers truly measure only one thing—client trust in you. Per-action pricing means clients trust “this single call is worth this price”; per-outcome pricing means clients trust “you can deliver consistently.” So pricing upgrades are never as simple as changing a price sheet; they’re monetizing accumulated trust. Jump tiers before trust is sufficient, and clients vote with their feet; accumulate enough trust but don’t jump, and you’re essentially giving clients free money.

To Charge Per Outcome, Learn to Charge Bonuses First

Many people understand outcome-based pricing as a binary choice: either charge subscription fees as a tool, or charge entirely per outcome. This understanding either scares people off or ruins them. There’s a transitional design called outcome bonuses: charge the base price for the work as usual, then charge an additional fee when results meet agreed quality standards.

The real value of this design isn’t the bonus itself—it’s the negotiation process. To negotiate bonuses, you must align with the client on three things: what constitutes a good outcome, who determines it, and what criteria they use. Once this conversation succeeds, the contract transforms from standard pricing into a customized contract, and your relationship shifts from vendor to something like a co-stakeholder. Customized contracts are extremely difficult to replace, because competitors trying to poach this client must restart this three-question conversation from scratch.

When selecting work, this logic also functions as a filter. In insurance, some companies specialize in policy renewals; in freight, some companies deploy two thousand Agents calling truck drivers to negotiate rates, arrange transport, and follow through until delivery. These jobs share a common trait: results are clearly quantifiable, the price ceiling is the original labor cost, and clients calculate the value effortlessly. Work with countable outcomes deserves outcome-based pricing; work where outcomes can only be determined by client subjective feeling, like “improved experience,” shouldn’t touch outcome pricing yet—go back and build measurement capabilities first.

Another phenomenon worth noting: many outcome-based contracts nowadays are proposed by clients first. Everyone’s uncertain about AI, so clients want outcome clauses to push risk back onto vendors—purely self-protective behavior. Don’t treat outcome pricing as some offensive weapon; first accept it as a defensive tool, then embed upgrade opportunities in the contract. Also beware of those early per-transaction fees that look like revenue but function more like trial samples. Clients haven’t seriously evaluated value; the real pricing negotiation happens at renewal, when many companies discover for the first time that the price they charged and the cost they delivered never actually reconciled.

The Cost Black Box: Money-Losing Clients Mixed With Profitable Ones

Pricing is only one side of the ledger; the other side is more hidden.

Agent companies universally keep fuzzy books. On the surface, costs are just model call fees. In reality, for a multimodal Agent that makes calls and executes operations, the cost heavyweights are often call duration, dialing frequency, third-party APIs, and speech synthesis—accumulating per minute and per call, rolling faster as client count grows.

What’s worse is this account is invisible. Currently everyone uses evaluation frameworks to chain Agent behaviors and run them, seeing only total cost at the end. Which clients are profitable, which are losing money, which steps are burning cash—all unknown. Traditional retail loses inventory but at least knows how much at month-end stocktaking; Agent companies’ value leakage can’t even detect whether there’s been loss. Clients extracting all value while you retain no profit—this often isn’t discovered until renewal negotiations, when bargaining chips are already scarce.

Here I must pour cold water on a popular optimistic judgment: models are getting cheaper, so cost problems will solve themselves. This is only half right. Token unit prices are indeed dropping, but reasoning models make each task consume multiples more tokens, so total bills per task don’t necessarily decrease. Two things are simultaneously true: tokens are getting cheaper, bills are getting more expensive. Companies betting on “costs will automatically solve themselves” will likely be solved by costs first.

Three Actions to Balance the Books

Specifically what to do comes down to three actions.

Calculate “human-equivalent cost” for each client. If people did this work, how much salary and time would it cost in the client’s location? The resulting number is your pricing anchor and your basis for future price increases. Clients won’t compare you against competitor quotes—they’ll compare you against their payroll.

Every quarter, check which tier you’re on, then proactively approach clients about upgrading. Don’t wait for clients to raise it. Renewals and expansions are natural windows for pricing negotiations; results are already laid out, making the conversation easiest. Miss the window, and clients will pull you back into the price comparison game in the next bidding round.

Break costs down to the client level. Which clients have negative gross margin—either adjust pricing, change delivery method, or let go. Leaving it unresolved, one money-losing client will quietly consume the profits of two profitable clients.

Finally, clarify boundaries. This playbook assumes your work has countable outcomes and clients are large enough to sit down for customized contract negotiations. If clients only accept standard products, or outcomes can’t yet be objectively determined, then honestly build measurement capabilities at the workflow level first. Climb tiers one at a time, but the direction can’t be wrong: Agent companies stuck at the bottom tier sign more contracts and do more work for clients.