My product is offering ML models (classification, recommendation, ranking) through web services. We are offering it for $149/month. You own your data. You can make REST api calls to get output of model. Let me know if you are interested. My email is aimlmodelfarm@gmail.com
Your product may be valuable to an organisation when:
(i) the org has a standard kind of decision they need to make regularly - for instance, sales forecasting. (ii) the org makes that decision sub optimally, in a way that could be improved by ML. (iii) managers/executives in the org are actually aware there's a problem (i.e. their sales forecasting is poor), and that sub-optimal decision making has serious consequences for them (increased costs, reduced profitability, reduced revenue, increased churn of clients/staff, increased risk) (iv) there are skilled personnel available who can analyse the situation and join the dots to realise that an appropriate ML model could form part of an effective solution. (v) the org has an existing process to generate and record the relevant input data necessary to make better decisions. (vi) there's enough executive support to do a proof of concept to evaluate if applying ML will produce a good return on investment. (vii) the proof of concept suggests there will be a sufficently profitable return on investment if the project succeeds. (viii) there are skilled personnel available who can productionise the proof of concept - this includes writing the REST API calls to your ML hosting product, as well as doing the remaining 95% of development work to integrate into the org's processes, such as data pre-processing, model output post-processing, data ingestion, report generation, preparing training documentation & training the orgs staff (ix) the org can retrain existing staff or hire new staff to perform the new job of care and feeding of data collection & monitor that the model isn't producing garbage decisions
Here's my guesstimate:
Suppose there's a business making repetitive decisions so sub-optimally that if your ML product were to be successfully integrated, it could lift that business' top-line revenue by 2% . Suppose it costs the business $150k to hire consultants to do an initial feasibility study that demonstrates that an ML-based solution for sales forecasting can lift revenue by a few percent, build a rough prototype, and productionise it. Excluding your product, the ongoing costs to maintain the overall IT system that incorporates this bit of ML might be $70k / year -- mainly staff costs to collect data & execute decisions based on the model output, say. There might also be $20k / year in maintenance costs of related IT systems & maybe a $10k / year retainer to consultants who give periodic tune-ups & support.
What's the smallest amount of revenue this business needs to have so it might possibly get a decent return on investment on this project? The business probably needs to have revenue of at least $7 million dollars.
Based on this kind of argument, your potential customers are:
* managers or executives of businesses with more than $7 million dollars in revenue that have opportunities where ML can be profitably applied. these people must be aware that they have a business problem and also aware that it might be possible to profitably apply ML to that business problem. they will only be able to consume your product if they have existing relationships with staff or consultants who can analyse, prototype, evaluate, integrate & deploy a ML-based solution for the business problem
* managers or executives of businesses with more than $7 million dollars in revenue that have opportunities where ML can be profitably applied but are not aware that they have a business problem where ML could help. it would be challenging to sell to these people as they would need to be shown that they have a problem/opportunity first, in language that they understand relating to their business's operations and goals. this would need to happen well before anyone talks about ML.
* consultancies who engage with businesses to identify and solve these kinds of business problems, that have in-house consulting expertise & ML modelling / feature engineering / domain modelling expertise, but lack the in-house software engineering & operations expertise to reliably run a web service that hosts a model
Here's some good news: the total cost to a business of deploying and operating an ML-based solution using your product is a lot larger than $150 / month. That means it is possible that you could increase your price by 5x to $750/month without it really changing the return on investment calculation from the business' perspective.
Less good news: the most profitable opportunities to apply these techniques are in the orgs with the largest scale -- large-huge enterprises, not small-medium ones. The complexity and cost of setting up an ML solution is often independent of the amount of volume going through the ML, but the benefits are proportional to the volume. These very large orgs may already have large internal IT teams with the in-house capability to do what your product does, as well as lots of enterprisey requirements around risks of their suppliers/vendors going bankrupt, compliance with data protection/privacy laws, auditability of decision making, encryption of data, network security policies, high availability, volume & latency of decision execution.