Work as a Service

The economics of running a gallery have quietly broken. Operating costs are rising faster than sales. Art fairs, the engine of the business, now consume roughly a third of total spending. Teams remain small, stretched across five to fifteen fairs a year, and cannot grow. Every new hire consumes the margin it was meant to create.

For thirty years, technology offered the gallery nothing, because software demanded the one resource a dealer does not possess: time to feed it. The industry's reputation for resisting technology was never backwardness. It was rational economics. The dealers who declined to become data clerks were reading the ledger correctly.

That calculation has now changed. Agentic AI is the first technology that performs the work rather than demanding it. Agents coordinate shipments, draft follow-ups, reconcile invoices, and trace provenance. They are briefed like colleagues, not operated like tools. Elite law firms, private banks, and global logistics houses have already crossed this bridge, and the relational side of their businesses emerged not merely intact but strengthened.

This paper describes that shift, from Software as a Service to Work as a Service, and what it demands of the gallery: a workforce priced against labor rather than licenses, configured around each gallery's bespoke way of working, and grounded in the operational record ARTERNAL has spent nine years constructing.

The dealer was right all along: all that matters is the art sale. For the first time, the technology agrees.

Start with the numbers, because they explain everything that follows.

According to the Art Basel & UBS Global Art Market Report, dealers' total operating costs rose by an estimated 5% on average in 2025. That is above the rate of inflation in most major art markets, and above aggregate sales growth. Costs are outrunning revenue.

Where does the money go? Payroll and rent each account for roughly 21% of a primary-market gallery's annual costs. Art fair booths take another 15%, and the travel, shipping, and accommodation around them add 17% more. Fairs alone consume roughly a third of everything a gallery spends.

Meanwhile the demands on the business keep climbing. Production costs are increasingly split with artists. Collectors negotiate harder. The fair calendar sends dealers across the globe five to fifteen times a year. Among the smallest galleries, the average number of buyers fell 40% in a single year, to twenty-nine, the lowest level since 2021. Mid-market galleries reported the sharpest profitability squeeze of any segment.

Here is the trap in plain terms. A gallery cannot scale its way out, because the model contains a structural flaw: doing more fairs requires more labor, and more labor erases the margin the fairs were meant to produce. Working harder is not available either. The hours are already spent. The only exit is to expand commercial capacity without expanding fixed headcount.

You cannot out-hustle math. You can only change who, or what, does the work.

The 15% problem

Beneath the financial squeeze sits a quieter one. Ask a dealer what their business is, and the answer is instant: the sale. The placement, the relationship, the eye. Yet a dealer spends only a sliver of the week actually doing it, by our estimate around 15%, in front of collectors, at dinners, with artists. The other 85% vanishes into the business of running the business: intake, data entry, condition reports, shipping coordination, invoicing, chasing. The most revenue-productive mind in the gallery is systematically consumed by the least revenue-productive work.

This is where the real opportunity hides. A gallery does not need its dealer to sell harder. It needs to give the dealer time to sell at all. Reclaiming the 85% is not a convenience. It is the single highest-return change available to a gallery today.

Move a dealer from 15% selling to 30%, and you have not improved the gallery. You have doubled its capacity to sell.

The art world is routinely called the last analog industry. "Luddite" is the word that gets used. Nine years of building technology for galleries has taught us that this diagnosis is wrong, and the misdiagnosis matters.

Galleries did not hate software. They hated maintaining it.

Every system ever sold to a dealer carried a hidden invoice: the dealer's own time. Enter the inventory. Update the contacts. Log the conversations. Reconcile the spreadsheet. Software stored information. It never did work. The labor of data entry, upkeep, and feeding the system always remained with the human. For a four-person gallery doing ten fairs a year, that was a deal worth refusing. Hours spent feeding a database were hours taken directly from artists and collectors, the only hours in this business that make money.

So the industry refused, rationally, for thirty years. A CRM is a filing cabinet that demands feeding. What a gallery needed was never a better cabinet. It was a colleague who does the filing.

The industry was not waiting for better software. It was waiting for software to stop asking.

Agentic AI is the first generation of technology that asks nothing of the dealer. There is no interface to master, no fields to fill, no system to keep fed. You brief a teammate the way you would brief a new registrar. "Get the Calder to Basel by the 10th. Condition report before it leaves." The work comes back done.

We call this Work as a Service. The unit being sold is not access to software. It is completed work: the coordinated shipment, the reconciled invoice, the drafted follow-up. The same transition is underway across the professional world, and it arrives in the gallery not as features but as roles.

The agentic workforce

None of these roles touch taste, judgment, or relationships. That is by design. The workforce exists to protect those things, and to return the dealer's week to the work only the dealer can do.

Augment, not replace

This distinction matters most, because the word "AI" makes people imagine job losses. The agentic workforce is built for galleries stretched too thin, not overstaffed. The problem is too much work for too few people, and the answer is not fewer people. Reggie does not replace the registrar. Reggie absorbs the administrative substrate of the role: the data entry, the document ingestion, the image handling, the shipping choreography. The registrar keeps the judgment. The sales team keeps the collectors. The role keeps its head; it loses its drudgery. The gallery does not shed people. It finally gives the people it has the room to do the work that grows the business.

The gallery keeps the final word

The workforce never operates unsupervised. Every agent drafts, prepares, and proposes. A human approves before anything ships, sends, files, or posts. Nothing that touches a collector, a shipper, or a dollar leaves the gallery without a person's sign-off. This is not a restraint bolted on after the fact. It is the core of the design, because in an industry built on discretion, control is not optional. The agents do the work. The gallery keeps the final word, on everything that matters.

Per-seat software pricing asks: how many of your people need to log in? Workforce pricing asks a better question: how much work do you want done?

The comparison that matters is not software-to-software. It is workforce-to-workforce. A human registrar costs $65,000 to $130,000 a year, plus benefits, payroll tax, recruiting, training, and turnover risk. That assumes one can be found at all. A digital registrar costs a fraction of that, works through fair week and every week, and never resigns three weeks before Basel.

Galleries already understand this kind of spending better than any industry on earth. A dealer will pay $30,000, $50,000, $100,000 for a single fair booth without hesitation, because the booth visibly drives the sale. It is not a cost. It is revenue infrastructure. The agentic workforce belongs in the same mental category, with one difference: the booth works four days. The workforce works the entire year, at every fair and in every week between fairs, where the follow-ups and the closings actually happen.

Three properties make workforce economics different in kind from both payroll and software:

Stop comparing it to software. Compare it to the salary you are not paying, and the booth you already pay for gladly.

Every relationship-driven industry believed it was the exception. The record now says otherwise.

The pattern

Across every case, three constants hold. The AI never touched the relationship. Adoption required no change in behavior. And the professionals who adopted recovered hours for their craft. The art world is not the exception to this pattern. It is the next chapter, with one structural advantage: having skipped the software era, galleries can move directly to the agentic one, the way economies without landlines went straight to mobile.

A fair question: why can't any AI chatbot do this?

Because an agent is only as good as the record beneath it. A chatbot can draft an email. It cannot coordinate a shipment, generate a condition report, or reconcile an artist payout, because it has no inventory to act on, no provenance history to consult, no collector record to draw from. Action requires rails.

ARTERNAL has spent nine years laying them. Since 2017 we have built outward from relationship management into the full operational architecture of the gallery: client records, inventory, correspondence, mobile tools for the fair floor, and online viewing rooms that show a gallery exactly who to follow up with, and when.

That foundation is what makes the agentic workforce real rather than theatrical. When Reggie coordinates a shipment, the work runs on the gallery's actual inventory, its actual artist agreements, its actual shipping history. When Sal drafts a follow-up, it draws on years of collector relationships already structured in the record. The agents do not start from zero. They start from everything the gallery already is.

For nine years we built the rails. Now we're putting the workforce on them.

Most software arrives as a login link and a help center. That model fails in the art world for the same reason generic software always has: no two galleries run the same way, and they should not.

So we work differently. Every ARTERNAL workforce engagement includes a dedicated engineer, the gallery's own technical architect, who spends time inside the operation and studies how it actually runs. The consignment terms. The artist relationships. The fair calendar. The habits formed over decades. Then the agents are wired around that reality. Not the other way around.

The gallery's bespoke way of doing business is not an obstacle to be standardized away. It is the blueprint we build to. When the engineer is finished, the technology disappears. The work simply starts getting done.

This is the same white-glove pattern behind the most successful AI adoptions in adjacent industries: a person inside the practice, the system shaped to the professional rather than the professional retrained to the system. It is the reason those efforts reached near-total adoption while conventional software dies in a drawer.

There is one more reason this industry has been right to be careful, and it is the strongest argument for building this inside ARTERNAL rather than alongside it.

The art world runs on discretion. Collector identities, prices, provenance, the quiet negotiation. Confidentiality is not a feature of this business; it is the business. And yet, in galleries everywhere, well-meaning team members are pasting exactly that information into consumer chatbots to save a few minutes on an email.

The numbers describe a quiet leak. 77% of employees who use AI tools have shared sensitive company data through them, and 82% of that pasting happens through personal accounts no one in the business can see or control. Nearly two-thirds of organizations have no policy governing any of it. IBM found that one in five organizations has already suffered a breach traced to shadow AI, at an average cost $670,000 higher than ordinary breaches, with customer data the most commonly compromised.

This is the part of the agentic era that nobody selling AI tools wants to discuss. The choice facing a gallery was never AI versus no AI; the team has already chosen. The real choice is whose walls it happens inside: ungoverned chatbots on personal accounts, or a workforce that operates within the gallery's own environment.

ARTERNAL is built on the second model. Reggie, Sal, Miya, Dana, and Finn work on the gallery's own record. The agents come to the data; the data never goes out to them, and it is never used to train public models. The team gains the capacity of AI, and collectors keep the discretion they have always trusted the gallery to provide.

The market has already priced this. 95% of executives say they will not buy from organizations that fail to protect their data, and 64% worry about exposure through public AI tools. Given the choice, people choose the walls.

You cannot recall a collector list from a public model. The only safe place for an agentic workforce is inside your own walls.