Whether AI automation is cheap or expensive for your agency has very little to do with the AI, and almost everything to do with how your management system lets data in and out. Two agencies asking for the identical automation can receive quotes that differ by a factor of three, and the difference is usually not the vendor being greedy. It is that one agency's system has a documented way in and the other's does not.
This is a guide to working out which situation you are in before you start asking for quotes.
What actually determines the cost
There are three tiers, and knowing which one you are in tells you most of what you need.
A documented partner API. The system exposes a supported way for an approved partner to read policy and client data and write results back. This is the cheapest and most durable path, because nothing depends on screen layouts that change or on a person moving data by hand. HawkSoft is a clear example: its Partner API documentation is openly accessible, and agencies opt in to sharing data with a specific approved partner.
Export-only. You can get data out reliably, usually as a book-of-business or production export, but there is no supported way to write back. A great deal of useful automation lives here, because analysis does not need a live connection. What you give up is the last step: a person pastes or uploads the result instead of it appearing in the system automatically.
Gated integration. The vendor treats integration as a commercial arrangement, with negotiation, fees, or a partner status that takes time to obtain. Automation is still possible, but the approval timeline usually becomes the project timeline, and that is worth knowing before you plan around a launch date.
Most independent agencies are in the middle tier and assume they are in the third. That assumption is expensive, because it stops projects that would have worked.
What works with no integration at all
This is the part most agencies underestimate. A large share of the highest-return automation in an agency depends on documents and exports rather than on a live connection.
- Policy comparison and coverage gap analysis. Reading two policies and finding what changed needs the policy documents, not the AMS. This is usually the single highest-value automation in a P&C agency and it starts working the day you have PDFs.
- Loss run summarisation. Same principle. The loss run is a document.
- ACORD and carrier form extraction. Pulling structured data out of forms is document work.
- Book analytics. Carrier concentration, retention by line, producer growth. All of it runs on the export your AMS already produces.
- Renewal preparation. Assembling what a producer needs before a renewal conversation is mostly reading and summarising.
What genuinely needs API access is the write-back: having the result land in the client file automatically instead of a person putting it there. That is a real efficiency gain and worth pursuing, but it is the last mile, not the entry ticket. Starting on the document work while integration approval proceeds in parallel is usually the right sequencing.
What to ask your AMS
Ask these directly, and ask early, because the answers change what is worth scoping.
- Do you have a partner or integration programme, and is the documentation public? Open documentation is a strong signal. It means integration is a supported path rather than a negotiation.
- Does the request have to come from us or from our vendor? Most systems treat data sharing as something the agency opts into, which means the principal or owner usually has to make the request. Projects stall for weeks on this, purely because nobody realised the request had to come from inside the agency.
- What does approval involve and how long does it take? Security review is common and reasonable. What matters is the elapsed time, because it becomes your critical path.
- What can be written back, not just read? Read access enables analysis. Write access enables the workflow to close without a human step.
- What does it cost, and who pays? Sometimes the agency, sometimes the partner, sometimes nobody. It is a short question that avoids a surprise later.
A note on switching systems
Every so often someone concludes their AMS is the obstacle and starts considering a migration.
Be very sceptical of that, particularly if it is suggested early by someone who would be paid to build the replacement. Migrating a management system is among the most disruptive things an agency can do, it consumes months of staff attention, and it puts your system of record at risk to enable a project that has not yet proved its value. The automation you want is nearly always achievable on what you already run.
Changing systems as a deliberate long-term platform decision is a reasonable thing to do on its own merits. Changing them so that an automation project can proceed means the project has stopped serving the agency.
How to sequence this
If you want the shortest path from here to something working:
Start with a document workflow. Pick the one your team complains about, where the input is a PDF or an export you already have. Nothing external blocks it, so it can start immediately and it will tell you whether the approach is worth extending.
Ask your AMS about integration in parallel. Send the question the same week. If approval takes six weeks, you want those six weeks running alongside the build rather than after it.
Add write-back once the analysis is proven. By then you know whether the output is good enough to be worth automating into the client file, which is a much better position to negotiate integration from.
We work across Applied Epic, EZLynx, HawkSoft, AMS360 and Jenesis, and the pattern above holds regardless of which one you run. If you want to know what is realistic on yours, the AI audit is free and will tell you which tier you are in and what it would take, or you can see how we structure engagements first.
Key Takeaways
- Cost is driven by how your AMS lets data in and out, not by the AI itself.
- Three tiers: documented partner API, export-only, and gated integration. Most agencies are in the middle and assume the worst.
- Policy comparison, loss run summarisation, ACORD extraction and book analytics all work with no integration at all, because they run on documents and exports.
- API access matters for write-back, which is the last mile rather than the entry ticket.
- The integration request usually has to come from the agency principal, not from your vendor. Projects stall for weeks on this alone.
- Ask about approval timelines early, because they tend to become the project timeline.
- Switching AMS to enable an automation project is almost always the wrong trade.