Your back office is an asset. Know what it is worth before anyone makes you an offer.
AI labs now license de-identified copies of everyday operating records: work orders, tickets, email, invoices, years of history. Anchor packages yours, brings several buyers to the table at once, and you keep the originals.
- Nothing leaves your systems to get a quote
- Samples only under an NDA you sign in the portal
- Nothing released without your written sign-off
A manifest: every system you run, how far back it reaches, and whether it can be exported.
Name the product, not the job it does. A buyer can price “ServiceTitan, 11 years, 386,000 work orders”; they cannot price “our scheduling software”. Volume is what gets priced, so a count beats a description. “Not sure” is a real answer in every column, and the retired systems count too.
Your manifest lives in the portal. Invite the people who know each system, save as you go, and submit when it is honest, not when it is perfect.
Example rows from a commercial contractor's manifest. Yours lists the systems you actually run.
What Google paid in Spirit Airlines' 2026 bankruptcy sale for the airline's internal emails, chats and operating records, with personal data removed by a third party before delivery.
Axios, Aug 17, 2026; Forbes, Aug 18, 2026The public internet is used up. The records that teach an agent how a company actually runs were never published.
The next generation of models is trained as agents that do real work inside businesses, in the tools businesses use. Labs build practice environments that mirror ERP, CRM, email and ticket queues, and they need long, ordinary operating histories to fill them: the decisions, exceptions and handoffs that synthetic data cannot fake.
Who is buying: frontier AI labs, and the data and environment companies that supply them. Most deals are licenses of a de-identified copy, limited by contract to model training.
Six steps, about six weeks, and a gate you control at each one.
Most companies hear from one buyer, once, and the offer looks reasonable because there is nothing to compare it against. Anchor runs the process so the price is set by competing offers rather than by whoever reached you first.
- Weeks 1–2
Step 1: Inventory and package
You list the systems you run and how far back each reaches. Anchor builds the offering memorandum: sources, record counts, date ranges, what makes the data valuable, and de-identified samples.
Gate: nothing moves forward without your written sign-off on scope. - Week 1, in parallel
Step 2: Buyer matrix
The frontier labs are a known list. Everyone below them moves with funding and with what each buyer is short of, so the list is built fresh for your data: operational workflow, communications or engineering history.
You see the list before anyone is contacted. - Weeks 2–4
Step 3: Outreach
Existing relationships plus direct outreach to every buyer that fits. A weekly update in your portal: who was contacted, who responded, and what the feedback was.
Your company is not named to a buyer until you say so. - Weeks 4–5
Step 4: Diligence and Q&A
Sample requests under NDA, use-case calls, structure conversations. Exclusive, non-exclusive and ongoing-feed structures are put to all bidders at the same time, not one at a time.
Gate: samples go out only under an NDA you have signed. - Weeks 5–6
Step 5: Compare and close
Offers compared on more than price: exclusivity, permitted use, retention, audit rights and refreshes, side by side in your portal. You choose. After signing, a copy is exported from the systems you already use and de-identified before any buyer access.
Gate: you pick the offer, or none. Counsel review is standard. - Ongoing
Step 6: Keep the field open
New buyers get funded and existing ones change what they need. A pass is usually timing, not rejection, and each is re-approached when the reason expires.
Refreshes, if any, are priced per refresh and signed separately.
Ordinary records, in volume. Identities, never.
Communications and operations
- Email, chat, meeting notes and transcripts
- Work orders, tickets and project history
- ERP and CRM records, with their activity trails
Documents, finance and quality
- Files, drives, SOPs, proposals, engineering and design files
- Invoices and accounting workflows
- Inspection and maintenance logs
Excluded or removed, every time
- Customer, patient, employee and vendor identities and personal information
- Credentials, keys and raw system access
- Anything you mark confidential, under NDA, or outside the signed scope
- Source code, unless you choose to include it
What is your data worth?
It depends on which systems, how many years, how complete the record is, and how many buyers want it. The desk shows how the first three move a range. The fourth is what a competitive process is for.
The figures are an illustration of direction, not an offer. Your real range is set by an analyst from your manifest and from what buyers are currently short of, and it arrives in your portal with what counts, what stays out and why.
The six we hear on every first call.
Do we give up our data?
No. It is a license of a copy. The originals stay with you and in use.
What about customer privacy?
Personal and identifying information is removed before any buyer access.
Is it legal?
Companies generally own their operating records; contracts define permitted use and exclusions. Review by your counsel is standard.
Could a lab use it to compete with us?
Permitted use, restrictions, retention and exclusivity are negotiated in the license before anything is released.
How much work is it?
Mostly exports from systems you already use. Little ongoing time from your team.
What is our data worth?
It depends on which systems, how many years, how complete the record is, and how many buyers want it. The manifest is how we find out.
Start your company's manifest.
About an hour, in your own time, with the people who know your systems. An indicative range follows in your portal.

