AI is increasingly part of the way that businesses operate. It can be used to automate the mundane, improve customer service and analyse huge volumes of data. There's no doubt about the opportunities it presents. The issue is in picking the right solution.
For the big organisations it's not just about buying the software. You've got to think about security, integration, deployment, cost and whether the technology actually does what the vendor has said it will do.
A bad decision can end up costing you money, disrupting your operations and creating security issues. That's why the Enterprise AI Solutions selection process has to begin with a very clear understanding of the business requirements and how the providers will meet those requirements.
This is the list of the key issues that the CIO, the IT head and the procurement team will need to consider prior to beginning the shortlisting of an enterprise AI solution.
Before you go out and start comparing AI vendors you need to have a very clear idea of what it is you're trying to improve.
It's so easy to get caught up in the whole demo and the promise of increased productivity. The truth is that no matter how sophisticated the AI platform is, it's going to be of very limited value if it's not actually solving a real business problem.
A company with problems around customer service response times would be a good candidate for an AI powered support agent. An organisation drowning in financial paperwork would also be a good fit for intelligent document processing.
These are two entirely different requirements.
Start by identifying the tasks that are the biggest consumers of resources, the problems that staff run into on a regular basis and the outcomes that the business is trying to achieve.
Set some measurable targets (e.g. reduce processing times, reduce operating costs, improve customer satisfaction). Having clear objectives makes it a lot easier to assess the options and to work out whether an investment is justified.
One of the initial technical decisions is how the AI solution is to be deployed.
Some organisations will go for a cloud platform simply because it's easier to get up and running and to maintain. Others will want more control over their infrastructure, particularly if they're working with sensitive data.
There are a number of common deployment options.
An external provider offers cloud based AI solutions. These generally provide faster implementation, automatic updates and the ability to scale as and when required. The downside is that you do have to know where your data is being processed and where it's being stored.
On-premises AI runs within the organisation's own infrastructure. This gives you greater control over the systems and the data but often requires a larger investment in hardware, maintenance and the required technical skills.
Hybrid AI solutions involve the use of cloud services alongside privately managed infrastructure. They can offer a degree of flexibility but can also add to the management complexity as you end up with multiple environments.
The correct answer is going to be down to the organisation's existing systems, its regulatory requirements, its budget and its security needs.
Before you go any further and pick a provider, ask where the data will be processed, what infrastructure is required and whether the deployment model can scale to meet future demand.
Security should be a prime consideration when you're selecting your Enterprise AI solutions.
AI platforms routinely have access to confidential business information including customer data, financial data, employee data and internal documents.
No controls and sensitive information could be accessible to the wrong people.
Imagine an employee using the AI to search the company's own documents. If the platform doesn't respect the existing access permissions then it could well return information that the employee isn't entitled to see.
That's a genuine issue even if the AI comes up with the right answer.
When looking at vendors assess their approach to identity management, access controls, encryption and activity monitoring.
Look for role based access control, single sign on, multifactor authentication and full audit logging.
Also consider whether customer data is used to train the AI and for how long the data is retained.
Vendors should be able to provide the relevant security documentation and be able to explain how their systems satisfy the applicable data protection requirements.
Don't just go on the security claims in the sales presentation. Ask for the evidence and test the key controls during your evaluation.
The vast majority of organisations already have a huge number of software platforms in place to run their business.
A company could have Salesforce for their CRM, SAP for finance, Microsoft 365 for collaboration and a number of internal databases.
Introducing an AI solution that can't talk to those systems is likely to create more problems than it solves.
Integration should be the key differentiator.
Ask whether the platform provides ready made connectors, APIs and is compatible with your existing auth systems.
Think about the flow of information between the applications and whether staff are going to have to manually transfer the data.
For example an AI sales assistant is a lot more useful if it can pull the correct customer information out of the CRM rather than the sales rep having to go from tool to tool to find it.
However integration is more than just integrating two apps.
IT teams should also be looking at data synchronisation, system reliability, error handling and the ongoing maintenance requirements.
A solution that works a absolute treat in a demo will quite possibly do something very different when you actually put it into an enterprise environment.
Testing the integrations early can prevent a very expensive surprise when you've actually got it all implemented.
AI vendors are always advertising the productivity gains, the cost savings and the automation they can provide.
While those claims may be very encouraging you absolutely should not accept them without evidence.
A platform that works a treat for one organisation can be absolutely totally different for another.
Ask the vendors for case studies, customer references, performance benchmarks and examples of them having worked with other businesses that have similar needs.
More importantly, run a proof of concept prior to you committing to a long term contract.
A proof of concept enables the company to test the solution against actual business scenarios.
For example a Procurement team looking at an AI document processing tool could test it on invoices of all sorts of layouts, in all sorts of languages and at all levels of incompleteness.
The team could then go on to assess processing accuracy, time saved and the number of documents that still required manual review.
These results give a far far clearer picture of what the platform can do than a pre-prepped product demo would have done.
It's also worth testing how the system deals with the oddball request, the request that contains incorrect information and the situation where a human has to get involved.
Reliability is more important than spectacular results under ideal conditions.
An AI platform could be absolutely spot on in a small pilot with 20 users. That in no way implies it will be equally good when 5,000 users start to use it.
Enterprise requirements can change very rapidly.
New departments could adopt the platform, data volumes could increase and new workflows may well require automation.
Before you even begin the shortlisting process you need to have a good understanding of how the provider's technology will cope with the ramping up of demand.
Ask about usage limits, processing capacity, performance under heavy load and the cost of scaling out.
Think about whether the platform supports further integrations and whether admins can have separate control over different departments.
Long-term flexibility also matters.
Companies should understand how easy it is to export their data, switch providers or migrate to another deployment environment.
Vendor lock-in can become expensive when your business changes.
A good enterprise AI solution enables you to grow without forcing you into any technical or contractual obligations that you didn't want to be in.
Implementation is just the beginning of an enterprise AI project.
Employees will need training, the integrations may need some work and there will undoubtedly be some technical issues.
The quality of vendor support can have a huge bearing on the success of the technology in terms of adoption.
Before you sign the agreement you should check out the provider's customer support options.
Find out if technical assistance is available during your operating hours, how rapidly critical incidents are dealt with and whether you'll have access to dedicated implementation specialists.
Service level agreements should define availability commitments, response times and escalation paths.
How does the vendor communicate product updates and changes to the AI models?
These updates can occasionally have an impact on system behaviour, particularly for organisations that have a lot of automated workflows.
A good provider should be able to tell you how major changes are tested and how they're communicated.
Customer references can also be of use here. A chat with a few of the enterprise customers might reveal how the vendor deals with things after the sale.
Good support is not a nice to have. It's absolutely vital to business continuity.
The cheapest AI platform isn't necessarily the one that's going to be the most cost effective over the long term.
Subscription fees are just one part of the total investment.
Organisations may also pay for implementation, system integration, cloud infrastructure, employee training, customisation and ongoing technical support.
Some charge you on the number of users, some on the number of API requests, some on the volume of processing and some on the number of AI tokens consumed.
These differences can make cost comparisons very difficult.
To get a more accurate picture, work out the total cost of ownership over a three year period.
Include both the certain costs and any potential increase in usage.
For example an AI platform may look rather cheap in a small pilot but can be a lot more expensive when you start rolling it out across a number of departments.
Procurement teams should be asking for full pricing breakdowns and realistic usage scenarios.
It's also worth bearing in mind the financial benefits.
If it cuts down on manual processing, improves operational efficiency or reduces errors then those are all factors that would go into the business case.
The aim is to look at the value delivered against the full investment, not just the purchase price.
It's difficult to compare the various AI providers as each is being assessed against a different set of criteria.
One sales pitch might be all about the features, another all about the security, another all about the price.
With no framework you're just left unable to say which is the best option.
A properly structured evaluation process will go a long way to addressing this.
Develop a scorecard against the key purchase criteria (security, deployment flexibility, integrations, performance, scalability, support and cost).
Assign weights in accordance with the organisation's priorities.
A financial services company might be all about data protection and compliance, whereas a fast growing tech company might be all about integration and scale.
Evaluate each vendor against the same criteria and provide the supporting evidence.
However some requirements should be mandatory rather than just being scored.
If a provider is unable to meet an essential security requirement then a lower price should not be the compensation for that.
This makes the shortlisting process vastly more transparent and gives the procurement team a far better basis on which to make their recommendations.
Once the shortlist has been reduced, the next stage is to test the most promising solutions in a controlled environment.
A pilot allows the organisation to see how the technology works in real business use.
Pick a specific workflow and define clear success criteria before testing even starts.
For example a company looking at an AI customer service agent would look at response accuracy, average handle time, customer satisfaction and the % of requests that need to be escalated to a human.
Include the staff who are actually going to be using the system.
Their feedback can point to usability issues that a technical assessment might well overlook.
The pilot should also look at the security controls, the reliability of integration and the operating costs.
By the end of the testing period the results are benchmarked against the original business objectives.
If the solution works then they can go on to develop a phased implementation plan.
If it falls short then the findings can be used to assess whether further customisation is worthwhile or whether a provider would in fact be the better option.
What should businesses be looking at when selecting an enterprise AI solution?
Businesses should be looking at security, integration, reliability, business value and total cost of ownership. The weighting of these various factors will be dependent on the industry, the technical environment and the operational needs of the organisation.
Is cloud based AI better than on premise AI?
Neither option is in any way universally superior. Cloud based solutions can be very easy to deploy and scale, on-premises solutions give you much more direct control over the underlying infrastructure. It's a case of which is best for your particular security needs, your resources and your business.
How can a company go about verifying the claims of an AI vendor?
Companies can ask for customer references, can look at relevant performance evidence and can carry out a proof of concept using real business data and pre-defined success criteria.
Why are integrations important when you're selecting an enterprise AI solution?
Integrations enable the AI platforms to work with existing business apps and data sources. Without them you end up with the situation where staff have to manually move the data around which in turn negates a lot of the benefit of the automation.
How can procurement teams reduce the risks of AI adoption?
Procurement teams can go a long way to reducing risk by defining their requirements clearly, assessing the security controls, reviewing the contract, looking at the total cost of ownership and testing the shortlisted solutions prior to going live.
Choosing an enterprise AI solution is a process which requires a lot of planning, a good bit of technical evaluation and a very clear understanding of what the business is trying to achieve.
Rather than get caught up in the bells and whistles or the vendor's marketing claims, organisations should look at how the platform deals with deployment, permissions, integrations, performance, cost and support.
A properly structured evaluation process, underpinned by real world testing and clearly defined objectives, makes it far easier to find the right solution for the business.
For the CIO and the procurement team the aim is not just to adopt AI. It's to invest in technology that provides real value, protects the company's information and is robust enough to grow with the business.
