I get asked to look at partner programs that are not producing, and a good share of the time the diagnosis is the same. The program was built for resellers and the partners signed up are implementation firms. Or the reverse. Nobody is doing anything wrong, they are just playing different games.
The model you pick decides who owns the customer, who does the work, where the margin goes, and what you have to build internally. Getting it wrong is expensive in a slow way, because it takes a year of quiet underperformance before anyone questions the premise.
The three shapes, and what they actually mean
Referral. The partner makes an introduction. You sell, you contract, you support. They get a fee, typically 10 to 20 percent of first year, sometimes a flat amount per closed deal.
Reseller. The partner sells and invoices the customer. You sell to the partner at a discount, usually 25 to 40 percent off list depending on the ecosystem and the tier. The customer relationship is theirs.
Implementation or services partner. They do not resell your software at all. You sell direct, and they deliver the implementation, the configuration, the training. Their revenue is their services rate, and your product is something they are certified on.
There are hybrids everywhere. Plenty of Acumatica and Sage VARs resell your product and implement it, which is really the reseller model with delivery attached. NetSuite has its own gravitational rules. The point is not the taxonomy, it is that these have different economics and different failure modes.
What each one asks of you
Referral is the cheapest to start and the least durable. You need a way to register a lead, a way to pay a fee, and someone to actually sell the deal. That is a light lift, which is why so many ISVs start here.
The problem shows up around deal twelve. A referral partner who sends you five good leads a year is making a few thousand dollars from you and a lot more from something else. You are a side item on their menu. When their quarter gets tight, they work on whatever pays them most, and that is never the referral fee.
Reseller is the most expensive to build and the only one that produces partners with real skin in the game. You need partner pricing, deal registration, margin protection, a way for them to quote, enablement so they can demo without you, and someone available when they have a question at 4:45 on a Friday.
That last one is not a small item. Reseller programs are a support obligation before they are a revenue channel.
Implementation partner is the one ISVs underrate. You keep the customer relationship and the full license revenue, and you get delivery capacity you did not have to hire. In ERP ecosystems this is often the natural shape, because the VAR is already doing the implementation of the platform and adding your module to that scope is a small step for them.
What it asks of you is certification, documentation good enough that someone can implement without calling you, and discipline about not competing with their services revenue.
The question that usually settles it
Who does the customer think they are buying from?
In mid-market ERP, the answer is frequently the VAR. The customer picked Acumatica or Sage largely because they trusted the partner who brought it to them, and that partner is their software person. If you try to sell direct into that account without the partner, you are asking the customer to manage two relationships and to go around the person they trust.
That pushes you toward reseller or implementation partner, not referral. Referral works better when your product is bought by a different buyer than the ERP (a compliance tool bought by finance, say, where the VAR is a helpful introduction but not the trusted advisor for that decision).
The second question: how complex is your implementation? If it takes two hours of configuration, a reseller can handle it and services revenue is not interesting to them anyway. If it takes six weeks, services revenue is the reason a partner will care about you at all, and a model that gives them none of it will not hold their attention.
The mistake I see most
ISVs pick reseller because it sounds like the real channel, then build a program with reseller discounts and referral-level support.
The partner gets 30 points of margin and no demo environment, no pricing tool, no lead time on roadmap, and a support queue that treats them like any other customer. So they sell the product once, spend forty unbilled hours getting the customer live, decide it was not worth it, and go quiet. You conclude the partner was not committed.
Margin is what a partner earns for taking on the customer relationship. Everything else in the program is what makes that relationship survivable. If you cannot fund the second part, run referral honestly instead, or run implementation partner where the money comes from their services rate rather than your margin.
Mixing models without creating a mess
You can run more than one. Most mature programs do. The rule is that a given account has one model at a time and everyone knows which.
Where it goes wrong is when a referral partner sends a lead into an account a reseller considers theirs, or when your direct team sells into an account a partner introduced you to eighteen months ago. That is a conflict problem, and it is solved with account registration and a written rule about how long registration lasts, not with goodwill.
The other version of mixing that works: start partners as implementation partners and let the good ones graduate to reselling. It lets you see who is actually competent before you hand them the customer relationship, and it gives the partner a reason to invest in certification early.
What to do if you already picked wrong
Do not quietly change the terms on existing partners. That is how you lose the two who are producing.
Grandfather the current arrangement, define the new model clearly, and offer a path. Most partners who are not producing will not take it, which is fine and tells you something. The ones who do take it are the ones worth building around, and you will have learned which is which for the cost of one honest conversation each.