Choosing a partner
Before appointing a firm to build an agentic system, establish what the project includes, what evidence shows the firm can meet your project’s requirements, and what your company will remain responsible for.
Set out the work before comparing firms
Describe the workflow you want to improve, the result you need, and the conditions the software must meet. Include the systems it must connect to, the information it will use, and the decisions that require human approval.
Separate what you have already decided from what the AI implementation partner will need to work out. You do not need a finished technical design to begin selection, but each firm should be responding to the same account of the work.
Record what your company must provide and any gaps that could delay the project. If reliable data or system access is missing, or no one is responsible for oversight, further preparation may be needed before implementation can proceed.
Look for evidence that applies to this project
When reviewing a firm’s previous work, establish what it delivered, what it was responsible for, and how that experience relates to your requirements. A project that appears similar may have involved different responsibilities or operating conditions, and a record of delivering conventional software does not establish a record of delivering agentic systems.
Keep confirmed facts separate from the firm’s own claims and matters that remain unverified. Missing evidence creates uncertainty; it does not, by itself, establish that a firm is unsuitable.
Where partner appointments go wrong
A demonstration leaves important requirements untested
A demonstration may show an agentic system completing a task with prepared data. It may not show how the system handles incomplete information, connects to your existing software, or waits for human approval. Choosing a partner on that demonstration alone leaves uncertainty about whether the system will work as your business requires.
The proposals do not cover the same work
One proposal may include preparing data and connecting existing systems, while another leaves that work to your company. Comparing prices without accounting for those differences can make a narrower proposal appear better value. Your company may then face work and costs it did not anticipate.
The delivery team remains unconfirmed
A firm’s past work may reflect the experience of people who are not available for your project. Without confirming the proposed team’s relevant experience and availability, you may appoint the firm on the strength of expertise your project will not receive.
Responsibility after launch remains unclear
A proposal may cover development and launch without establishing who will maintain the system or respond when it fails. Unless your company or another provider is prepared to take on that work, you may reach handover without the support needed to keep the system operating.
Understand what the recommendation supports
A selection report should explain why a firm is recommended for your particular requirements, not simply describe its capabilities. You should be able to identify the evidence behind the recommendation, any risks that affect the choice, and what remains uncertain.
The conclusion should reflect what the assessment establishes:
- One recommended firm
- The evidence supports recommending one firm over the other firms assessed.
- Several recommended firms
- The evidence supports more than one firm without establishing a justified order of preference.
- No suitable fit
- The evidence supports none of the firms assessed against your requirements.
- Insufficient evidence
- There is not enough evidence to determine which firms, if any, are suitable.
Any recommendation may depend on stated conditions. A conclusion of no suitable fit applies to the firms assessed, not to the entire market.
Check what has changed before you commit
The report should state when its supporting information was checked. A change to the proposed team, its availability, or the project scope may affect whether the recommendation still applies, even if the report is recent.
Before appointment, check whether important information has changed and confirm that any conditions required before proceeding have been met. Any remaining conditions should be clear to the people making the decision.
Sablecrest’s role
Sablecrest supports the decision on an AI implementation partner with a selection report based on your confirmed requirements and the available evidence. Your company engages Sablecrest and retains the final choice of implementation partner.
Recommendations require human approval. Firms under consideration cannot pay for a recommendation, and any partner fees are standardized and disclosed.
Sablecrest does not design or build the software, or manage its delivery. The appointed partner is responsible for that work. Selection does not guarantee implementation success or financial results.