A single museum running a single audio program is the easy case. The hard case — and the one most of this category's vendors quietly avoid — is a museum system running interpretation across multiple sites: the Smithsonian's 21 museums plus its more than 200 affiliate institutions, a regional historical society with eight county properties, a university museum network with a flagship and three discipline-specific galleries, a national-trust-style portfolio with hundreds of heritage sites under a single brand and budget. The operational questions multiply: who owns the script, whose logo appears on the splash screen, where do the analytics roll up, and what happens when the one person who knew the Civil War wing leaves.
This piece is the working answer we give to multi-site directors who land on our pricing page, then realize the math doesn't change the underlying question — which is how to run interpretation across a portfolio without either centralizing it into beige sameness or letting each site improvise its own incompatible program.
What counts as a "multi-site" tour program?
A multi-site program is one where a single org has editorial, financial, or brand responsibility for interpretation across two or more visitor-facing venues. The three shapes we see most often: a museum system (one parent, many directly-operated sites — Smithsonian, Tate, Te Papa's regional outposts), a federation (one umbrella brand, semi-autonomous member sites — a state historical society with county chapters, a diocesan heritage program, a Smithsonian Affiliate-style network), and a portfolio (one operating company across heritage sites with distinct identities — National Trust, English Heritage, a university museum system).
The operational distinction matters more than the org chart. A museum system has central editorial authority by default; a federation has it only where the charter grants it. A portfolio usually has central platform decisions but per-site brand sovereignty. The right governance model for tours is the one that matches the model already governing your wall text, your collection records, and your hiring — not a new one invented for the tour program.
What should be shared across sites, and what shouldn't?
Share the things that don't change by site, and let each site own the things that do. The reusable layer is bigger than most teams expect. Artist biographies, period overviews, technique explainers ("what is encaustic," "what is a daguerreotype"), accessibility descriptions for common object types, glossary entries, and conservation context all write once and travel everywhere. A shared library of 200 of these pieces, professionally edited, raises the floor of every tour in the system and lets site curators spend their time on the interpretation that's actually site-specific: this room, this object, this provenance, this community.
The site-specific layer is the framing. Why does this site have this object? What does the local community know about it that nobody else does? What did the donor's family ask be remembered? A Frederick Douglass biography is the same biography across a network; what's said about Douglass at a Rochester site is not what's said at an Anacostia site. The platform's job is to make the shared layer easy to inherit, easy to override, and easy to track ("this stop uses the central bio, edited 2025-03-12, with site-specific framing added 2026-04-02"). When inheritance is invisible — when a site curator can't tell if she's editing the shared copy or her override — the program gets brittle fast.
Central editorial control or distributed? Pick one.
The two patterns that actually work are central editorial with distributed authoring, or distributed editorial with central standards. Pick one in writing; the hybrid drift kills programs.
Central editorial with distributed authoring means the parent org owns voice, register, accuracy standards, and final publish authority. Site curators draft and edit; a central editor — usually the system's head of interpretation or chief curator — approves. This is the Smithsonian-style pattern, and it produces consistency at the cost of velocity. It's the right answer when the brand is the product (national institutions, marquee university museums) and the visitor's mental model of the system is "one museum, many locations."
Distributed editorial with central standards means each site owns its voice and publishes on its own authority, but signs a charter that defines the floor: factual accuracy review, accessibility requirements, multilingual minimums, prohibited language, brand guardrails, citation standards for sensitive content. The center audits; the sites publish. This is closer to the National Trust or Smithsonian Affiliate pattern, and it's the right answer when the sites have real local identity and the visitor's mental model is "this place, supported by that organization."
The pattern that doesn't work is the unwritten hybrid where everyone assumes everyone else has authority. Tours stall in approval, sites improvise around the system, and the central org finds out about a problem when a board member visits a satellite location and reads something off-message. Write the model down before you scale. Convo's three-status workflow (draft → review → published) can support either model, but the model has to live in policy, not in software.
What does portfolio-level visibility actually look like inside one platform?
A multi-site program on one platform is comparable; the same program spread across different vendors isn't. When each site uses a different vendor — one on a legacy handset platform, one on a different web platform, one with nothing — there is no portfolio view. The director of the state historical society cannot tell a board which sites are over- or under-performing on visitor engagement, because the metrics don't agree on what a "start" or a "completion" is.
Inside a single platform, the practical pattern that's worked best for multi-site orgs we've talked to is one admin tenant per umbrella org, one tour per site, all sites sharing the same admin. Each site's tour is its own unit — its own scripts, its own languages, its own Q&A. Completion numbers can be rolled up across all tours in the org, so a director gets a single view of how many visitors made it through interpretation across the whole portfolio in a quarter, alongside the per-tour breakouts for the sites that need attention.
What this view doesn't include: dwell-time benchmarks across sites. Most platforms in the AI-narrated category — including Convo — don't track dwell time at the gallery level. It's a gate-sensor measure, not a phone-based one. A portfolio dashboard built on tour analytics will tell a director about scans, completions, and questions, but not about how long visitors stayed in front of an object. The right next move there is gate-data integration, not pulling more from the audio platform.
Per-tour Q&A clustering — anonymized themes from what visitors at each site are asking about, grouped — is itself unique to the conversational category and gives a system director a separate view of curiosity per site, comparable in shape if not in detail. The board-meeting version: each site's most-asked-about themes lined up next to its completion data, so a director can see where wall text is leaving curiosity on the table and where a particular interpretive frame is connecting or not across the portfolio.
How do you survive staff turnover at one site?
Treat every tour as something the next curator will inherit, because the museum field's turnover patterns guarantee they will. The American Alliance of Museums and field surveys from 2024–2025 have made clear what's been quietly true for years: museum workers face significant precarity, and turnover among curators, education staff, and visitor-experience leads is high — particularly at smaller and regional sites, which are exactly the sites a multi-site system depends on. The U.S. Bureau of Labor Statistics projects about 4,800 openings per year in the curator/archivist/museum-worker category through 2034, many to replace exits.
What this means for a tour program: the curator who wrote the Civil War wing's audio guide may not be there in three years. If the only place that knowledge lives is in her head and in a finished MP3 nobody can edit, the tour becomes uneditable. Updates stop. Errors compound. Eventually the whole tour is replaced because no one can confidently revise it.
The structural fix is reference-material-first authoring. Every tour is built from uploaded source documents — exhibition notes, catalog entries, scholarly references, wall card text, the curator's own working files — that live in the platform and are tagged to each stop. When a new curator inherits the wing, she inherits the sources, not just the script. She can see what each stop is based on, regenerate or edit any line with the context preserved, and know what is settled versus what is her predecessor's interpretation to revisit. This is the way Convo is designed to be authored, and it's the single most important design choice for a multi-site program — because the people who built the tour will not always be the people running it.
The secondary fix is published-state freeze with auditable history: nothing visitors see can be silently changed by a successor without a logged review. Combine the two and a tour becomes a maintainable institutional asset, not a personal one.
What about traveling exhibitions across the system?
A traveling exhibition that visits three sites in a system is the cleanest possible test of a multi-site platform. The interpretation is identical; the surrounding framing is local. The platform should let a central team build the exhibition's tour once, ship it to each site as it opens, and let the site add a short local intro and any community-specific framing without forking the canonical content.
We covered the broader pattern in audio guides for traveling exhibitions; the multi-site case is a subset where the host sites all sit inside one platform tenant. The key advantages are version control (when the central team fixes a label error, all three sites get the fix; the exhibition doesn't become three slightly-different exhibitions) and consistent analytics across the three runs.
How does this affect what you'll pay?
Multi-site programs change the pricing conversation from "per tour" to "per platform." Legacy audio-guide vendors typically scale linearly per site, because each site is its own production line. Most AI-native platforms — ours included — scale on platform tier rather than per-tour, which means the marginal cost of adding a site is low once the system is in place. We get into the full taxonomy in audio guide pricing models; the multi-site short answer is that a single Institution tier can usually cover a small system, and an Enterprise contract is where larger systems negotiate the volume and the institutional terms.
Where the multi-site case breaks down
We try to include an honest "where this doesn't work" section in every operations piece. For multi-site:
A multi-site platform is the wrong call when the sites genuinely don't share editorial DNA — when, say, an investment vehicle owns an aquarium, a science center, and a historic estate that have nothing curatorially in common. In that case, "multi-site" is an accounting artifact, not an operational one. Each site is better served by its own program decision.
It's also the wrong call mid-contract. If three of your eight sites are on a five-year handset deal with two years left, the move is to plan the migration for that contract's end, not to break the contract early. Run the new sites on a phone-based platform now and converge at renewal.
And it's the wrong default for a small system — two sites, both small, both run by the same one or two people — where the operational overhead of formal governance exceeds the value. Two sites can usually run one tour program informally; eight sites need policy.
FAQ
The verdict
A multi-site audio tour program lives or dies on three decisions made before any platform is chosen: what's shared and what's site-specific, who has editorial authority and where, and how the program survives the people who built it. The platform's job is to encode those decisions cleanly — shared content with override, comparable analytics across sites, and reference-material-first authoring so a tour outlasts its author. Get the governance right and the technology question gets easier. Get it wrong and the best platform in the world becomes a fork-tree of incompatible sites in eighteen months.
For the broader operational picture, see the operations pillar guide. For the pricing implications, audio guide pricing models. For how a single tour gets built and kept current — the foundation everything in this piece sits on — see product.
About the author
Eric Duffy is the founder of Convo, a platform that helps museums and cultural institutions publish multilingual audio tours their visitors can have a conversation with. He writes about the operations of interpretation programs from inside the category, drawing on discovery conversations with directors of multi-site museum systems, regional historical societies, and heritage portfolios. Reach him at eric@convo.app or on LinkedIn.