Conference presentations, awards and supplier case studies are selected stories. They can show that a tool or experiment exists and rarely reveal how widely it is used, which controls surround it or whether the claimed value survived normal operation.

That does not make public claims dishonest. It makes them incomplete evidence for another firm’s investment decision.

The practical question is not which AI logo competitors have bought. It is which client or operational capabilities may be changing, what evidence supports that view and whether the same problem matters in your firm.

Separate talk, trial and operation

Use a simple evidence ladder when reviewing a claim.

An announced interest or strategy shows leadership attention. A vendor demonstration shows potential capability under prepared conditions. A proof of concept shows that a team tested something. A pilot can provide contextual evidence. A production service has real users, ownership, monitoring and consequence. A measured outcome connects operation to a defined change with limitations stated.

Do not promote a claim to the next level without evidence. “Piloting a multi-agent workflow” does not establish production use, scale or benefit. Equally, lack of a public announcement does not prove inactivity. Some firms sensibly keep internal experiments private.

Record source, date, exact wording and confidence. AI products and programmes change quickly, so a competitor profile needs an expiry date.

Research capabilities, not brands

Start with work that may affect client value or operating economics:

  • first-draft support;
  • research and retrieval;
  • knowledge access;
  • document review and extraction;
  • meeting and communication assistance;
  • workflow routing and administration;
  • service monitoring and reporting;
  • client-facing interaction;
  • software development or quality assurance.

The same product can be used safely for one bounded task and poorly for another. Brand comparison misses information, workflow, evaluation and human accountability.

Ask what outcome a competitor might be improving: response time, capacity, consistency, risk detection, new service or client access. Then assess whether that outcome changes buyer expectations or the economics of your market.

Avoid treating every productivity feature as strategic advantage. Widely available tools may create a new baseline rather than differentiation.

Use legitimate evidence

Relevant sources may include public service descriptions, client communications, recruitment, product documentation, approved case studies, procurement notices, regulatory filings, events and direct conversations where participants are authorised to share.

Evaluate source incentives. Vendors highlight successful deployment. Candidates may describe tools without knowing firm-wide governance. Public job descriptions show intended capability and do not prove it exists.

Do not seek confidential information, misrepresent identity or encourage employees and suppliers to breach duties. Competitive intelligence should have a clear ethical, legal and information-governance policy.

Triangulate material conclusions. A case study, specialist vacancy and client demonstration together may support higher confidence than one conference sentence.

Expect ordinary uses to matter

Useful adoption often begins in unremarkable work: summarising an approved information set, preparing a first draft, extracting fields, assembling a recurring report or helping a person find a relevant source.

These tasks can produce value when they are frequent, bounded, reviewable and measured. They can also fail through weak information, displaced verification effort, confidentiality exposure or overreliance.

Do not repeat universal time-saving percentages. Establish the existing task time and quality, run representative cases, measure review and exceptions, and decide whether capacity is genuinely released.

The discipline around a simple tool may matter more than model novelty: clear eligibility, good information, evaluation, training, monitoring and an accountable owner.

This is a more useful competitive lesson than copying whichever tool receives the most attention.

Investigate shadow use inside your own firm

The best evidence of near-term risk may be internal. Colleagues can adopt public or embedded AI features before leadership calls it a programme.

Create a safe route to disclose tools, tasks and information used. Review expense data, browser or software inventories only under appropriate employment, privacy and security governance. The aim is to understand and control use, not create surveillance that drives it further out of sight.

Provide clear approved, conditional and prohibited categories. Explain how colleagues can request a tool or experiment. A blanket ban with no practical alternative may be ineffective; uncontrolled permission is worse.

Assess current licences as well. AI features may arrive in productivity, CRM, document and service platforms through product updates. Confirm whether they are enabled, what data terms apply and whether permissions or workflows changed.

Learn from discontinued experiments carefully

Public silence after a launch is not proof that a tool was shelved. Where reliable evidence exists, a stopped experiment can be valuable.

Ask which layer failed:

  • the problem was unimportant;
  • the use case exceeded available information or capability;
  • output quality or review cost was unacceptable;
  • clients or users rejected the service;
  • governance or legal conditions were missing;
  • integration and operation were too costly;
  • the supplier or product changed;
  • success criteria were never defined.

Avoid categorical claims that client-facing chatbots universally underperform or that complex drafting always creates more rework. Consequence and task vary. The defensible lesson is to test the complete service with real users, hard cases and human escalation before scale.

An AI strategy document that receives no owner, funding or decision is not operational capability. That is a governance failure shared with many non-AI strategies.

Compare your firm with the problem, not the noise

Assess your own portfolio across use, value, information, technology, people, governance and operation. Different practice areas may sit in different positions.

If individuals are experimenting without controls, formalise visibility, policy and one structured test. If the firm has no relevant use, investigate client and operational problems before choosing a pilot. If services already operate with evidence, improve their evaluation and consider adjacent work only where foundations transfer.

Do not impose a universal 90-day pilot or “this quarter” deadline. Urgency follows client need, shadow risk, market change and reversibility. A rushed experiment can create more delay through weak evidence or lost confidence.

The related AI adoption curve offers a language for these states. Use it as a profile rather than a ranking against peers.

Build a competitor-informed decision

For each plausible capability, create a short record:

  • market evidence and confidence;
  • relevant client expectation;
  • current internal problem and baseline;
  • available information and workflow;
  • value and risk hypothesis;
  • build, buy or existing-product options;
  • evaluation and guardrails;
  • owner and next decision.

Competitor evidence may raise or lower priority. It should never substitute for the internal problem. A rival’s production service can still be irrelevant to your clients, operating model or obligations.

Consider response options beyond copying: improve the human service, communicate an existing strength, partner, wait for product maturity or deliberately avoid an unsafe category.

Report uncertainty to leadership

Present claims in three columns: what is publicly evidenced, what is inferred and what remains unknown. Date each item. This prevents a repeated industry anecdote from hardening into strategy.

Avoid statements about the “average” mid-market firm unless the dataset, definition and date are robust. Exploration, licence availability, informal use and governed production are different measures.

The AI adoption self-assessment can help establish the firm’s starting point. Confirm its destination and questions before publication. Its value lies in prompting evidence, not telling leaders they are ahead or behind.

The opportunity is not to deploy in silence while competitors talk. It is to ignore theatre, understand real work and make decisions the firm can defend. The tools producing durable advantage will be the ones connected to a useful service, governed in operation and measured against an outcome that matters.